📚┃suno-school

[teal raft]

I'm only getting hip hop now

cookout, grooveout, catchy hook

[neat nexus]

I kind of managed to force it on one with [wailing vocals] but the second version continued the instrumental and faded it.

[neat nexus]

That's not horrible hiphop though.

I don't know what they're saying but it sounded good.

[teal raft]

No, Suno seems to always nail it pretty well

Boss can get bent, I'm not going back to work (Get. Bent.)
I'm fed up, I'm done with his smirk (Eat Me Bossman)
No more time sheets, no more cubicles for me (Cubeless Jambori)
I'm gonna ride this freedom like the wind, can't you see? (No you can't because bossman is blind))
9 to 5 more like time to die since I will not be attending my job any longer (Unemployed!)

[broken lotus]

😂

[neat nexus]

Oh wait! I also changed the style to this in the continuation: grooveout, bonghit, tokeout, catchy hook, mellow daddio, male vocals Then I used the [wailing vocals] just above the [verse] tag.

I think it works...

[shadow timber]

Is there some sort of problem? I keep getting error messages on everything, shouldn’t be a copywrite issue since I tried it with random lyrics

[neat nexus]

Might be a glitch?

Try adding several lines of spaces of indeterminate length to the end of your lyrics. See if it goes through with that.

You can always remove the spaces when you get to the next segment.

[shadow timber]

It’s doing it on no lyrics at all

It finally went through, I think the issue must have been some sort of memory thing where it got stuck

[radiant badger]

hello! i am player! i am new to discord and this bot!

[mystic grail]

i've been using chirp on website for over a month now, and lately i stopped receiving free 50 chirps a day. Did they change anything?

[kind token]

Support the company they are doing a good job even the 8 dollars worth it 🙂🙂

[merry saffron]

Hi, I have a question that has been bothering me for a long time.

When I'm composing Chinese lyrics, I find that Chinese itself has branches
(apparently, in China, due to diverse regions and ethnicities, even though it's the same Chinese language, the pronunciation varies).

Is it possible to specify when using the Suno AI algorithm to only use Beijing Mandarin or Taiwanese Mandarin?

[arctic lantern]

The credits refunded thing is happening again and this time I made sure that the clip I'm trying to continue has audio on it up until the last second.

Anything else I could be doing incorrectly

[versed valley]

Welcome! For help, check out Suno Help

Welcome! For help, check out Suno Help

[neat nexus]

I had a lot of hopes for this one, especially when that chorus came in. Unfortunately the chorus altered the vocal completely and it never went back.

I tried again and I think I managed to get something out of it but I'm still not completely happy with it. And I'm not completely certain it's a rock song any way....

If you make it alway to the end, please laugh along with me at the sheer number of fade outs I inputted in sheer frustration to get it to end the song...

[neat nexus]

I've been playing with some of the lessons I've learned with this one. Some interesting stuff...

Some more testing.

style was: glam metal, ballad, bluesy hard rock, melodic hooks, soulful warm vocals

Then: glam metal, ballad, bluesy hard rock, melodic hooks, soulful warm female vocals

With that second one I added some old lyrics I was writing a bit back to get it to stop adding its own. It really wanted to sing something the second test so I gave it something to chew on rather than let it hallucinate vocals.

As you can hear though, I lost my vocalist in the transition.

Gained this other dude...

[thin timber]

how do i 100% force the algorithm to put two lines separate from eachother

ive tried putting them one newline apart and two also

but i keep getting issues where big parts of the chorus are said during the verse

and where lines of the chorus combine together

https://cdn.discordapp.com/attachments/1083129341870878780/1184360896500338748/okay_shit_AI.mp4?ex=658bb10a&is=65793c0a&hm=bf224ada39887008133e8cd758108bba4a2366ef4da4e97e43e200426181ad88&

[neat nexus]

Can I get a screenshot of the lyrics in the field?

[thin timber]

thats me ttrying to fix it with hyphens but it didnt work

https://i.gyazo.com/4a0fdc925ea221d13fc2c1614e954c23.mp4

heres another example

it merges those parts together

dont want that

[neat nexus]

Did this just start happening?

Recently, I mean?

[thin timber]

no ive had a consistent issue with this prompt

[neat nexus]

When did you start this prompt?

[thin timber]

i think yesterday

[neat nexus]

I have linked this in #🪲┃bugs-web and in the moderator chat to make sure it is seen.

[thin timber]

Is there a way that I can also do the opposite?

for example if I want to have lyrics like this:

Fire fire fire fire fireball (Fireball)
I'm a fire fire fire fire fireball (Oh ooh)
but I want I'm a to be played just before the line starts

also, how do i also create instrumental sections?

for example if im using Dubstep
how do i create a "Drop"

[Verse]
Lyrics...

[Pre-Chorus]
Lyrics...

[Chorus]
--- DROP, 4 BARS LONG ---

[Break]
--- SMALL INSTRUMENTAL BREAK ---

[Verse]
....

do I have to use something like this
[Drop, 4 Bars]

[Break, 2 Bars]

or ar they not valid metatags

another issue too

they are separate lines but chirp doesnt like it

i think the issue is its trying to merge lyrics on its own

but not properly respecting my lyric placement

also does this work for copying the sort-of punk "false start" chorus

[Verse]
1
2

[Pre-Chorus]
1
2

[Verse]
1
2

[Pre-Chorus]
1
2

[Chorus]
......

[untold briar]

I think it's just a visual difference -- the blank lines are there in the prompt. I also think double blank lines are reduced to a single blank line, so it's not making them farther apart. Though possibly this is different in V2, not 100% sure on that.

[thin timber]

in my case it depends on the generation but most of the time the lyrics are merged when I don't want them to be

[untold briar]

Either way, a few ideas. Be careful with ~~~~ and or any unusual punctuation in general. I use >>> a lot to intentionally confuse the timing and get overlapping vocalizations, but it's a risk if you are trying NOT to be weird.

You can also try make space in other ways. (Yeah) on a line by itself. A single period on a line by itself. That shoudln't be stripped out.

[thin timber]

does using the period occupy the space of a full line?

or will using punctuation like

Line.
Line.
Line.
make it 100% certain

[untold briar]

That could just be a hallucination too. And (long shot) I wonder if [echo] is doing something interseting here...

A period will occupy the space, but I honestly haven't tested it in Chirp V2, so I can't say for sure it'll help on the web version.

[thin timber]

another thing i want to test also is a delayed vocalization

for example singing something right at the end of a line

i.e.
where we go.......
away!

(next line)

maybe this might work?

[Chorus]
So we go (Ooh) to another place
Where we know (Ooh) we can't be replaced
So we know (Ooh) that we matter to ourselves
(Hey Hey)

but im not sure

dont think i have enough credits to test the song yet

[untold briar]

This isn't a solution, but for illustration, check this clip. Sometimes the singer stops mid sentence for a a mini bagpipe instrumental, other times she does something different. So clearly Chirp can see the spacing but it doesn't always respect it.

She sings the word "bagpipes" (line 1 and 2).
"I (pause)" on line 3.
line 4 she pauses to play the bagpipes.

It's a mix when you try for more fine trained control.

[thin timber]

maybe one day chirp can accept additional guided input from composers

that way I can insert midi guidance for singing notes or music

kinda like my other software

[untold briar]

I wouldn't spend too many credits trying but this format for melodies does work more often than chance

Hap[C4_1]py[C4_1] birth[D4_2]day[C4_2] to[F4_2] you[E4_4]
Hap[C4_1]py[C4_1] birth[D4_2]day[C4_2] to[G4_2] you[F4_4]
Hap[C4_1]py[C4_1] birth[C5_2]day[A4_2] dear[F4_2] Su[E4_2]no[D4_2]
Hap[Bb4_1]py[Bb4_1] birth[A4_2]day[F4_2] to[G4_2] you[F4_6]

[thin timber]

that works for when I want a melody but not necesarily a rythm

[thin timber]

https://i.gyazo.com/03fe45f0f65234050da553e5c2000a76.mp4

[untold briar]

Subscription doesn't increase the text limit. To make long songs, you still have to use the continue button. But in case you are using Discord, the web version can generate up to 80 seconds

Under the ... menu

No, sorry, on the web version it's '...' menu "Continue From This Clip"

In Discord it's just the right arrow emoji

[cerulean cloak]

Continue from this clip definitely exists at least on my web version

[untold briar]

That sounds lke a bug. What are you browing on? Can you try in a different browser? Or even rotate your phone or something?

It should look like this

[silent willow]

is there open source code for chirp singing ai?

[untold briar]

Ahh yes. Sorry about that. That is sort of a bug, in that it is horribly unclear. Discord songs are Chirp V1 and right now can't be continued using the web Chirp V2.

In the future they should be interchangable. At least, the devs said it's a goal of theirs.

[cold gull]

Better filter coming soon? Norwegian words such as "fagprøve" (an apprentice test) and "fagbrev" (like a diploma for completion of the fagprøve) dont work due to the first three letters...

[tender eagle]

So okay, I used the following "lyrics" , gpt produce "coherent" poetry mixing English and Ukrainian, nonsensical but rudimentarily cohesive, including physics formulas spelled out in English:

I drink Heinz 57 directly from the bottle, this strange yes, but the vinegar taste is sharp. Sour sweetness while aberrant to some is not so bad. Perhaps a little vodka with it up. Кохання любимая любимение Thor Zeus lightning, we speak only of such things, the quantum state of the electron it's velocity and location unknown , we express as such. pq minus qp = h divided by 2 multiplied by π times i, And it is so uncertain, uncertainty, uncertaintous, what is not known what is known. Всё, everything. All things where there is nothing there is everything and where there is everything there is nothing.ahhhh bluhhhh going hffgh high Hefner. Loose lips sink ships, loose ships are interdimensional. And thus, it is uncertain whether the ship is really there. Whether we are really there, or if anything is nothing at all. Who am I and who are you? That is not a question but the answer.

So I've been experimenting with using acapella spoken word, and it gave me this which is not at all what was said because it was a bunch of random stuff

the second gen was correctly done but with an African American Urban accent voice the second gave me this...I don't even know if it says anything but sounds African French maybe?

https://cdn1.suno.ai/bc415bb3-e3ed-43d8-838c-699deb79b2d4_0.mp3

I asked it for East African Male, acapella spoken word. As style. The second gen was

[untold briar]

That's a hallucination essentially. Though the meaning in audio models is not quite the same, typically refers to any output which isn't speaking or singing the text prompt it is supposed to. But the extreme edge of this is basically "unconditional" output, almost as if you hadn't written any lyrics at all (which you can do, though it works better as a continued segment)

In Chirp specifically unconditional language output tends to only sound like a real language on the surface. It sounds like Fake-English or Fake-German for example, much like "SimLish" the fake english in the Sims does.

[tender eagle]

Ah okay. I just don't understand. French, only Ukrainian, English and Russian, but it absolutely worked on the second generation that it created, it said all the words correctly

It used properly the East African Male, but since the acapella spoken word portion didn't align with the lyrics and the language model. Perhaps it failed to understand that in that sense, which is why. Also it includes music, as it only sees audio as audio, not lyrics and music per se?

[untold briar]

There will be some real words or even phrases, usually really common song phrases, mostly it only resembles real language. Bark actually hallucinates vaguely coherent sentences, that's one thing it does "better" though it's not exactly a commonly desired feature.

The less the prompt looks like a typical song the more likely that is. Also just using way too many lyrics (even in standard song format) makes it more likely.

[thin timber]

do you happen to know why chirp v1 doesnt work very well when i try to generate more slow and soft music

[tender eagle]

And what is the difference between the first and second generation? They always seem quite different in how well they've applied the request. Does it have a different settings value for ...I don't know chirps settings insofar as things like image generation CFG, k values etc.

[versed valley]

Welcome! For help, check out Suno Help

[thin timber]

#🐣┃chirp-beta-1 message
on this one the timing is out of sync and pitch jumping around

[untold briar]

This is my personal speculation, but I think Chirp V1 has a less effective music style prompt, so if you use more lyrics (longer lyrics) it tries harder to sing ALL the lyrics, and songs that sing faster tend to be also faster music.

[thin timber]

interesting

[untold briar]

Chirp V1 is less musically coherent in general. It's only still used in Discord because devs haven't quite finished the nuts and bolts infrastructure to hook V2 up yet. it's not supposed an alternative long term. Though there are some things you might prefer V1 for, for example if you want to make a song that changes music genre every 30 seconds that (at this time at least) is much easier to pull off in V1. But most users want the opposite of that, they don't want a song to change music genres and V2 is way more consistent. But there are edge cases like that.

[thin timber]

https://cdn.discordapp.com/attachments/1128313332273782784/1184424243526647848/chirp.mp4?ex=658bec09&is=65797709&hm=4be6cf4f69a2343f771cdfee31af365524bbe09a80a4620308bc576bea5b8a14&

[tender eagle]

@untold briar me and my brother have been also experimenting without lyrics, but with audio requests built like the generated lyrics but with blank spaces, such as beginning the song with

[Breakbeat]
[Electric Guitar]

[Guitar Solo]

[Hook]

[Chorus]

[Outro]

And requesting as

No vocals or no vocals instrumental.

Sometimes it returns what is asked for and sometimes It produces music without vocals but not at all. What was in the vocal prompt. Which is pretty cool and interesting, but it's almost always a good sound. Sometimes it'll include one of the instruments in the prompt heading, along with other unexpected instrumentation

https://cdn1.suno.ai/d71f710e-8b40-4059-ae05-f9549c78faf6_1.mp3

Begin it with steel drums and accordion, but came out more of a metal style with a somewhat accordion like synthesizer. And I have a typo in the request where it says according solo lol.

