🏞｜general-with-images

[smoky oak]

maybe on paper, but not in practice. Again, bus width constriction

[wispy ether]

its a benchmark.

[smoky oak]

sigh

are you just gonna argue with everything I say?

Loses to the 3080 here, as well as the 6800XT and 6950XT which are much more powerful cards for the same price

[hasty nova]

Anyone know if VAE's work on both 1.5 and 2.1?

[smoky oak]

Loses here as well

[hasty nova]

I would assume they would

[wispy ether]

you really think i buy amd for ai

[smoky oak]

From what I have seen, 2.x has no support for VAE's

[hasty nova]

Ah, dang. Thanks though!

[smoky oak]

6000 series isn't good for AI, but 7000 series is getting good at it surprisingly

[wispy ether]

cuda still matters

and tensorcores

[smoky oak]

with current tests showing the 7900XTX could be close to 4080 performance with 24GB VRAM for $950, which is a massive win

And I mean actual tests run by this community, not by sales people

As it is currently, its already around 4070 it perf

[wispy ether]

ok ill save for a 4070

i dont wanna buy new psu

[smoky oak]

Still gonna get a 4070 after its been proven to be balls for AI, alright man

[wispy ether]

im fine with my 3060 and if 4070 is like 3060ti then its fine

i dont train

[smoky oak]

I mean, if you are ok with very little improvement, then thats valid

[wispy ether]

my pc bottlenecks with resolution in games lol

[smoky oak]

I would still go 3080 tho as it is faster at gaming for the same price, but alrighty

[wispy ether]

i got a 3440x1440 monitor

gpu bottleneck

[smoky oak]

Yeah, 3080 would be better then cause its stronger on average in gaming vs 4070

[wispy ether]

4070 doesnt bottleneck with 5700x

and 3080 is mid

[smoky oak]

3080 > 4070 in most games, and if the 3080 is mid... that makes the 4070 bad/ low mid?

I don't get what your point is here-

[wispy ether]

what

[smoky oak]

What? You said the 3080 is mid, yet its better than the 4070 in most games, so that would make the 4070 what?

[wispy ether]

3080 is older

4070 has newer architecture

some features like rtx upscale thing with tier 4

or 3

[smoky oak]

So you care about slightly better features rather than actual gaming performance? If so, then yeah sure, the 4070 has flashier features, but the 3080 is the better gaming card

[wispy ether]

the wattage tho

[smoky oak]

Yeah, i will agree that the 4070 is much much more efficient

that is a massive W for the 4070

[wispy ether]

id need a new psu

i got 750w

[smoky oak]

Ahhh, if thats the hold up then I can understand that

I was just saying how the 3080 is a better gaming card than the 4070, which is true. But if you would need a whole new PSU then I can understand going with the 4070 instead

[wispy ether]

https://www.youtube.com/watch?v=rKOWCYEx26Q

funni

[smoky oak]

Testing the GPU perf in minecraft, a notoriously CPU bound game

[wispy ether]

not always

if you use fancy mode its also gpu

or shader

https://www.youtube.com/watch?v=plt49WrpUL8

[smoky oak]

But yeah, the 3080 is a better card for gaming, but also draws like 2x the power, which is a big concern for sure

[smoky oak]

Well I mean, you are comparing a pretty bad GPU for gaming to a new middle tier GPU, I sure hope it would be better 😅

[wispy ether]

i mnean

in gaming 4070 is almost the double as 3060

+~1.8x

[smoky oak]

Yeah, thats a good point. The 3060 never was a good gaming card 😅

it was a GPU made just for the 12GB aspect

[wispy ether]

lol

[smoky oak]

Oh awesome, a graph showing how good of a deal the 7900XTX is haha

[wispy ether]

4070 missing

[smoky oak]

7900XTX, 24GB VRAM, much smaller card with less power requirements, and 3090ti perf for ~$950

and those perf numbers are rapidly improving as more people learn how to use the AI cores on the RDNA 3 GPU's

[wispy ether]

i cant find a benchmark with 4070

[smoky oak]

Cause it sucks lol

People have done their own testing

[wispy ether]

how many it/s

[smoky oak]

you can find benchmarks on reddit and such

[smoky oak]

averaged out? About 11.5

[wispy ether]

mine is 7

[smoky oak]

some as low as 8, some as high as 14

[wispy ether]

7 at 514

or 524

i forgot

[smoky oak]