Gets really nice in the mid with the accordion break

[untold briar]

Yeah it works but not 100% consistently. Though if you try to get too specific with details a lot of them probably get ignored. But sometimes it works.

[tender eagle]

I really should read the paper if there is one on it, not sure if one is published. I guess. I wonder if there be a way to refine chirp to understand [Instruments: instrument type 1, instrument type 2].

Like with some LoRa models for image generation that can be fed style:tags, color:tags and so forth

Also [] vs. ()? How are those understood, is () and [] viewed differently or the same? And maybe {} enclosures

[thin timber]

when i tried it [] is tags and () is additional lyrics

[tender eagle]

It also does well with acapella alone, singing to a style of music, I really enjoy 1990s New York City hip hop, a capella

[thin timber]

normally for backup singers/sounds

[tender eagle]

Ah. Because I noticed sometimes the generated lyrics will use. () Even for the style heading such as chorus etc

I wonder if a capella and acapella is differentiated. Didn't consider that.

[thin timber]

like
Oh oh oh, abc efg will make them want to sing everything but
(Oh oh oh) abc efg will normally put backup singing for that part

[tender eagle]

Yeah I saw that. But sometimes you see if you generate lyrics in the web UI, it will

(Verse 1)

Yeah we do Y when X (Y Y X X). And work out as you describe

Which parens both but contextually gets it

I have on occasion , had it say chorus as a backup but that's been very rare.

[thin timber]

as far as i know it will see it but normally ignore it

happens even when I use single word lines

[untold briar]

It's quite freeform, in that any text could work. In Chirp V0 even [chorus] tags were a discovered feature, not a design decision. Stuff like that was only by design in later versions.

[] is seen as different input from the model than lyrics, as non sung instructions or song structure like [chorus].
**() **is treated as a background or additional sung voice.
{} are not special, at least not on purpose. They will have some effect and could even be useful or consistent, but would be a discovery if so.

These are not hard rules and you have probably noticed Chirp might decide to sing the word "flute" when it sees [flute] from time to time.

A fairly small set of tags like [chorus] and [verse] have some extra elbow grease from the devs to make work extra consistently. This set is probably growing over time.

[thin timber]

is there a way to force chirp to finish the chorus or last part completely

[tender eagle]

ah, sometimes, the generation will use parens for the headers of each section, which is why I asked, that may be whatever NLP llm you use on backend for lyric generation, GPT? or local?

[thin timber]

my generations stop as soon as the lyrics have finished being sung

cuts off most of the songs making it hard to continue from them

[tender eagle]

@thin timber I have just removed sections of lyrics and began the "continue this " with that, then I've had to use audacity to combine the two .

[thin timber]

i tried doing that too but it starts singing the continued section in the middle of the chorus

[tender eagle]

yeah but change the lyrics at the top, it will repeat the chorus a second time

sometimes*

[thin timber]

ah i see

have to get lucky then that it loops good

[untold briar]

Not technically, but adding a bit after the last lyric can help. Could even be a blank line, a line with a single period on it. Or a line like (Ah Yeah) that you wouln't mind if it was cut off or not cut off.

[thin timber]

I tried using the . thing but i cant figure out how to make it generate a blank line

like i tried doing something like this to get a good intro forced into the song

[Intro]
.
.
[Verse]
Lyrics—

but it seems to just skip straight to verse

[untold briar]

You can try something like [intro][piano instrumental] or something that implies a longer bit of content.

[thin timber]

will the piano instrumental bit hang on to the rest of the song? if so, how do i remove it

ive tried using markers like [Swing] and [Solo] but sometimes it affects a lot of the song

[tender eagle]

@thin timber if you force it to loop, just use audacity here, this is a break section where it flips the intro, not what was wanted included too much of the continuation, but I chopped it down to

it will align "at some point"

I force the loop by changing lyrics to remove the parts it has done already and it seems to understand to use the new lyrics (the old - where it cut off)

[untold briar]

What do you mean by hang on? Like the rest of the song has too much piano? You can generically say something like [instrumental] or [melodic instrumental] or [some-descreptive-word instrumental] that may not be as distinct, piano was just random example.

[tender eagle]

but you don't see the part I had chopped in this edit of it, at the top part which is I think a kludge of 3 , but this is my remaining file for it

you see where I ended up cutting the ending part though , where it is placed in the editor

[thin timber]

yes they seem to have a very strong global effect

maybe can benefit from negative tag

[Swing]
[!Swing] or [-Swing]

something like that maybe good for generation

dont know if supported

[tender eagle]

here's the audio you cannot hear the actual change you can SORT of see it if you zoom in on the combined audio, but this is the "hack" i use for that, anyhow

the chop between the two is right after the piano keys ,after the short break, begin, so the alignment was good, and lucky, with both options, both aligned well, so it was 2 for 2, the vocals just went better with this generation, so it may be more likley than not to properly do so.

The piano's initial placement is across the line of the chop so the tone is almost perfectly the same, with only a small amplitude shift but not freq.

this is how it originally sounded with the cut section, dramatically different, but the continuation allowed for a much better shift to it. as well as completed the song.

@thin timber I think, akin to image generation, that a combination of the generated audio + post production is probably always going to be the only way to get exactly what is wanted, for a while, anyhow, at least from a closed UI as sorts, an no way to really get control of generation elements.

A negative prompt in general would be amazing as would weighting ability for style, with extended style application, maybe like prompt travel

@untold briar is there any base model of chirp available to the public directly or all unreleased? pre-v1 etc?

[untold briar]

Nope, just Bark the TTS model

[tender eagle]

yeah I've played with bark

is it based on bark as a base model?

or is it completely different

I think I am going to just implement some bark nodes in comfyui maybe I can get it to do some interesting things that would be applicable to chirp. Again not so much a quality, but output negotiation that I am looking.

[untold briar]

Unknown. Suno devs chime in from time to time if you search this Discord but just bits and pieces.

[untold briar]

I'd love to see that, keep me posted

Does ComfyUI have tools that would apply to Bark?!?

I haven't used it awhile...

[tender eagle]

Yeah, i am slow :p but I'll let you know IF I get it to work, I'll have to make a loader and see.

not per se.

but it's just a node based system

we can make it apply

people have made NLP LLM applicable nodes for it

[untold briar]

I would love a node based Bark UI but making it work at a boring nuts and bolts UI code level is a nightmare. But maybe coudl build on Comfy

Gradio is RIGHT OUT

[tender eagle]

Gradio works great, just is too superficial and the under the hood customization is too granular

on the end user pov

[untold briar]

I use Bark sort like a node based UI, but just in notebook cells

Is there an example of a LLM using comfui in some cool ways?

[tender eagle]

ah, yeah that allows recursion and processing at various steps with your choice of when and what executing each cell

yeah there was a redditor, though it may be using GPT, and not local LLM, but application of that is just about a loader and conditioning implementation but he made a whole image + pick your path kind of visual game with comfy

it has gpt incorporation in some nodes, but that's mainly prompt generation etc. not sure if he has a workflow lemme find it, saw it the other day should be in my history

yeah so it does use GPT, but everything, rememeber, that works in comfyui, is a node, with the main stuff in big combo node suites within the primary comfyui nodes directory, you can implement the same, as custom_nodes for loading, as comfy did with his primary checkpoint loaders in the nodes directory for checkpoint loader etc. so loading a local model would be no different I do not believe,

so here is his layout,

I do not know if they include the workflow meta when posted to reddit, didn't check

https://www.reddit.com/r/comfyui/comments/18fn0fj/i_created_a_dialogue_game_in_comfyui/

someone wrote some oogabooga api integrations. as well

but it could all be written to nodes, it's all just python encapsulated, so I see no reason why not

[untold briar]

The only way I've gotten some unusual stuff like music-ish output from bark is very strange sampling, typically recursively sort of building on a single output but inferencing again and replacing sections. (In a total mess of notebooks, nothing that work reliably lol)

So would be nice if I could setup a workflow and save it better. It's real easy to to something in a notebook you can't reproduce by clicking around nonlinearly which drives me NUTS.

ahah

[tender eagle]

https://github.com/hushaudio/rivet-oobabooga-plugin

But still api,

[tender eagle]

implement an output that saves workflow as mp3 meta lol

[tender eagle]

yeah that wouold drive me nuts

I am looking for local llm checkpoint loaders, without the use of the api, I mean... it should be doable, if not I'll add it to my 1.8E+3 list of things I intend on doing... 😢

[untold briar]

I would 100% do something like this. Literally all the relevant Bark Python code is way smaller than any typical album art! Jam ALL OF BARK in every mp3 for rela.

[tender eagle]

hahaha what if your playlist was just a list of jsons

"sorry cannot play your song, 3 nodes needed are not installed"

"please connect to the internet and clone the node base repos"

[untold briar]

It doesn't really solve the notebook problem, or at least not without extra stuff. Somehow it has to know that you ran a cell, set a value, then edited the cell to do something different, but that code is still needed, etc

Every mp3 includes recorded keystrokes and mouse targets for the entire session, solved

just recreate the whole thing

these are my kind of terrible ideas

[tender eagle]

well consider this, take each cell, make it a function , and just call functions in order with the settings as an array of each step sequence.

[untold briar]

yeah there's probably a way. it's still quite silly but you know, that's my vibe

[tender eagle]

an array of function call order + arrays for settings of each

wouldn't be good for the human reading it, but

[function1,function2,function1] , [(12.323,122,5),(32,2,-1),(12.21,102,-5)]

[untold briar]

People who nkow what they are doing dont' have this problem in notebooks, it's not a limitation, it's me being sloppy ultimately

[tender eagle]

save that as the metadatat

where each function = your cell

as meta data, + cell as function, it could be implemented as a single python script with all the cells, and the implemented array execution

[untold briar]

I know the fast.ai notebook extensions have a lot of cool stuff that I keep meaning to learn some day

[tender eagle]

never tried it

never even colab really, just jupyter.

I use jupyter via colab locally but it is annoying and goofy,

[untold briar]

the pitch was like, you just work in a notebook, click a button, and it generates aclean python library and documenation from your notebook

[tender eagle]

ah

[untold briar]

Colab is always useful just because it's a standard setup, so people put "try in colab" links in their readmes, even small projects. And sometimes it might be a real pain to setup locallly otherwise but the work on Colab is done

[tender eagle]

nbconvert and some automated postprocessing or template extending coudld transform the notebooks into standalone, reproducible scripts with ordered function calls and separated parameters. some other things like mlflow, dvc or simple logging decorators can also help track experiment metadata during runs

okay so here we go, this is just a hypothetical, but if we could convert the notebook to a single script of cell funcitons, and store the json as calls to functions + params

{
"workflow": [
"cell1",
"cell2",
"cell1"
],
"parameters": [
{
"k_thresh": 1,
"amp": 12,
"sigma": 3.5
},
{
"preload": false,
"noise": 0.5,
"text": "trumpets"
},
{
"k_thresh": 1.25,
"amp": 12,
"sigma": 3.25
}
]
}

and the converted notebook,

import json

nb cells as functions

def cell1(config):
k_thresh, amp, sigma = config

Use vars

return output

def cell2(config):
preload, noise, text = config

Use vars

return output

json params

with open('params.json') as f:
data = json.load(f)

workflow = data['workflow']
params = data['parameters']

for func, config in zip(workflow, params):
if func == "cell1":
out = cell1(config)
elif func == "cell2":
out = cell2(config)
... ... ... and so on

oof forgot it did that to my comments lol

that could be how meta would be saved lol

it wrecked the actual script and indentations, but you get the idea

import json

def cell1(config):
k_thresh, amp, sigma = config
return output

def cell2(config):
preload, noise, text = config
return output

with open('params.json') as f:
data = json.load(f)

workflow = data['workflow']
params = data['parameters']

for func, config in zip(workflow, params):
if func == "cell1":
out = cell1(config)
elif func == "cell2":
out = cell2(config)

like that,

just stuff all the cells in a script and write a json output from a log,

a small ui front end with "nodes" based on "cells" would allow passing of output to the next cell and each parameter, that would be not at all hard as a comfy implementation with a offload to the cell conversion py

I mean tbh, it could be directly implemented if the output from the conversion scripting simply generated custom nodes, instead of using a monolithic cli script, it could just encapsulate it in the node structure. and pass vars directly, with comfy handling the workflow as it does already

[untold briar]

You got me Googling for something that just handles this, This is interesting https://github.com/ipyflow/ipyflow

[tender eagle]

oh neat.

[untold briar]

Does a lot more than I need, but seems super cool.

[tender eagle]

I think I am going to work on this... to allow a notebook parse to custom node set for comfy. I will probably have to require some custom tagging within the cells for include/exclude cells, cell type, , parameter variables which wold normally be changed within the notebook and so on. But I do not think this will be too difficult™ for jupyter to comfy nodes, for non sequential notebooks,

[untold briar]

Feel free to DM or @ me any interesting progress, that's a tool I would be interested in for sure

We should probably have been in bark-technical channel, oh well, too late now

[tender eagle]

let me preface that by saying I have about 10 repos I work on in an overly distracted manner for various "projects" I will one day complete/not complete hahah, I really would like to do this though. This may be useful in other AI applicaitons, which might make me much more focused

I'm just overly distracted..but I am an EMT and my 24 hour shifts often (don't say slow) have periods of me just sitting around, with my laptop, so maybe I can get something working... lol

do you have a good example of a notebook which is minimal but could be to follow the kind of method you currently use for this that I could use as a test case?

[untold briar]

Ideally for my problem I'd prefer something mostly invisible, sort of a magic safety net. Silently tracks whatever is done in a notebook, edits, execution order, and according to some criteria logs a copy of something that can reproduce the output start to finish. So for bark could simply be when you write an audio file to disk, there's a second Python code dump next to it.

[tender eagle]

ah hrmm that would be even easier than comfy node from parse

[untold briar]

I'll extract a notebook the nest time I'm messing around. They are mostly purely exploratory so it's not even obvious what should happen (you can't 'run all', it's a mess of non linear ideas) in current state.

[tender eagle]

I don't know "tons" about notebooks, unfortunately. but maybe even a modular inclusion that is called from each script which stores the cell data, and follows the execution until a "restart" or the "final" cell is executed, that would write the info out, and when the final step is executed it could take what was stored and collate into execution order / variables for repeatability.

[tender eagle]

yeah, that's what I need, a linear cell notebook would not follow the expected purpose of all this anyhow

I'll see if I can do that first, and from that I'll have a basic output which I could then use later to do the more complex comfyui node implementation, cos really, that may be really nice to just have each cell as a node, whith its params adjustable at each node step, comfy would handle the storage of the workflow as it does any

just shoot me a direct when you have have a chance to grab that for me.

[untold briar]

Tons of custom nodes I see. Like just pure python if you want https://github.com/IcyIbis/ComfyUI-Code-Node I'll have to make time to try comfy again it's way bigger than I remember.

I'm happy to see your Comfy LLM example plays audio, last I looked I couldn't even figure out that much

[tender eagle]

Yeah there's tons I have a few commits on some of the node projects, mainly just fixing things, but none of my own really except local nodes I've made which, as always, I find someone did already better lol it is just python encapsulated in the node structure comfy uses

oh that's interesting, that node...

[tired spruce]

/Tem como gerar musica apenas instrumental?

/Is there a way to generate only instrumental music?

[harsh yacht]

Is broken continuations from stitched together audio a fixable bug?

Hmm, it seems to be working a bit better

[merry saffron]

Can someone answer my question?

[broken lotus]

There aren't a lot of accents by region, it seems to work only when the genre expects it. (a music genre like reggae will sing with a Jamaican accent)

we sometimes spell words in a way to force Chirp to pronounce it with the accent..., but not viable on a whole song.

[arctic lantern]

What does (...) do?

[neat nexus]

Spaces things out a bit so that it doesn't try to do everything at once.

[thin flint]

() also works for stuff like instrumental builds and such e.g. in intros. More effective for me than, say, [Instrumental Build].

[thin timber]

i wish suno had an auto continue option

you could write really long lyrics and it will automatically partition them (or manually)

then after the first generation you have the next set of lyrics saved and editable

and when you choose Continue for a clip it will ask you if you want to continue where you left off

if yes it will keep going

you could write an entire song out and have it generate it without needing to worry about cutoffs

[neat nexus]

Not sure if this worked or not... 🤔

Having a chance to listen to it on a better sound system I can see it does not.

[shadow timber]

Is anyone connected to chirp here? I’m freaking out because all my music is gone except for a few little things I made today, I have one page and everything else is gone

[normal grove]

There is a continue page button at the bottom right hand side usually (website version)

[shadow timber]

I go to page two and it says I haven’t made any clips

This is new

[normal grove]

Maybe they're updating something.

Try refreshing the website

[shadow timber]

I tried refreshing and I tried looking at it on a different computer just in case and everything is fine

Gone not fine

I can’t even type I’m so upset

I upgraded my account to the higher one and it looks like everything made before my upgrade is gone

[normal grove]

It'll still be there, but just a hiccup, ask support, they might be able to find out what is going on.

Sorry I am unable to help.

[shadow timber]

How do you get support

No one here is answering

[normal grove]

Could be a holiday day, but I think the email is [e-mail]

[shadow timber]

Well this is terrible, I can’t believe this is happening, I didn’t download so many things I love and now I can never hear them again, the only thing I have left is a song about my cat I just made for fun

[golden pasture]

you broke the rule of having multiple copies of things you care about.

[shadow timber]

It’s hard when I work on an iPad and it won’t download from their site

So I have to make a special trip to download anything

[broken lotus]

I'm sorry this is happening to you, please don't panic.

Notice how no one else seems to be having this particular issue...?
That's a sign it may be a user error. I am currently logged in and see all my songs.

I suggest you make sure you are logged in with the correct account (google? Discord)?

If you can see your songs on mobile then they are all still there.... I understand your frustration with the download 'share' menu on mobile.... the website is getting an update within a few days (I'm not dev, just a mod on discord), I will make sure someone is testing mobile links so hopefully this is easier for you in the near future.

[shadow timber]

It’s not a user error, I am logged in with the correct account and I can’t see any songs except the ones I made this morning after upgrading

I can’t see my songs on mobile

They are gone

I was explaining why I hadn’t downloaded every single song I liked out of hundreds

[broken lotus]

I see. These were songs made under a free subscription, and after changing subscription, the old ones aren't with that acct...

[shadow timber]

No they were made under pro

I upgraded to premier and my hundreds of pro songs disappeared

[broken lotus]

I think this is an issue for
[e-mail]
they will know more....

OH! from Pro to Premium?
please contact billing with your discord username, and I gess the email acct that you signed up with on the website...?

and maybe any other infor you can remember from your pro acct...

[shadow timber]

I haven’t made any songs on discord so I’m not sure what that would have to do with it

And my pro account is exactly the same email and everything, it’s literally the same page I was working on in my browser

[broken lotus]

OK I can't help, I am only a mod on discord. You will need to explain to billing in an email...
They will straighten it out within 24hrs...

Please email them.

[shadow timber]

I already have, they haven’t responded

[broken lotus]

they will have a record of your Pro subscription (I assume)

[shadow timber]

I just hope they haven’t wiped everything I cared about because I made the mistake of giving them more money

It feels like someone died

[broken lotus]

ok. I am sorry.

Aww, no it will be a wait. I'm sorry. It is frustrating.

[shadow timber]

You know I noticed when I upgraded that they didn’t prorate my old credits, it looks like they gave me a new account instead of upgrading, but it’s the same account

[smoky hollow]

/sing

[dim glacier]

psst... Have a look here: #🐣┃how-to-chirp

[elder flicker]

Does SUNO can change the singer voice?

such as use a child voice to sing my lyrics?

[versed valley]

Welcome! For help, check out Suno Help

[limpid spade]

What the heck does that have to do with Suno????? Oh it's spam posted in almost every channel.

Tested this and [muted guitar] [muted electric guitar] in Style will get you that background music, corporate, inspirational, motivational sound which is all over the royalty free music sites. I need to improve my song structure though and melodies are not that impressive but it's a start.

[hybrid plover]

Can i use the same chorus or clip i used before ?

[graceful condor]

time to see if this thing can make harsh noise

well that sucked

[viral lotus]

Are there any prompts that can force Suno to generate nonsense vocals?

Also, is there any way to ensure that pure instrumental generations go the full generation length? (Or at least have a better chance)

Also, with the new Chirp web version, what is the best way to try to pick a song back up that has been cut off in mid-verse or mid-chorus?

[limpid spade]

For me, Continue Clip will usually pick up the same chorus from the first Chirp

[raw swift]

How do you get it to make a actual intro instead of just starting the instrumental in some random area.

[astral ferry]

does anyone know how to control [Instrumental] flow? like, are there any commands to introduce some instrument in a particular segment? or intensify a bit? etc

[shell jungle]

is suno v2 on discord now???

site model

[midnight brook]

Anyone have the problem of chirp deciding to only generate the back half of a really long track.

A lot of credits and time burnt on this track only to not have it print the whole thing.

[iron blaze]

Are credits renew?

[untold briar]

Free credits renew once a day. Paid plans renew once a month

[iron blaze]

YES. Thank you.

[spark walrus]

Guys, how can I use the command?

Guys, how do I continue the song and make it big?

[pearl gull]

Great questions. I wanna how to do that too,

[versed valley]

Welcome! For help, check out Suno Help

[pearl gull]

My AI songs only have 1:20 minutes

I'd need at least 2:00

[untold briar]

That's normal. To make a longer song you click the three dot menu ... and pick "Continue From This Clip". You can do this repeatedly. On a clip that isn't the first the ... menu will have a new option "Get Whole Clip"

[pearl gull]

Thanks so much

Doing this, it'll get some other credits or it doesn'?

My songs stop at the half 😕

[graceful condor]

Hello. Is it possible to import song created in web version to chirp bot? I want to continue my song created in web. do you know?

[untold briar]

Not at this time. The devs said they hope to eventually make this work.

[idle inlet]

chirp

[versed valley]

Welcome! For help, check out Suno Help

[versed valley]

Welcome! For help, check out Suno Help

[floral dagger]

I keep getting a lot of generations that don't have lyrics sung. Is this a common problem? Any fixes?

[untold briar]

Can you post an example prompt?

[unreal basalt]

Came here to say the same. The quality of music has improved, but the lyrics is gone in about 80% generations.

[unreal basalt]

I have purchased a Premier subscription, switched to the browser GUI (from Discord, I used before), and made about 50 chirps, mostly with Russian lyrics.

[frank isle]

The browser UI changed? I seem to only be able to insert lyrics and not the style of the song

[light ether]

Depending on your monitor size, might need to scroll

or, make sure to hit the toggle at the top "Custom Mode"

[frank isle]

Ahh I didn't see it

There's even a title setting now, huh

Does it change anything in the song?

[light ether]

No, just for organization and sharing

[frank isle]

I see

[floral dagger]

Sure:

#🐣┃chirp-beta-1 message

Text
[Verse]
Yeah, the time is here, everybody's near
We're gonna dance all night, no time for fear

[Chorus]
We're gonna stay up all night till the morning light
Let's keep the energy high, shining oh so bright
Style of Music
deep techno, hard bassline, 120 bpm, sexy female voice

I'm still playing around. I tried it on the 'website' as well. Lyrics were missing there as well, and after many lyrics generating errors.

[floral dagger]

Agree.

It's a shame because I believe I end up using credits for that, and from what I gather, there aren't enough credits to actually try it out enough to decide on going for a subscription.

[fallen radish]

Example clip that refused to generated the lyrics: https://app.suno.ai/song/cd5bf7cb-5d00-4989-bb19-8ed75cc604c9 ... clip 1 had lyrics, clip 2 no lyrics

[untold briar]

I bet it's the techo in style, and possibly the smallish amount of lines of lyriccs in general. I'd try a few adjectives used to describe a voice instead of sexy. Maybe sultry female vocals also female singer airy female vocal and just other words in general that are voice-ish like velvety raspy piercing maybe close

[floral dagger]

Well, I was going for the 4 -8 lines is best guidance, but thanks for the other advice. I'll give it a shot. Appreciate it again. 👍

[untold briar]

Does it still say 4 to 8 lines somewhere? It should recommend two versus as a good minimum now for V2

[floral dagger]

I didn't even know it's v2! I tried it on the website first (today), and yes, it does mention 4-8 there. I'll try that too. 👍

[light ether]

Whoops. We'll edit that

[unreal basalt]

I tried using both similar and different lyrics, and attempted to vary the styles, but it seems to be irrelevant. Also, I've noticed that in most cases, the music in the chirps sounds like one repeating sample, lacking any progression.
Here are some examples without lyrics:
https://app.suno.ai/song/9a67d0c7-0bbe-4fd2-a43c-50366e863c07
https://app.suno.ai/song/29acd608-1646-4dc3-9c8a-2b416e558b6d
These also have a "sample" issue
https://app.suno.ai/song/2fc72cf4-8d19-446f-a583-8e85923f3624
https://app.suno.ai/song/98053fa1-eb46-491d-8c3e-adde6fa17382

[fallen radish]

OK that's 6 in a row that generated nothing but instrumentals... taking a break from wasting any more credits... was only doing it half the time before, now it's both clips on every attempt. I hit "report" on all of them...

[light ether]

Do you mind DM me linking some of them?

[light ether]

These are all custom lyrics right?

[unreal basalt]

Yep

[untold briar]

I tweaked your style prompt to deep techno, hard bassline, 120 bpm, sultry female vocals, close singer, female and it's 4/4 singing. Change is adding sultry female vocals, close singer, female to the end. (Sometimes repeating helps too.)

[floral dagger]

Excellent thanks! 👍

And you know what, @unreal basalt mentioned 'one repeating sample, lacking any progression'.

Very true! That's a shame. 😦

You've been very helpful. 💯

[frank isle]

How do you make the AI end the song?

[floral dagger]

@mc #🐣┃chirp-beta-1 message

You might want to have the team look into the following issues being raised:

1. less adherence to the prompt as far as vocal generation (either no vocals, or in the case of the 2nd generation in the link, the woman is a man}
2. frequent cut-offs in the lyrics, even if they are only a few lines (again, 2nd generation)
3. the lack of progression/change between the verses and the choruses

Wonderful job otherwise, but these are credit chompers, and onboarding new subscribers will become a challenge if these issues become a long-lasting problem.

Again, incredible platform otherwise and an improvement from what I tried a few months ago. Thanks.

[inner steeple]

so.. you now get 2 different lyrics variations on single chirp with web version?

I guess this is intentional

[cold gull]

The new TOS is made a lot more confusingly. Reading through and cant get a clear answer on ownership of output?