Also, this graph is far from accurate-

The more I look at it the worse it gets lol

[woven dawn]

These numbers will change drastically when higher batch sizes are introduced

[smoky oak]

The Arc A770 16GB is actually about 10it/s, not 2

and the 3060ti is closer to 11.5 it/s

and the 4090 should be well over 30it/s

[wispy ether]

[woven dawn]

25its

48 at batch 10

[smoky oak]

Wait, what? The 4080 is only 25 it/s? That does not sound right

[wispy ether]

i always use 1

[smoky oak]

Ah, thats inefficient

[woven dawn]

4090

At batch size 1 it is sub-30 its

[smoky oak]

Jesus, thats upsetting

damn

[wispy ether]

idk what batch size does

so i leave it

[smoky oak]

more images at once

way more efficient

[smoky oak]

The more you do, generally the faster it goes per image

[woven dawn]

At batch size 10 the 4090 effectively hits 48it/s

[smoky oak]

my 3060ti at batch 1 does abourt 11 it/s, wheras at batch size 16 its closer to 14it/s

[smoky oak]

I wonder how it would do at 36

my GPU starts to slow down past 16, from VRAM overflow I presume

[ripe cedar]

I think batch count and batch size are best upped like squares. Ie if Batch count is 3 then batch size should be 3/4. They should be upped about the same

[woven dawn]

The it/s worsens at over batch size 12

[wispy ether]

im making a seed travel

[smoky oak]

Very interesting

mine seem to plateau at 16, once I go past that I start to go slightly slower

[wispy ether]

[smoky oak]

if you did 4/4 it would be 16 total images, but 2/8 would be a way faster way to get to that result

let me test, just a sec

trying 4/4

[ripe cedar]

I see. I wasn't sure how far up that would go. Maybe only for small numbers. I saw someone posted a test of it on reddit but only up to like 4 batch size

[smoky oak]

32.44s for 4/4

That seems sow for some reason

oh yeah

2/8 was 26.5s

or my bad

27.40

read the wrong line

let me try 1/16

[ripe cedar]

Interesting. I think I'll try different ones. I was OOM so I'll give your way a shot

[smoky oak]

My workflow always goes with making as many low sample gens as possible and refining using my upscale process

ok, so 1/16 was faster again

26.21s

4/4 - 32.44s
2/8 27.40s
1/16 26.21s

[ripe cedar]

I'm more of a controlnet guy

[smoky oak]

oh, my overclock if off

rerunning just for curiosity

trying 4/4 with OC

oh wow 26.79s

[wispy ether]

now back to my topic tho

how to i enable fp16 on a fp16 model

[smoky oak]

trying 2/8 again

[ripe cedar]

Yeah if you try 1/16, 2/6, etc you can see which one is best for your setup and keep using that to get optimum each time

[smoky oak]

it also depends on res

[ripe cedar]

yeah true

[smoky oak]

as if you do a bigger image you will OOM/overflow, which has a speed penalty

ok, 2/8 did 25.21s

[ripe cedar]

wow new record

[wispy ether]

how to i enable fp16 on a fp16 model

do i need safetensors?

[smoky oak]

1/16 took 24.77

ok, so 1/16 is still faster here, but the big takeaway is this OC affected 4/4 WAY more than the others

how weird

[wispy ether]

.

[ripe cedar]

Try it and see, it may also be a command line argument linke --no-half or something, not sure

[wispy ether]

what

[ripe cedar]

For that model I just use anything v5 its like the same thing

[smoky oak]

No OC:
4/4 - 32.44s
2/8 - 27.40s
1/16 - 26.21s
OC:
4/4 - 26.79s
2/8 - 25.21s
1/16 - 24.77s

[wispy ether]

wdym

where do i use the command line

[ripe cedar]

I'm guessing 16/1 is slower than 1/16?

[ripe cedar]

When you open it, the black window

[smoky oak]

only one way to find out-

trying

[ripe cedar]

Maybe also 8/2, etc.

[wispy ether]

i cant type anything there

thats just the bat

[smoky oak]

ok 16/1 is way slower

31.24s

I wanna try 32

[wispy ether]

that?