[neat nexus]

You can try [outro] and then [fade out] and some people have good luck with [end]

[neat nexus]

Is this clearer? https://suno-ai.notion.site/FAQs-b72601b96de44e5cacd2cd6baa985448

[frank isle]

I see

[cold gull]

Thanks for digging up the link 👍 . I have previously talked about the TOS and even read the whole thing multiple times but in this new TOS update today, it got a lot more wordy and confusing https://www.suno.ai/terms

Btw, anyone else having issues downloading their audio on the new version of the site?

[neat nexus]

I've been focussed on music videos and issues involving my personal employment most of the day so I haven't made anything yet today.

[cold gull]

Seems like it fixed itself 👍

[white topaz]

I created/generated a song that I really like a couple of the singers voices. And would like to make more with those same singing voices. I was trying to think of a way to this. One thought was just to continue from the end of the last song. But then chirp would, I'm thinking, just add it as part of the original song. Then I'd have to manually separate them. Is there a way I can start a new song, but with those same singers voice?

[untold briar]

Not yet but better control over continues (like for example "continue but from 25 seconds not the end of the whole clip") are on the roadmap, and this may be able to do a version of what you want.

[white topaz]

@untold briar Okay. Thanks for your help!

[solar geyser]

How can I get a better intro?

[thin flint]

Interesting. This was an AI-generated prompt from Suno, no lyrics.
https://app.suno.ai/song/7816afae-5b46-4835-afdf-57e988b574f6

@untold briar Are you on the team? Is there a roadmap to look at?

[untold briar]

Just a discord mod. But if you check posts from Suno team like mc, you will see chat messages. (Not a real roadmap) #💡┃suggestions-web message

[thin flint]

How are people sharing stuff in #🐣┃suno-showcase via generations.ai bot? Is that people creating clips in Discord?

[untold briar]

In the new web UI you can share the sing song page, it's available either by clicking the icon the puts it ni your copy and paste buffer, or click the title on the song to navigate there

[earnest willow]

Are servers overloaded or something? This has been going on for more than 5 minutes. 4 chirps not generating.

[thorn hill]

yeah, happening to me too

I just hi thumbs down and generated new ones, then it got ready really fast again

[earnest willow]

It disappeared from the queue when I did that. I guess it's a way to get rid of chirps.

[thorn hill]

yeah

I mean, I'm using it with that purpose LMAO

[thin flint]

Well it didn't get the female vocal right but I did get an intro that was good. https://app.suno.ai/song/9194bb28-a814-4145-aec0-b429ec35170d

[thorn hill]

Can I put mopre than one style of music? If yes, do I separate them just using commas?

[neat nexus]

That's how I do it.

[brave trellis]

Chirp

[teal raft]

~Chap

[earnest willow]

25 minutes later and songs still not generating on the website. It's a continuation I am trying to get. Force refreshed the webpage just in case anything had changed but it keeps acting up.

Oh well 🙂 Hope it's just a minor fluke.

[thin flint]

This is a proper intro. I wish the instrumental build tag would work consistently.
https://app.suno.ai/song/f240b5a1-ff57-430b-9a45-1f689647c3a1

[light ether]

sorry, i think some a small % of continuations and Get Full Songs are failing rn, team is looking into, but wont cost credits if fail

[thorn hill]

what does the remix button do? I really liked one song here, the vocals are spot on, but the instrumental is just a coontinuous annoying noise, Can I fix it somehow? Or just try again and again and hope to get it right ? D:

[teal raft]

Remix is just what 'Reuse prompt' used to be

[thorn hill]

ooohh

[teal raft]

It just uses the same prompt yeah

[naive jungle]

Is discord using the old version?

[neat nexus]

At this time, yes.

[signal herald]

How to I get the bot to remake the lyrics with different style music

Not remake lyrics but sing it in a different syle

[stone heath]

Problems with V2 web app, new interface

1. No Control for Intro prompts. Vocals start immediately. Needs vocalization to hold the vocals at start.
2. Continuation failure. Lyrics break in continuation to a continuous noise.
3. Vocal change severly in continuation.
4. Quality of music and vocals reduced tremendously as compared to V2 previous version.
5. Languages other than English have taken a severe setback in terms of vocals.

Hope the team would look into the matter soon.

[graceful cliff]

Try now

[stone heath]

No Intro Melody, as it used to be in earlier version. Have you changed the prompt requirements?

[graceful cliff]

Intro is never consistent

you are always pulling from probability distributions

btw, you can try [start] and [end] and see what they do

[stone heath]

Tried all options like Intro, Instrumental, Instrumental, everything

[uncut leaf]

I'm having the same problem
Generation is no longer stable even when generated with the same prompt as the previous version of the website

[graceful cliff]

your experience will always vary with the generative stuff

we do try to make your average experience as good as possible

if you have a wishlist, please do post, these are very helpful feedbacks, I really appreciate them

[stone heath]

I am quoting this from my experience only. I have been working on Chirp since long and now the experience is totally different and painful

[graceful cliff]

I understand

and I am sorry 😦

[teal raft]

You are personally responsible, tsk tsk

[graceful cliff]

lol

I'd not assume the experiences that works well for you before will carry on

but there might be new tricks

[stone heath]

Please do not misunderstand me. I am not complaining. Just informing. You may not even know that this platform means to people like me, who always wanted to create music all their lives, but never could. You have given them a magic wand. Without mincing word, I say, it is the best thing happened to AI. So keep it up with your development and best wishes.

[graceful cliff]

in general, I think a better experience happens when you prompt with just clear genre tags and have less control on the tags but more on the input lyrics

genre tags should be genres, not a paragraph of what you want the song to be, try to annotate that along the lyrics with [] prompts, and prompts sometimes work, sometimes doesn't, AI needs to make the whole song so there are trade offs.

We always try to give you more controllability, but also hopefully it gets smarter so you have to control it less

try to embraces the changes? Personally I think the updates offers a lot more different experience and options to explore. Again we always appreciate your feedback.

I am going to sleep.

[versed valley]

Welcome! For help, check out Suno Help

[robust ermine]

oh nice, seems you made streaming of generations? i.e. you can start listening before it's complete

[runic forge]

I might be the unluckies man ever, but I think [intro] and [outro] stopped working in webpage generations...

I'm also having trouble continuing the song when the cut is inside a verse - it was always a hit or miss, but now it's miss all the way.

[still steeple]

Can I use the browser version of sunoAI in the same way as the discord version? For example, making a song a little longer,

[untold briar]

Yes

[hasty fossil]

Is there a list anywhere showing the genre or styles of music that SunoAi will recognise and create?

[untold briar]

Not yet, there is unofficial wiki just getting started at https://suno-ai-wiki.notion.site/Metatags-22a69f63b1814aa9afbec9d21f7b55c4

[hasty fossil]

Useful, Thank You.

[still steeple]

Thanks, but how can I make a longer song with the browser version?

[untold briar]

In the three dots menu ... you can pick "Continue from this clip"

[still steeple]

I already tried that but it only generated about 2 seconds of music, is it just a bug? It was an instrumental song with no lyrics in it.

[untold briar]

Yeah that's probably just back luck for that specific clip

[still steeple]

Okay thanks I'll try again

Nope, another song generated for only 3 seconds again, and my credits have been reduced

oh I just looked and the credit was restored,

I guess I'll just have to give it some time to try

[normal grove]

Is there an issue at the moment, I am not getting any generations and I want to use my credits before my subscription renews.

[thin timber]

yea mine just crashed too

[viral lotus]

I came here to ask about problems as well

[thin timber]

stuck generating

[viral lotus]

I kept getting "Please wait a minute before generating more lyrics"

[thin timber]

i think maybe technical difficulty

ill use the discord one instead

discrd one is down too

either all services down or restarting

all of my jobs are queuing but none of them are completing

i hope it doesnt drain all of my credits when its complete

[untold briar]

Looks like the Chirp server is experiencing some turbulence. 🚨

I don't have any info but the main Suno team will be awake in just a couple hours, worst case.
Also IMPORTANT: If you have stuck Chirps they will eventually timeout and refund credits. (If they don't you can email support). Sometimes this takes a long time they just keep spinning (it's a known issue.)

[thin timber]

is it possible to recover them

or do i have to save the lyrics

[untold briar]

I think they'll probabloy disappear when you get refunded (at least on the web) so save any lyrics you might lose just in case

[thin timber]

ok ill keep that in mind

[normal grove]

What about in the scenerio where your credits are about to expire for the month, so you want to use them all, but they refund, I guess I'll just lose them.

[thin timber]

also do you have any tips for how to tell chirp to choose a hyper specific genre to work in

[untold briar]

It shouldn't take that long, but if you get super bad timing or something you can email [e-mail]

[thin timber]

me and a friend are trying to get ideas for hood/nightclub music but most of the generations are pop
maybe i have to wait for negative prompt so i can make it focus on drums bass and rap

[untold briar]

What is the prompt?

[thin timber]

and not singing or instrumentals

trying to generate more stuff like this

#🐣┃suno-showcase message

a lot of the other generations are too “musical” for the genre

similar thing happens when working in other rap genres

[untold briar]

Never forget the simple trick of finding the "key" style prompt word and SPAMMING it, repeating it over and over in the style prompt.

[untold briar]

If "atonal" doesn't work try "atonal atonal atonal"

[thin timber]

the one i saw that did the most heavy lifting was “strip club music”

i put that in one of the songs and it was more accurate

trial and error i guess but whatever works

or at least until it went down

didnt get to try other things yet

[tulip palm]

कोई मिलग्या

[thin timber]

the commerical terms for the song and whatnot, do they apply while the subscription is active?

i dont really like subscription model so i would prefer to just buy and cancel immediately after

but i would assume everything stays until the end of the month?

[limpid spade]

check the fine print but pretty sure that anything you create while subscribed is fine ongoing.

[light ether]

Yep, this is correct

[cyan merlin]

I have generated my chirp. However, my chirp is not complete my lyrics how to continue it without retype my lyrics?

[broken halo]

Vocals slowly becomes robotic

[fierce stump]

hi, can i still continue after i gave a ❤️ ?

[broken halo]

Yes

[hushed tulip]

Okay, so I've tested generations on this new website, and there seems to be something wrong with the model right now. Here are some examples. Skip to halfway through it.

[light ether]

We're looking into this one!

[hushed tulip]

Also its repeating the same line non stop. Edit: Actually, it was my fault. I wrote a [Chorus] instead of [Verse]. That font is not very readable for me in this new website.

[hushed tulip]

I also noticed that the time to create clips has been reduced by a lot. Maybe it has something to do with it? Like the model doesn't have enough time to 'think" and just go with "whatever"

[native crypt]

Why is generation now so much faster than before, but the result is so much worse than before?

[hushed tulip]

Exactly!

[sly steppe]

just listened to quality of my stuff from 2 days ago compared to now and wow... hopefully they get it fixed

[brave gull]

this is bonkers

(dont mind the lyrics and the image)

[fallen radish]

I DM'd lots of examples

[normal grove]

The quality definitely fades a lot in the continuation of a chirp, much more than it did previously.

[topaz willow]

I'll be honest: the new Web version is much better in my opinion. However, the issue is that you need to find the right initial chip to create a great song. In my experience, I have to generate around 10-15 chips before finding a voice that doesn't sound robotic and has a good melody. Overall, the improvement in V2 is significant, but it requires a lot of credits to work effectively.

[glacial rampart]

I see that the website has had an enormous upgrade, it's more user friendly. But it seems that every time I try to generate lyrics by writing some sentences in Swedish, it changes the lyrics into English instead. It didn't do that before. Is there a way to fix this?

[normal grove]

It gets better all the time, I'm really impressed by it all.

[dusk moat]

Is chirp an opensource project like bark?

[neat nexus]

It is not.

[dusk moat]

I guess they are taking OpenAI's Route.

[winged drift]

Hi guys. Would anyone have any advice on how to prompt specific durations and have the song keep evolving? right now I can prompt so the initial couple of bars is good, then itll just loop a part continuosly for the rest of the clip. Any ideas about how prompt to keep making musical changes instead? thanks

i should add this is for instrumentals

[tight frost]

I'd like to know this as well. What I did was reuse the same prompt but change the emotion. The result was a subtle change.

BTW, what happens to the credits for the song that are obvious fails ? Like the duration is less than 10sec and we hear neither melody nor lyrics. Are the credits refunded ?

[autumn quail]

Just wanted to add to what some other people have said, the new UI is amazing but the generation quality is substantially lower for me, same issues as mentioned, robotic voices, and often getting stuck in loops that it just wont get out of no matter how many generations I try, with the amount of credits I've gone through for just one full song I like today I could have probably made 5+ yesterday/ before update and I just pushed my sub up to the $30 plan 😦 (something about that was also off, payed just shy of $30 but only received around 7500 credits?)

I understand that as training progresses these things might happen, maybe allowing the user to chose a older checkpoint of the model, so that if they feel performance degrades they can switch to a previous checkpoint they like more?

[autumn quail]

I think this is a defect or not by design, I wasn't experiencing this a day or two ago, now it's happening quite often

[neat nexus]

Please email [e-mail] about this issue and they will check into this matter.

[tight frost]

I meant, is it automated ? It's not reasonable to have a team member manually listen to hundred thousands of failed generations.

[neat nexus]

I believe that if the track is under a certain number of seconds that refunds happen automatically. So if you have a short track like this and did not get a refund then you should contact them about it.

I'd show you a screenshot of what the message looks like except I thumb'd those down and cannot find them now..

[light ether]

Thank you!

[light ether]

Sorry, please email [e-mail] and someone will take care of this

[autumn quail]

Ok cool thanks, so this isn't expected behavior? Thought I might have misunderstood. As for the loops and stuff happening, was the model changed or is it just in my head?

[light ether]

We are looking into it, the loops are definitely an unintended recent bug

Hope to have solutions to all of this over the weekend

[autumn quail]

That's great! Still the best text to music gen out there, so can't complain to much lol thanks for the help 🙂

[winged drift]

thanks ill try

[winged drift]

Yea any prompts to generate evolving pieces with multiple parts are just gving me loops or feedback loops

[autumn quail]

Sure they'll figure something out if they're aware, might be something like the duration of the lookback window used for continuation being shorter, if I had to take a wild guess. It might just be me but it also felt like voice consistency wasn't quite as good from one continuation to the next, just guessing though. It's difficult to gauge just how much different the quality over all is

Just one more thought, it also seems like, if you haven't added enough lyrics for the full 1:20 or 1:00 for continuation, it tries to max out the length with instrumental rather than just cut the clip short like it was doing most of the time in the past, kinda feels like this is part of the looping problem, if there is some type of lookback period in the model. Hypothetically if the lookback is 10 seconds and the last 10 seconds had no vocals, then it gets stuck in the loop 🤷‍♂️

[hushed tulip]

Quality issues + instrumental unintended tracks

[proud crag]

Is there a way to only generate a batch of 1 track at a time? Automatically generating 2 songs at a time runs through credits faster. I'm okay if you want to give me two variations and have me pick which one I like more but I don't want to be charged credits for 2 songs on every click.

[native crypt]

Will the quality be fixed? if a new model or logic comes out, then it is advisable to give the user a choice of models, as in midjourney

[storm sierra]

Any tips on how to get Suno to do a guitar solo? I have a section after the chorus where I'm trying to get a guitar solo part

[neat nexus]

This is something that the team is aware of and is working on: #💬┃general-chat message

[fallen radish]

Has anyone got a working way of setting bpm?

[timber fog]

Am I right in thinking that most chirps cut off one beat short of a measure (on the third beat)? I figure this is probably set up to allow for continuation chirps to kind of "pick up" where the previous clip left off... which makes sense. However, it might be cool to allow users to declare a clip as an ending clip, which would allow the chirp to end after a full measure (and perhaps with a bit of space after, to permit a reverb tail to resolve). Anyway, just a thought...

[fallen radish]

[Guitar Solo] is supposed to do it... but like most tags its spotty

OK... I just generated a continuation clip 9 times... and it begins with "afraid to go to the beach" ... which is no lyric that's anywhere in my song...

ARGHH... I can't get this extra lyric out of my song...

It's the first continuation off the original gen... so I can't go back to anything else to try and unstick this.. OK wow... now it's just generating gibberish

[fallen radish]

You can try [End] or [Outro] after the last verse/chorus

[sharp merlin]

is suno offline? i'm getting system error on pc and mobile

[fallen radish]

Strange stuff is happening to my clips... I'm getting a lyric from someone else's clip at the start of everything I generate... and the resulting clip sometimes comes out completely gibberish...

[sharp merlin]

getting this when i try to remix

[fallen radish]

I'm not getting that... just strange output results no matter what do... other peoples lyrics, weird loops, etc...

[sharp merlin]

tried to gen the normal way then got this.

it seems to be generating only 1 song and failing the other song

[unreal basalt]

Got the same, no matter what I'm trying to do.

[kind token]

🤔

[ember sierra]

My account is messed up, popped up server error, now it it just spinning trying to make a song.

Apparently this is an on going issue I see.

[thin flint]

Dang. I really don't know what happened to this model with the update, but none of my instrumental tags are working especially for intros. I really would like the intro and first verse to be in one clip.

[edgy otter]

What happened? I can't enter chirp command (there is "required field")

[bronze harness]

Stuff like that's always been hit or miss for me. I have yet to put the new update through its paces though.

[autumn timber]

I wish there was some sort of preview before generating

[light ether]

audio preview? working on this 🙂

[inner steeple]

I think for actual lyrics?

if using the noncustom path

audio preview would be really neat tho!

I have no idea how that would work 😄

[steady granite]

Has anyone got any info on the new Remix feature present on the website version? Haven't seen any info on it or did I miss it?

[neat breach]

Not always, I had a few dozen that were 3-8 seconds each and only got refunded for one. Actually had a little pop-up underneath it that said something along the lines of "Credits Refunded."

[light ether]

Same thing as "Reuse Prompt"

[neat breach]

This is a feature on the website.

I do NOT like the thumbnail look, btw. A lot of my generations create similar pictures, so even among liked clips I can spend some time searching through them for the ones I REALLY liked.

[teal raft]

You telling me you don't enjoy birds? Boy

birds birds birds

[neat breach]

Do a lot of interludes, huh? I'm pretty sure [interlude] would work if my guess is ironically correct.

[teal raft]

No, actually, a lot of instrumentals

Whatever image genreation they are using just defaults to birds

[neat breach]

Ahhh. Because "chirp." I haven't really tried instrumentals with the new release. The old one it was like 3 seconds of noise and then silence.

[teal raft]

It's quite good at music but extremely slot machine

You do get to sidestep a lot of clip degradation though, voices seem to be the major carriers of it

[light ether]

Gotcha, yeah, we'll probably add a slider for this one or maybe a grid view

[thin timber]

i would try [120 BPM] but otherwise i only was able to use [Fast BPM] [Slow BPM] or typing the equivalent in the genre box; “fast paced” and “slowed down” also helped in my case

[neat breach]

IDK if that would solve my particular issue with it. I usually batch mine with a few big obvious changes in lyrics and style, which is what I'm looking for. It just shows the thumbnails and the title, I have to listen to each and every one for a few seconds at least to find what I'm looking for, and for similar iterations with only a few key changes, I may have to listen to the entire thing.

[light ether]

I see, want to optionally see lyrics then? We will also have search at some point

[neat breach]

Yes, underneath it in wide lines, not crammed off to the side.

[light ether]

Kk, we will look into some sort of setting here

[thin timber]

is there a way to search or sort by genre

[neat breach]

Much appreciated.

[thin timber]

i guess its a bit difficult when u use more than one genre in a song

[light ether]

Hopefully in by end of year we will have text search

[neat breach]

Any plans to isolate the vocal track away from the bgm?

[thin timber]

stems are coming

[neat breach]

Neat. I dig it.

[thin timber]

would be fun to mess with it

[feral basin]

?????

i cant create ?!

[steady granite]

please can we have a search function for our creations to narrow down the pages and also renaming of creations woulld be awesome

[thin timber]

The continue feature on website i dont think is working?

[steady granite]

someone please correct me if this is the wrong place for feedback/feature requests

it was earlier for me

[thin timber]

i pressed continue and it didnt continue

[steady granite]

about an hour ago

am about to launch a new one so standby and i'll let you know

[thin timber]

maybe it was just a bug

next one worked

[steady granite]

yeah its working off website too

[lone copper]

can i replace the default generated picture in Suno?

[thin timber]

afaik no, only your profile image

[thin timber]

how do i report only the generated image from a clip

the song is fine but the image is somewhat nsfw

[glacial rampart]

Does it work again? I did some chirps last night that got error messages, but they're still in process mode. I've also made bug reports of them

[stuck dove]

Hello! Can u help me? How can I pay pro version from Russia?

[hushed tulip]

That's exactly what I've mentioned yesterday in the suggestion channel. To not only bring back these nice big covers with lyrics visible underneath them but also the font from the previous version. This new font right now is not very readable to me. And I already made some mistakes in my lyrics/style text. Previous website had nice, big and bold text. Very easy on my eyes. I'm on a laptop so I don't know how others are seeing it but for me the old website was perfect.

[broken lotus]

ugh, I thought it was just me...

👓

[random minnow]

I really like how this starts but then more and more background fuzz/noise/static or whatever it can be called keeps creeping in. You can hear the fuzz/noise building as the song goes. What exactly is causing that? Is there a way to edit that out or reduce it in GarageBand or Audacity or some other program? Thanks for any tips or suggestions.

[broken lotus]

😓 it's just 'noise' ... i haven't found a more specific word.

I've found sometimes the prompt
[Break]
will clear it, but once chirp continues and thinks it's part of the song, it doesn't really go away

in RipXDAW it gets sent to it's own layer because the AI can't figure out what notes are playing... (they call it 'un-pitched sound')

UltimateVocalRemover might scrub it with one of the noise remover models, or echo/reverb remover, but it will scrub a lot more too usually...

[bronze harness]

It's full frequency noise, going to be very difficult to isolate because it occupies so much of the frequency spectrum. But, I must say, if you generated a song that long and the artifacts were the worst thing, that's still pretty good. After 3-4 generations, I can't really get anything useful out of it.

[pliant cave]

Some Feedback for the Team on the Web UI Update

Streaming Audio Chunks: The feature of streaming audio chunks as they're produced is a significant improvement in the latest UI update, enabling audio playback in seconds. This clearly reflects extensive backend work. Well done!

Tagging Feature: Initially, I planned to suggest adding tags in the server response to indicate the specific model/checkpoint used, which would enable power users to provide targeted feedback on changes. Impressively, it seems you've already implemented this feature. Bravo.

A/B Testing and Recent Updates: I understand the necessity of A/B testing, which appears to be extensive recently. In the most recent model, coinciding with this UI update, there's an evident improvement in vocal clarity. This addresses a previously widely reported issue of unintended instrumental-only outputs. However, as others have mentioned, there are new bugs, including repetitive loops and fading sounds. There's also a potential decrease in "creativity". I would need a lot more samples to confirm the latter, so it's currently just a gut feeling, but I find it interesting others have reported this as well.

Custom Mode Observations: As of today (Saturday morning), 'custom' mode seems to have reverted to an older model checkpoint. This model is free from the new issues and seems to produce highly creative outputs. However, I've noticed that this mode no longer streams audio but waits until the full MP3 is generated. Is this a limitation of the older model, or simply a UI design choice/temporary situation?

Hope this feedback helps in some way. Looking forward to seeing continuous improvements and innovations.

[winged drift]

ncluding repetitive loops and fading sounds. There's also a potential decrease in "creativity". ....yea the last 100 ive done since update this are the issues

im trinyg various metatags seeing if it makes a difference but no its def something the devs need to work out

[light ether]

This is very helpful mind if I dm?

[thin flint]

I think I found a bug where I've several continuations, but I'm no longer getting the get whole song option. Anyone else getting this?

[final hazel]

you only get that if you continued a part.. it is not showing if you rendered only the first part

[winged drift]

so the "fading' into feedback loops and smearing, seems to be worse when several music styles are used as tags. the more disparate the descriptors, the harder it finds to meld them coherently. Understandable because inputing many concepts is confusing. however, if one could weight the descriptors in affect this could improve output . gonna suggest this

[thin flint]

Yes, and I have continued several clips and have reached the end. No get whole song option.

[final hazel]

😮 That's weird then indeed

[unreal basalt]

I'm still receiving this error every time I attempt it, but it seems like others can generate chirps without any effort. Is this just my bad luck, or is it a global issue?

[sharp merlin]

it's been fixed for me, but I haven't tried to remix anything since then

[serene current]

I think the problem could be a "mix" of different voices/music/noise. I think that the Ai get different relevant "pieces" of sound and put them together. If you pass through a program to analyze the sound it perceives not only the singer and the instruments but also some kind of "pieces" or "sounds" from another music. I don't know how it works, but one of the problem could be the "low quality?" of the data given to the Ai, or the sampling. (it's not a matter of quality of the program, Suno it's great, i mean it's a matter of data available or created.)
You see the same in stable diffusion on some Models. (for example some pruned models)

[unreal basalt]

I even tried to create a new free account to check but got the same error.

[final hazel]

do you use the website or discord to create chirps?

[unreal basalt]

Website

[final hazel]

Did you try a different browser?

[unreal basalt]

Yep, I even tried different devices. No differences at all.

[final hazel]

you create the progression by building further. I personally work with a verse and chorus in the first prompt and for me it creates progression.

[unreal basalt]

A-a-and it began to work right after I wrote about my trouble 🙂

[final hazel]

Cool!

[unreal basalt]

Wow, a new thingy

[bright ember]

/chirp rock song

[final hazel]

wrong channel @bright ember

[thin timber]

oo

i wonder if it works in post

[random minnow]

Thanks. I'm looking at RipX DAW right now to see if that can help.

[broken lotus]

it's this button, (unfortunately only in the more expensive version)

will show certain instruments that have the ringing noise
(usually voice, guitar, or piano... all around the same frequencies)

[random minnow]

Okay, fascinating. RipX Daw separated out 8 layers from Sorceress. By isolating them, I can hear the Percussion layer is a main culprit in creating that "noise." This is quite interesting. Thanks for the heads up on this software! They have a 21 Day free trial so taking advantage of that for now.

[pliant cave]

Do not mind at all, feel free.

[graceful condor]

wait what? so just delte this and no noise?

[random minnow]

I don't think it's that easy. That also would get rid of all the percussion. This is my first look at RipX Daw so don't know much about how to really work it. I'm also noticing there's noise in the Strings part. What's cool about this software is that you can mute layers (one, some, all, etc) so certain things really can stick out. Now how to actually use this software to clean up the song and get it more to where I want it is a whole nother story...

[graceful condor]

well you can delte percussion and add some new^^

iam downloading this now also^^

[random minnow]

True. Just don't know what I'm doing at the moment 🙂

[broken lotus]

it's weird software... and it's only for this...
i couldn't get it to import a clip without ripping it, like if I just want to use a sample...

[final hazel]

So it is just splitting? For that I use this: https://vocalremover.org/splitter-ai

[graceful condor]

Is there a programm when you splitted i can use the notes in a extra programm?^^

[broken lotus]

I selected all that noise layer, and tried a low-pass filter, and just simple volume...
I also deleted some but that seemed a bit janky, lol
percussion might be the easiest section to replace? it has tools for that, but I haven't used yet

[graceful condor]

Melodyne for example can wav into notes or?

hmn

[broken lotus]

it's more, it's some other processes where every note (!) is made into a graphic element that can be dragged around the timeline...

you can just grab vocals and change the timing and pitch.

[graceful condor]

nice, so it extract all notes sperately??

man cant wait i must wait 8min now

[broken lotus]

for just ripping stems, try the free UltimateVocalRemover, it downloads different ai models and writes new audio files with the ai processes applied...

[graceful condor]

it splits every instrument??

[broken lotus]

well.... 😏

[graceful condor]

?? writes new audio files what you mean?

[teal raft]

This software is RipX?

[graceful condor]

sorry my english sux^^

[broken lotus]

here start with this one, it's free.
https://ultimatevocalremover.com

[teal raft]

I'm a native speaker and same

[graceful condor]

iam downloading^^

[broken lotus]

yes, RipXDaw to remove the white noise that creeps into our Chirps...

[graceful condor]

how to remove the sound in rip?

this?

[broken lotus]

(Dev's gonna kill me for pulling this cannel off topic.. lol
it's this function #📚┃suno-school message

[graceful condor]

ah nice ty

😄

my idea is sounds who are very noisy, i just want to get the notes and replace it in good quality

or can you clean with ripxdaw the noise complete?

[broken lotus]

i think it's a case-by-case.... Once it reads the notes and converts them, you can convert them again to midi and record a new track....

it's a really different workflow..., I'd say it's still evolving...

[graceful condor]

hm i will try 🙂

[still steeple]

What prompts when specifying a voice would give a better result for the song?

[idle inlet]

you

[sharp merlin]

i tried continuing a song with a bunch of meta tags for [trumpet solo] and variations of that to test this, also changed the music style to trumpet solo, but it didn't work. interestingly, i added a bit for silence and then for the trumpet to come back and it added the silence, but the part where the trumpet was supposed to be just had a creepy laugh by the singer

so, i don't know if this is possible currently or not. seems like no

[stark wadi]

Suno, please go back to the previous model!!

[final hazel]

What is the previous model? They told me yesterday they went back to the previous?

[stark wadi]

It has improved somewhat, but nothing to do with the previous model

[final hazel]

But what is the previous model then? V1?

[stark wadi]

v2

[final hazel]

They say we are still on V2

[stark wadi]

Well, it will be here, the user interface does not work as well as before.

[final hazel]

maybe https://studio.suno.ai/create works better for you?

[stark wadi]

Ohh my dear interface, great, I'll try it and see how it goes...thank you very much.

[pliant cave]

@final hazel This is speculation, but when they said they "went back to the previous model," I believe they are referring to a checkpoint/variation on v2. To further clarify, they seem to be doing some A/B testing. When you create via the "custom" mode, it uses the model from before the UI update, which does not appear to support streaming. However, it seems that when using "simple" mode, it still uses the (assumed) newer but reportedly problematic model that does support streaming. So, right now, as of this message, you can experiment between those to see the results. It's hard to say which version is newer since their naming convention is not clear, as it's intended just for their internal testing, I'm sure.

[final hazel]

That could be, you might be right. I never use the simple mode, so I couldn't tell.

[pliant cave]

This one also appears to use the previous version of the model for both 'simple' and 'custom' modes. I'm glad they left this UI up.

[final hazel]

Yeah I assume it is the same... But sometimes it feels different in another UI.. I think that is what JavidL felt.

[broken lotus]

this is some good speculation
(checks join date... ahhhh)

[hasty fossil]

I seem to be having a problem downloading the MP3 version of one of my generated songs. Tried everything on my iPad, but it just opens the song player and won't download. Started up my laptop to do it from there, but same problem. Is it just me or is it broken?

[cold gull]

Some songs i have just refuse to get downloaded. even older ones i made unless i open in incognito. Using brave browser.

[broken lotus]

can you get to an ios 'share' menu?
you can share to a message or an email to get the mp3 url.

[shadow timber]

I never can download on my iPad, I have to do it from my desktop

[eager elm]

hi i dont know if this is the correct place to ask? but im on the website and it keeps erroring when i try create stuff and also doesnt always give me my credits back when it does that or sometimes makes a 10-20? sec track and still counts it as used credits or even a 1 min one that isnt music or half is but still useless and doesnt refund me. who do i talk to for help ?

[broken lotus]

You can email
[e-mail]
but if you're having consistently broken gens, you need to fix your prompt.

[eager elm]

what could be wrong about the prompts ? they seemed normal

[broken lotus]

How would I know what your prompts look like.

If you are getting consistently bad gens, as you say you are, and you would rather get something else, then you need to change your prompt as that is literally the only control option you have.

[broken lotus]

https://www.notion.so/Metatags-22a69f63b1814aa9afbec9d21f7b55c4

[shadow timber]

The beginnings of songs are a little funky in this version, I keep having to clip a fraction off where it sounds like it started in the middle of a previous verse before starting my lyrics. This latest song starts with someone saying “us” like it came at the end of a line. And it is totally ignoring instrumental lead ins.

[thin flint]

I use [Instrumental Intro] followed by verse to make an instrumental lead in. Either that or [Intro]
(Instrumental build)
works. Yes, some tags can be put in parentheses.

And intros were always flakey in both versions including this update.

Have to wait until credits refresh on Thursday to test myself.

You aren't using metatags...?

[scarlet mica]

Absolutely none. My friend wanted Nevermore, and I didn't bother.

Used all my accounts' free gens

[thin flint]

Ah, I see.

[scarlet mica]

Still going to have to use tomorrow's free gens to finish it.

[thin flint]

I wish those of us on pro got some free daily credits.

[scarlet mica]

Same lol

It feels bad to miss out on them when you're paying

Oops. I thought this was chirp showcase. my bad

[gloomy lotus]

Now that I'm a paid subscriber, it seems like I'm wasting mad credits just to get the vocals to render... Anyone else have this problem? ( I'm using the app) Update... after wasting hundreds of credits trying to get vocals to show up in the app, I tried coming back to discord with the same prompt. It almost worked perfectly first time. (wrong gender of voice) Second render output voice too! Not as expected, but at least it worked. I think the problem is with the app and the new version of chirp. While the sound quality and speed are much better, its still a little buggy.

[versed valley]

Welcome! For help, check out Suno Help

[modern pine]

free

[fallen radish]

Sigh... continuations are making gibberish lyrics again

[frank elbow]

Hello. I lose a hundred credits just to try to continue the song and add lyrics but to no avail. Is there a solution?

[inner steeple]

if you had no lyrics chirp in the middle, it almost never recovers

[scarlet mica]

Yeah. The failures in free mode... well, they suck, because they use your free chirps, but eh.
Failures in paid mode, you're suddenly realizing that each chirp that fails is 30 cents that is just burned.
But it's AI, and basically to be expected.

[uneven furnace]

I made some instrumental electronic music using Chrip, but it's always the same "hook" repeating over and over again. Has anyone been successful in doing instrumental music with different sections? I tried using [Verse] and [Chorus] (without words), but that seems to work only for songs with lyrics. Thanks for any help or advice.

[sour dune]

hi, is there some [ ] that we can use to affect the result?

to have more control on it?

[broken lotus]

yes, describe the 'verse' and 'chorus' sections with a style-word. Since they have no lyrics to make them different...

I've been using similar to:
[Melancholy Instrumental verse] [dramatic instrumental chorus]

I'm not sure the words actually work as described, or just help Chirp tokenize them differently....

(Translation, it matters less WHAT the words are, as long as there is something there, chirp will do what it wants but more likely to hear the difference)

also use [Break] and [Bridge] to put a little section in between

[sour dune]

ok thanks i will try

[winged drift]

have you had any success with words to help break traditional song structure? say if you wanted a different fill every four bars and a change in the intstruments or melody every 2 bars etc.

[broken lotus]

I haven't been able to change that quickly.... (my experience)

I get good results giving 1 structure prompt per song cycle (probably 16bars)...

you can try a bunch of
[percussion Break] ... [bass break] ... [Cowbell] ...

but I just do the main song sections because chirp usually makes a better song if I don't try to micro-manage it.

[winged drift]

thanks. the reptitive nature of parts is the main issue im having atm

[sour dune]

so the [ ] are something that will be ignored by the lyrics, and tried to be apply to the music?

[broken lotus]

Chip was trained with the prompts
[Verse] and [Chorus]

afaik, all the other prompts are 'discovered'

so it's not a guarantee that words in [] will be understood – sometimes Chirp will sing them if it's not a prompt it understands..., and also Chirp won't sing
verse 1:
even though it doesn't have the []

[sour dune]

what do you mean by : afaik, all the other prompts are 'discovered'

that everyone is traying and check the result?

[broken lotus]

it's AI being spooky and just knowing what musical terms mean...

https://www.notion.so/Metatags-22a69f63b1814aa9afbec9d21f7b55c4

All of these were discovered through trial and error

https://www.notion.so/Style-Genre-List-f7d1bb28193c4eac910f03db3f602508

[sour dune]

mm ok, thanks i will try

[neat nexus]

You're not wrong here. Like at all.

Somehow it doesn't bother me like it would literally tossing away thirty cents does? Maybe it's because I'm still stuck in the wonderment stage. I get irritated when stuff doesn't work and then I kind of laugh at myself for expecting so much when really it's amazing it works at all, ever. I still half suspect that Suno has a house band somewhere wearily playing music at minute intervals for our enjoyment....

[sour dune]

is there a ways to restart from the beggining a song?
I generate a song with random lyrics, but i like it, so i would like to keep the melody. but change the lyrics. Is there a way to do that?

[broken lotus]

no unfortunately... you could try continuing the song, and edit the beginning in a DAW

there is also a feature coming that will allow use to continue a song from a specific time in the clip (like maybe before the verse started) but that feature my be in January after the Discord upgrade...

[sour dune]

Ok that's good to know

yes i guess for the moment it will be on DAW 😄

[warm iris]

Why is the web version unstable?

[spice stratus]

how'd i fart?

[neat nexus]

🤷‍♂️

[pliant cave]

I'm not sure how you calculated the cost at $0.30 per chirp. If you're doing a lot of experimenting, it might be more economical to choose the premier plan for $30 a month. This plan offers 10,000 credits, effectively reducing the cost to about $0.015 (or 1.5 cents) per MP3 generated. As you mentioned, since the service is based on an unpredictable transformer model and is currently in beta (recently transitioned from alpha), there are frequent updates and A/B testing. This can be frustrating when you encounter bad chirps or experience regressions in the model, as has happened recently. However, this is a common challenge when working with cutting-edge AI technologies. Despite these issues, I think paying 1.5 cents per MP3 for experimentation purposes is quite reasonable.

[neat nexus]

I'm sure we all agree....until we run out of credits.

LOL

[pliant cave]

This actually brings up a different, but good discussion to have: the credit top-up system appears to have significant flaws, or perhaps it hasn't been adequately updated and given attention. For example, after Suno reduced the pricing for chirp plans a few months back, they didn't seem to adjust the rates for credit top-ups accordingly. As a result, for $30, you get 40% fewer credits in a top-up compared to what you receive in the regular plan. This disparity seems to encourage users to create new accounts and abandon old ones rather than purchasing credit top-ups. Ideally, it should be the opposite: top-ups should be more cost-effective than the plan's initial credits. This would motivate users to maintain their accounts, build a history and loyalty, and purchase additional credits when they run low each month. The inconvenience of creating a new account, coupled with the frustration of feeling 'ripped off', might discourage casual users and lead them to quit the service altogether. Just my two cents on the matter, but Suno is likely more focused on the models themselves right now, and rightfully so; I see rapid progress.

[thorn hill]

Is there a way to force a guthural man voice, or very deep voice as vocals in a music?? I want to make one where a Werewolf Bard sings a song but I'm only getting High vocals D:

I tried using Death Metal as style, but then the vocals dissapears LMAO

[pliant cave]

"deathcore" has that vocal style.

[broken lotus]

not had a lot of luck getting 'deep' voices...

try
'doom vocals'

[thorn hill]

OOOH riight, gonna try that

[pliant cave]

Here is an example, be warned this is NSFW due to excessive cursing.

[thorn hill]

THAT is what I want, gonna try to get close to it

[neat nexus]

....

Anyone feel like doing some tests?

[final hazel]

https://tenor.com/view/shiba-inu-head-tilt-question-gif-3521016

[neat nexus]

I've been playing about with trying to get some Japanese heavy metal generated and noticed something...

[tender cypress]

how do I make a song in the style of some artist using AI?

what kind of AI exist?

[neat nexus]

This:

鋼の叫び 火花散る夜 テールライトは闇を切り裂く アスファルトが囁く、この熱狂 クレセントムーン、鼓動高鳴るレース

and this:

Hagane no sakebi hibana chiru yoru Teeruraito wa yami wo kirisaiku Asufaruto ga sasayaku, kono nekkyou Kuresentomuun, kodou takanaru reesu

Gives different results despite theoretically being the same wording.

The style I've been using is: ジャパニーズ・メタル, glam metal, progressive rock, blues, catchy hook, extended solos

[neat nexus]

You cannot just enter in an artist's name and hope to trigger their style. Your best bet is to research their style and try to list that in the music style field.

[final hazel]

So what do you want to test then @neat nexus
I mean, you already tested it, right?

[neat nexus]

If anyone feels like it. I'm just trying to confirm the results I seem to be getting.

The choice of alphabet seems to be making a melodic change akin to a music style change.

Hmm...

I should test swapping out ジャパニーズ・メタル for Japanīzu metaru in the music style field and see what that does (if anything) too.

[final hazel]

So it's about the style, not the words? Because I don't hear any lyrics at all

[neat nexus]

Yeah the sound of the instruments even plays differently.

[final hazel]

I know I am not that long here... but what I noticed is that consistency is still not in reach. At least not with what I am testing with tbh. I usually work with raps, but still get "singing" rappers sometimes lol

[neat nexus]

::nods::

Part of the reason I switched was in hopes of catching where the vocals end.

Not sure if it's going to help me all that much though if the sound changes.

[final hazel]

I wish I could help you, but I am personally not good in recognizing the genre you are going for in the first place 😬

[neat nexus]

No one has to jump onboard here, by the way. I just noticed something that might be impacting things and wanted to mention it.

[final hazel]

gl 😉

[frank elbow]

Why I couldn't add lyrics to my song. The result I always get is music without lyrics

i use web app

[broken lotus]

https://www.notion.so/Why-can-t-I-hear-my-lyrics-6dea581cf3d5414d968fccd3f940994a

[scarlet mica]

I dunno how accurate that assessment is. If you're a person paying 30 a month, and still running out, you're probably a power user that is more comfortable with a higher expense for that which you extensively use.
By business practice, subscriptions are much cheaper than single purchases. Not quite half off. But still.

[neat nexus]

Gave up on that attempt as its vocals got increasingly distorted as it went on. 😦

[scarlet mica]

Yeah. Even if it is mostly fine, if the last bit of it is corrupted, it ruins the rest of the continues.

[broken lotus]

I think those prices were from back when paid option was $10/mo for 1500 credits, (free was also very limited, they didn't renew every day)

(waaaay back like 8-9 wks ago, lol)

A dev mentioned the top-up prices need a review, but I'm sure that's a boardroom decision....

(it feels expensive now, imo)

[scarlet mica]

Yeah.

Wasn't arguing that the current prices are right. Just that the previous comment train of thought was not what's proven successful by common business practice

[broken lotus]

(I also did not intend to sound argumentative)

[random lily]

hello ,where can i talk with customer service with my account

[neat nexus]

[e-mail] unless it's billing related or account related in which case you should email [e-mail]

[pliant cave]

@scarlet mica , I see your point, and I understand the typical business practice you're referring to, but there's a crucial distinction in Suno's approach. Typically, subscriptions offer more value as an incentive for commitment, with one-off purchases being pricier but available to all. Suno's model, however, only allows credit top-ups after a subscription purchase, which should logically offer even more value to those who've already committed. Instead, these top-ups are less cost-effective than the subscription itself, creating a loophole that not only skews metrics like the churn rate but also disincentivizes continued subscriber and account loyalty. By making top-ups more appealing to existing subscribers, Suno could close this loophole, which could actually lead to a healthier business model with better user retention and more reliable data.

[sweet bridge]

how do you make it do an instrumental [intro] and [outro]? These seem to rarely work for me.

[thin flint]

On this new model, I did:
[Instrumental Intro]

[Verse]

Or, the AI suggested to me once to do:

[Instrumental Intro: Build]

and it actually worked, creating a shorter instrumental section to fit vocals into the minute long clip.

Outros are harder to get them to do what you want. I took what Suno did with the Turkey Time song.

[Piano, outro]

[Outro]

The [Piano, outro] starts the outro, and the [Outro] tag by itself does its best to end it.

[limpid spade]

Any tips for when a continuation ends mid sentence? I've had some cut out 2nd line of a 4 line chorus. Tried starting next chirp with full chorus, no chorus or literally continuing from the next word without success. I tend to take a step back and try my luck with a new continuation from the first chirp and hope for the best.

[untold briar]

Usually I try 1) Start exactly where it cut off 2) Repeat just the last line where it cutoff, possibly with [chorus] tag. But sometimes it's just a dead end clip.

[ionic bloom]

I've been using the website to make music, and I'm just wondering... How far off script is it suppose to be able to get? I gave it lyrics. It didn't like them, I guess. Is this normal?

[untold briar]

It's hard to say why but that "SimLish" output is pretty common. For some reason it sang the end of that prompt and then just kept gonig so it tried to make up the words and Chirp is not a master of language yet

[ionic bloom]

Ok. It sounded good enough that I would have believed it was doing actual lyrics but in a different language or something. The other generation from that pair also only did the chorus part, but it ended after. I'm guessing there is a bit of an issue with the ending of the first section of the song that caused it to want the chorus.

[untold briar]

Bark is much cooler (or creepier) when it does the same thing because it generates actual words. One thing Bark still has over Chirp that I love.

[ionic bloom]

I had been thinking that using a German named music style might have somehow caused it to try singing in German or something.

[untold briar]

That might make it more likely yeah. Especially now I see it uses a full 3 german words. But it should typically do okay with that.

[ionic bloom]

I don't actually know German, but I looked up what Rammstein's style is called. It did pretty good, but may have messed with the vocals a bit.

[untold briar]

Chirp Unprompted: No countries, no language, no words, pure music the world can understand, as one.

[untold briar]

"click" is strange, but chirp nails "click instrumental"

[still vine]

is there any promt documentary? I want to trying generate a hip hop beat but they most of the time are same

[jade mist]

Could you increase the max clip lenght beyond 80? Continuing is not always the solution…

[broken lotus]

Chirp likes nouns, I think.... and maybe thinks 'click' isn't noun enough....

[normal grove]

I created a GPT that tells you, genre, sub genre and standout features of any song, So I use Chat GPT to base them on other songs. Do you have one in mind?

[broken lotus]

add descriptive style-words
https://www.notion.so/Style-Genre-List-f7d1bb28193c4eac910f03db3f602508

[echo crag]

Hi, is there a command to show all the chirp I made?

[broken lotus]

search for 'Mentions' (your name)

[echo crag]

thank you!!!

[digital jungle]

What's up with the "I'm sorry I'm sorry I'm sorry before the lyrics start?

[sturdy flax]

wet circuit I wish I knew more about this. I would think the template has cross over, but would differ depending on specific vocal style prompts or the genre of music prompted

I thought about creating a GPT custom bot for specific genres

[jade mist]

Any news about Suno working max lenght of a clip?

[digital jungle]

I think chirp is down now. I keep getting "I'm Sorry" lyrics so I made a short video for it! https://www.tiktok.com/@graywaterops/video/7314385554181786922?is_from_webapp=1&sender_device=pc&web_id=7307109712536847914

[sour dune]

hey

is there a way to control the lyrics rythm?

or it's 100% random?

[midnight night]

@neat nexus hello good sir, is there a way we can chat on a private channel or direct message?

I need to report a possible bug with the latest change they made to Suno-web-app and its better for me to talk on DM, but i can also talk here......
Since that day i have issues generating audio and keep coherence when resuming from another made song, to continue the song in full mode.

In a summary, i have created 3 songs from intro and outro and pre chorus and bridge and chorus and all, it was a perfect song, NOW after the update, songs will not make sense or even read the verse that you wrote to it.....

I'm a pro user, so i wish to know if there is a new BUG or something.....

[neat nexus]

I'm sorry that you are having these dificulties. You may be best served by contacting [e-mail] since I have no special knowledge I am just a user of the software like you.

[midnight night]

ok is that an EMAIL address?

[neat nexus]

Yes, yes it is.

[midnight night]

ok let me try then there, so there is no LIVE helo here in any channels?

live help

[neat nexus]

https://tenor.com/view/dies-cat-gif-21483615

[broken lotus]

if you can describe the rhythm, or find out what musicians call it, you can prompt for it.

[sour dune]

😄

[neat nexus]

There were changes made to the model and then some reversion of those changes and then patches to the changes and...

Bottom line is some of these things are known issues and they are being worked on. It's part of why there is less of a presence of the team in the Discord despite them being here in the listing. They're working to bring things back to par and in getting the long awaited handover of v1 Chirp to v2.xx Chirp in the Discord version.

Plus all sorts of things happening in the background that they don't necessarily announce until things go live. Like the recent Co-Pilot announcement.

[open magnet]

Yeah sorry lots going on 😓 but promise we’re very heavily working on all this

[graceful condor]

No Porblem you guys do a great JOB!

my ai song got played at my christmas party from my job D:D:D:

freeky feeling

german schlager is hard rocking with ai

[neat nexus]

That's awesome! 😄

[graceful condor]

ya i gave him the usb stick and he liked it a lot!

i wondewr why german schlager is always very clean and no noises?

secret bug?xd

[autumn timber]

Yea i made a family Christmas song like a vintage 1940’s

[thin flint]

Hmm. Why didn't Suno announce the partnership with Microsoft in the Discord here? Also I guess you can't create full songs with it.

[bronze harness]

Any clues as to why I'd get an error with this prompt? "lo-fi beat with no lyrics". I'm on the web app, typing in the normal create box, no custom.

[neat nexus]

They did: #🔔┃social-media message

[neat nexus]

No idea. Seemed to make lyrics but didn't actually use them: https://app.suno.ai/song/bcec54d9-5561-470e-8145-5a577c1d3a31

The other prompt sung the lyrics it created.

[bronze harness]

It did in the end of that sample. That tracks with what it finally did for me when I changed the prompt slightly. It does really well with lo-fi it seems; however, I wish that there was a sure fire way to say no lyrics. I also tried to continue and it did half a bar and quit.

[neat nexus]

How about this one? https://app.suno.ai/song/78556d18-f87b-4527-ab3d-699bfb818f56

https://app.suno.ai/song/3864417d-8f26-480c-9be8-9a39ca80b63a

[hidden forum]

Yeah, I don't really know
https://twitter.com/JordiRib1/status/1737142899903811784

https://blogs.bing.com/search/december-2023/Turn-your-ideas-into-songs-with-Suno-on-Microsoft-Copilot

I wonder if it's v1 or v2

[neat nexus]

Has to be v2 I'd think.

[hidden forum]

Have you guys seen this?:
New AI first hardware
https://twitter.com/LinusEkenstam/status/1737100105852014683

[crude night]

#🐣┃chirp-beta-4 chirp

[versed valley]

Welcome! For help, check out Suno Help

[placid dove]

is Chirp open sourced at all?

[untold briar]

No, just bark

[placid dove]

RIP, sad.

[meager geyser]

Hello there! Which command to use to generate song with male voice?

[midnight compass]

I need minimum 5 min lenght... to create my songs. I think 5 min it´s a very good time to create our songs.

[broken lotus]

Now that Microsoft and Suno are BFF

can Bing help with Chirp prompts...?

[still vine]

thank you

[jade mist]

How to find other releases besides trending

[graceful condor]

Does suno have an API?

[neat nexus]

Not one that is open to the public as far as I know. (I as a non-programmer assume there must be one internally, but I don't actually know how that works.)

[graceful condor]

Yeah, I don't see any mention of it. Instant customer if they did though

[neat nexus]

Hmmm...

Now that I'm thinking about it more, there must be something for them to plug into Microsoft CoPilot with, so there must be an API, it doesn't seem like they're making it available externally except to specific partners though.

You can always send them an email to ask about it though. [e-mail]

[turbid arrow]

I can use the Chirp bot on another server 🤔

[neat nexus]

Right but that's Discord more than anything. But yeah, there must be an API for Discord to plug into...

[turbid arrow]

Ok Thanks

[neat nexus]

Just remember the Discord bot remains at v1 Chirp for the time being. A transition to v2.xx is planned but there have been delays.

[turbid arrow]

Ok thank you

[outer furnace]

hello everyone!
It may be just my imagination, but I feel like I am getting a lot more misses or just bad quality generations since about 2 weeks ago when the website update dropped.
I am not using a paid subscription just to be aware of it and it might just be something related to the "Continue this song" feature as it fails a lot more than before.

I also noticed songs generate a lot faster now. Almost instant. Could it be related? Perhaps the generation was sped up at the cost of quality?
If you have any information please let me know! It seems harder now to get good quality vocals or songs than before in general...

Also I noticed now the random generation behavior has changed. A new song generates two sets of lyrics instead of the same lyrics with two different melodies.
Is that also something new?

[untold briar]

During the website update there were some issues, but if you try it now, everything should be the same as it was before (for music quality)

In custom mode

[full sluice]

Hi there, just curious. I just subscribed to the 10/mo plan. But, when I come back here and type in slash info, it shows that I’m still on the freebie.

[broken lotus]

Ahh, please email
[e-mail]
tell them your login email on website and your username here at Discord.
They will straighten it out
(it is maybe a glitch between google login and discord login, they just need to connect the accts)

[full sluice]

K

[sour dune]

Hy, i know we can do some background voice by using ( ). Is there other little effect like that we can use?

[bronze harness]

I've found you can ask for anything between []
Whether you get it or not is up to chance and how well it's feeling, but you sometimes do.

[zenith depot]

yo

Therre is a Suno Plugin on Copilot:
https://twitter.com/Saboo_Shubham_/status/1737675390586351719

but I cannot find it

[thin flint]

The 21st, today, is when my credits should refresh. What time should I be getting them? What timezone is the server in/ PST?

[umbral loom]

Wdym by credits refresh

[thin flint]

I'm on the pro plan. I currently have 0 credits and they don't refresh daily like free users.

[west pine]

I keep getting this error which is preventing me from finishing my song. Everything I'm doing is 100% original. "Couldn't generate that. Suno is for generating original music. Try different lyrics, and use genres and vibes rather than specific artist names." Tried many variations but keep getting it. Any workaround?

[thin flint]

That error might be comming up because the AI thinks you're writing an NSFW song.

As well as the original stuff

[west pine]

Thanks

[graceful condor]

hi! great software! how to make acapellas?

[sour dune]

is there a way to control the background music. i notice everytime you extend of retry, the music change... so it's difficult even after in a DAW to recombine good part.

[broken lotus]

use [a cappella] as a metatag before the lyrics section

or 'a cappella' in the style description, if it's the whole clip

[broken lotus]

Might be the luck of the 🎲
kinda depends on where the song cycle is when the continue happens....

if it's in the middle of a cycle, Chirp will try to finish it (and then has a tendency to keep repeating) but the song will usually sound continuous (may need to add a [Break] to get it to stop)

if it's at the end of a cycle (chirp will sometimes cut the clip just as the new cycle is about to start) the next section is free to do something creative (harder to keep it on the verse/chorus pattern it had before, may need to get back to the Chorus sooner to make sure it is remembering you main pattern)...

Might need to generate a few more options, and go with the best one....

[sour dune]

unfortunally that's what i do, but i was thinking you knew already how to fix those kind of things

[broken lotus]

I waste a percentage of chirps when I'm working on a longer song...

Only real advice I have is to listen to the previous clip and be sure where it's at in the song instructions (doin instrumentals it gets hard to know if this is 'verse' or 'chorus' sometimes...)

and it's ai so there's some luck involved...

also editing in a DAW is my last resort...

[broken lotus]

I think mine refresh in the afternoon, around 2pm EST... (anecdotal, from memory)
so maybe check back after noon and see if they've been refreshed?

[stone mantle]

I'm having issues on the Suno website with song lyrics. It seems that the Create function will generate instrumentals without an issue, but only some creations will include the lyrics. Any suggestions?

[broken lotus]

some suggestions here:
https://www.notion.so/Why-can-t-I-hear-my-lyrics-6dea581cf3d5414d968fccd3f940994a

[stone mantle]

Understood, thank you!

[brave gull]

i wonder what is being flagged as not original here because I doubt anybody has ever written such stupid lyrics

[neat nexus]

Try it without depressive as a style modifier, maybe?

Otherwise without translating the text I can't say.

[brave gull]

i reworded it progressively and it worked somehow

i did a binary search to try and pinpoint what was triggering it

:^)

[midnight compass]

please, what is the name of the music style? Anybody has a list about it? Thanks.

[swift steppe]

Hello, guys! I wanted to ask if there's a way to instruct Suno to create a wind arrangement or a bass solo in a specific section of a song. Is it possible to give Suno these kinds of instructions? If so, how can it be done? I've tried several approaches, but Suno seems to be ignoring my commands.

[broken lotus]

some tips here for instrumental breaks
https://www.notion.so/How-do-I-make-an-Instrumental-352abcb0bf6a4f77b8fcf76a768f18af

[swift steppe]

Great thank you very much !

[umbral loom]

Probably a bug

[wintry adder]

is it intentional that i should have to log in to notion to read this page? i see the content flash for a second and then it puts up a login screen

[broken lotus]

idk. I am not a fan of Notion, lol.
I think I used my google id to log in.

[untold briar]

Chirp might secretly also be the best TTS AI

[broken lotus]

I don't know what to think, lol

Anyone want to do a Shakespeare project...? All the plays...?

How many chirps is that...?

[untold briar]

I think it's doable if you accept the voices will not be consisten from day 1. Even in a single clip like this they don't properly alternate, it would take 100x more Chirps if you tried to keep it straight. Just go for raw performance each segment.

[broken lotus]

well I got exactly 1 (out of 10) MacArthur Park-inspired Lady Macbeth 'out out spot' scene.... where she actually sang. 🧐

(the disco is excellent, but no vocal – normally I'm kinda good at this so... I blame the Disco)

Maybe disco cancels out iambic pentameter...

[untold briar]

Maybe start with like pure disco chirps, continue each with shakespeare, if the lyrics are effecting it too much

[broken lotus]

I mean... The 🪩 is 🤌

maybe it's the extended remix...

Maybe I'll just stick with disco, lol

[untold briar]

it's awfully hard to find an animated discord emote that doesn't look like it should have an epilepsy warning

[thin flint]

Dang. Been trying to do an instrumental intro on this new cheerful song I'm creating and it just absolutely will not do it. I don't put instrumental into the style box becaise I want an instrumental lead in to the first verse.

[untold briar]

You can put instrumental in the style box. If you use lyrics, it will still sing them quite often.

[thin flint]

Interesting. Why are the vocals so low quality on this? Was it trying to put some sort of effect on them? Also, my pre-chorus and chorus were hallucinating. Melody is great though.
https://app.suno.ai/song/7c8ae984-ae55-44b5-921c-59dc07a50c8f

[solar sparrow]

I did an experiment with Suno, what do you think? I was amazed that he understood everything I told him, except for what I asked him at the end and I ran out of credits, I couldn't continue. But he made all the rhythms that I gave him, obviously he added some variations to make it better, but the basis of the rhythm that he plays is what I asked of him.

I ask a question because I don't know anything about how to train a bot, but couldn't you give Suno visual examples and make him listen to what it sounds like what he is seeing at the same time? So that he associates what he is seeing with the sound and he can learn to read. This way you could tell it what the rhythm you want is like for the song you are going to generate.

I don't know if what I did was an experiment or if he already knows how to read and I spoke to him like a caveman. (skip the part where I talk to him as if he were a human being and motivate him, come on Suno, you can do it!)

Links:
https://app.suno.ai/song/e977240f-0316-44e7-9647-7de5118a152e/

https://app.suno.ai/song/72b81606-ff65-4eb8-86c6-0785e253bb43/

https://app.suno.ai/song/ad931214-0f7b-4b9e-bbd4-dc122a321e19/

https://app.suno.ai/song/36fee187-8b6e-4324-b8ab-c1f5a817bef9/

[thin flint]

I don't think its as good as you say it is. Things in brackets are very hit and miss, especially instructions that the AI doesn't explicitly know E.G. Chorus etc.

[thin flint]

So if I put instrumental in the style box and add lyrics in, it should make a short lead in to verse 1 before vocals? Trying to figure out if my style prompt of stuff like instrumental build etc is actually helping. I'm noticing that putting stuff like end in style is much more effective for me at least with these new updates.

[solar sparrow]

I didn't want to generate something good, my only objective was for him to understand the drum rhythm that I wrote and reproduce it, that's what surprised me, that he did it and generated the drum rhythm that I asked for.

[untold briar]

It can make it more likely. However maybe I should repost this, other things to try too:

intro prompts that at least occasionally worked for me:

[intro]
[instrumental]

[intro]
[instrumental interlude]

[intro]
[instrumental]

[intro]
[instrumental hook]

[intro]
[interlude]

[build]
[melodic instrumental]

[melodic instrumental]

[intro] [intro] [intro]

[intro][intro][intro]

[intro] [intro] [intro]
[intro] [intro] [intro]

[intro] [intro] [intro]
[intro] [intro] [intro]
[intro] [intro] [intro]
[intro] [intro] [intro]

If Chirp keeps skipping over anyway, try adding a bit of sung lyrics in front like this:

(ooh-yeah)
[intro]
[instrumental]

[untold briar]

I'm busy I am quite curious about this. I will talk with you later for sure. (Got to do some work right now.) Very cool.

To be clear, Chirp actually sort of followed the drum notation?

Do you know if that notation is used in song lyrics sometimes?

[solar sparrow]

In the first example he does it perfectly, the same as what I asked him, in the next he makes the rhythm that I asked him but he adds variants because he adds music to it, I forgot to tell him that it does not generate music and that it is only the rhythm of battery

My language is Spanish, sorry because it is difficult for me to communicate and I am using the translator

[runic forge]

[Long intro] works often good for me.
But I usually do something like (and occasionally get that letter sang, but mostly not):
[Long intro]
(o)
[Verse]

[untold briar]

I understand you just fine. I am quite curious about this. I'll be back later! Feel free to @ me any results if you find out more, any time.

(I'll still be online, I just have to do work, so I'm afk)

[solar sparrow]

I'm a drummer I show you here how is the drum notation, y try to simulate this in the promt

on top is the hi hat, below drums next down the bass drum

[runic forge]

Wait, you're talking with it? 😄

[untold briar]

That's just the translator, it uses "he" like Suno is a person. It's cute.

[solar sparrow]

sorry on the top the hi hat, in the middle the snare, and on ther bottom the bass drum

[solar sparrow]

Yes don´t read that

[untold briar]

BTW this format for melodies or notes sort of works in Chirp:

Hap[C4_1]py[C4_1] birth[D4_2]day[C4_2] to[F4_2] you[E4_4]
Hap[C4_1]py[C4_1] birth[D4_2]day[C4_2] to[G4_2] you[F4_4]
Hap[C4_1]py[C4_1] birth[C5_2]day[A4_2] dear[F4_2] Su[E4_2]no[D4_2]
Hap[Bb4_1]py[Bb4_1] birth[A4_2]day[F4_2] to[G4_2] you[F4_6]

I have not had time to experiment with it yet.

(It does not work well but it works more often than random chance.)

[solar sparrow]

I will try

[solar sparrow]

Is it possible train my friend suno and he could learn to red and reproduce the sound ?

[untold briar]

Suno Chirp is not open source. So only Suno can train it. But the developers have talked about some ideas for better control and fancy inputs like you are using

[solar sparrow]

great !

[buoyant cedar]

So I have been playing with suno all night - it's really amazing! But the fascination faded when I made a different song that had the exact same vocal style, baseline, and rhythm as another song I had done. This made me wonder if this is not truly a full audio diffusion model creating sound samples millisecond by millisecond, but instead it's just layering multiple clever human-understandable layers on top of one another (write lyrics, send to text->voice, assemble sound samples, etc).

Does anyone know how suno works? Is it truly generative audio for the instruments, or is it just pulling from a library of sound loops/samples for the music, and then it is only generating the vocals?

Here was one song I generated: https://app.suno.ai/song/f706782e-194d-4bb3-9912-b293b3cb8720

And here was a totally different song: https://app.suno.ai/song/5e577467-9d15-44d7-9284-c664cb1a84fc

They are almost identical

[tender eagle]

Did they tweak something on the chirp2 side on the web? The quality seems off from what it was a couple weeks ago

[untold briar]

This is a Suno user interface failure. Both of those clips were a continuation of the same exact song (they were both continued from the same exact song clip) but that's not obvious at all in the UI at a glance, you can't tell without inspected the source code, or clicking "get whole clip" on both files and comparing.

It may even a UI bug where it continued from the same song when it shouldn't have, but either way that's why they sound the same, they were the same up until that clip

And it is absolute true that keeping track of which clip is a continue of which other clip is a challenge right now.

[untold briar]

A couple weeks ago it was potentially worse, but for the last week it should be the same as it ever was.

[tender eagle]

Well today was the first time I'd used it in a little bit. But it seemed off. Like it was using the original chirp. Is how it sounded. The vocals were very Microsoft Samish lol

[untold briar]

The Simple Mode model is different, it's more tuned for short two verse outputs, but custom create shouldn't be different, at least not purpose.

[tender eagle]

Yeah it was custom. I didn't figure it was on purpose. I just wondered if something was going on with it. It was still a short output, as expected because it wasn't very large set of lyrics to work with but three different generations, six songs total and none of them sounded very good to be honest The vocals were all the bit strange sounding

I'll see if it gets better next time I try

[versed valley]

Welcome! For help, check out Suno Help

[normal oracle]

Sorry if it was answered several times before, but I can't wrap my head around the "get whole song" feature in the web-app. No matter what I am trying or clicking, every "full song" starts with the very same song I once had selected some weeks ago. I now have a large pile of useless "full songs" in my already overflowing library, all starting the same. Can someone please explain it to me? It feels like an itch in the brain to not understand this feature. 🧠

[untold briar]

Hmn. Could be a bug. Can you link two different songs (or dm to me)

[broken lotus]

Are you able to [CLEAR] the Continue From clip?

[normal oracle]

Nah, its not "Continue From", it is the "get whole song" feature. I created several "continue from" songs since the song I now have in every "full song". But I will try it anyways, thank you for your help!.

[normal oracle]

https://app.suno.ai/song/968841e3-b700-43cc-b383-6e1ff8f98362
https://app.suno.ai/song/e474051d-918e-4047-aaf7-16bbedff9497
https://app.suno.ai/song/44b4e405-4f41-4972-bd1f-48deb4fe7357

[normal oracle]

Thank you for your help!

The first song is always the same, followed by a various second song.

[untold briar]

I think this is either a bug or a terribly confusing user interface problem, which is also a bug. Because a lot of users are having this happen. I think the "continue from" clip is being hidden or maybe just not showing up clearly, or is just easily missed frankly. So these are all starting from the same song. It may also be a straight up UI bug, I haven't ruled that out. Can you try closing all the tabs on Suno website. Opening one new tab. Go to custom create. Then make double triple sure you don't see a clear button like this image or the word continue from #📚┃suno-school message