[ripe cedar]

It's been a while since I used it, I use the exe installer because its simpler and can add from its window instead. I think you have to edit

yeah

[wispy ether]

so what do i add

[ripe cedar]

add it next to the xformers one

--no-half

that might do it

[smoky oak]

ok 1/32 maxes VRAM and suffers cause of it

[ripe cedar]

yeah

[smoky oak]

no half uses more VRAM as a warning, and is a good bit slower

I have tested it on my card

around 15% more VRAM use and around 20% slower gens

[ripe cedar]

I made a lineart out of a 3d scene and made a depth map

[smoky oak]

ok, so at 2/16, I did 32 images in 48 seconds

[smoky oak]

I am happy with that haha

[ripe cedar]

Turned into stuff like this

[smoky oak]

And I am not dropping the quality in order to get these speeds either

[ripe cedar]

but

[smoky oak]

oh thats cool!

[ripe cedar]

Outputs got weird outlines like this after a while, not sure what happened

[wispy ether]

@ripe cedaryou halfed it

🗿

[smoky oak]

Yeah, I warned you lol

[wispy ether]

you didnt

[smoky oak]

--no-half forces FP32

[wispy ether]

what

bru

[smoky oak]

--no-half means do not use half precision, use full

[wispy ether]

how do i force full

[smoky oak]

which forces FP32

[wispy ether]

wat

[ripe cedar]

Yeah but I was trying to make it work better, you can just erase it

[wispy ether]

bruh now im confused

[smoky oak]

FP32 is full, FP16 is half

[wispy ether]

how do i force fp16

[smoky oak]

thats why FP32 uses more VRAM and takes longer

[smoky oak]

I am not sure you can

[wispy ether]

the model is fp16

[smoky oak]

Then is should default to FP16

[ripe cedar]

https://civitai.com/models/9409/or-anything-v5v3 I use this one. Its a safetensors

[wispy ether]

what about int8

[smoky oak]

let me look through the A1111 commands and see if I find a force half

oh oh

try --precision half

Hell, I wanna try it, see if I get gains

[wispy ether]

[--controlnet-annotator-models-path CONTROLNET_ANNOTATOR_MODELS_PATH] [--no-half-controlnet]
[--ldsr-models-path LDSR_MODELS_PATH] [--lora-dir LORA_DIR] [--scunet-models-path SCUNET_MODELS_PATH]
[--swinir-models-path SWINIR_MODELS_PATH]
launch.py: error: argument --precision: invalid choice: 'half' (choose from 'full', 'autocast')
Drücken Sie eine beliebige Taste . . .

[smoky oak]

Oh, I suppose there is no half option

only full or autocast

damn

[wispy ether]

is there a int8 option

[smoky oak]

not that I see

here

https://github.com/AUTOMATIC1111/stable-diffusion-webui/wiki/Command-Line-Arguments-and-Settings

Full list of CMD args

I need to get back to my LoRA training

[ripe cedar]

I want to make a short comic. Workflow will probably be making scenes in Blender, exporting to Maya, and then using the Maya plugin (Pencil+ 4) to generate lineart as well as depth maps, for both backgrounds and characters (the plugin only recently came out for Blender). Then putting those images into a 2d program and manually adding panels, speech balloons, and cleaning up. A nice trick for anime is that because it suffers from same face syndrome, I rigged only few different heads. That way, changing just the hairstyle makes a whole new character. I basically kitbash it so I just mix and match heads and bodies and clothes instead of having to do stuff from scratch.

I need to learn about instruc2pic and stuff so I won't have to manually color mismatched clothes and such though.

[wispy ether]

I broke the .bat

I cant open it anymore

I click and nothing happens

[ripe cedar]

Restart PC and try again?

[wispy ether]

I did

[ripe cedar]

oh...

[wispy ether]

I reinstalled it

[smoky oak]

I am assuming you didn't back it up before you messed with it?

[wispy ether]

💀

Now for some reason my it/s is 6 instead of 7

What is doggettx

Nvm

[smoky oak]

IIRC, its an optimization

I think its cross attention, which makes things faster and uses less VRAM, I think

Jeebus its crusty here-

eleventeen percent wotah

[ripe cedar]

Are there any "free" optimizations? I'm only using xformers for far, with a 3060 12gb

[smoky oak]

I can offer my optimized generation settings, but it does depend on the models you use

my go to is 15 steps of DDIM, which gives me very good results very fast

at least on my GPU, its the fastest out there for me

these images are all 15 step DDIM

with my OC on my 3060ti, I get about 11.5 it/s

in DDIM

[smoky oak]

@dense tapir Are you here by chance?

I am having problems with my stable diffusion and I cannot find the root

my images are getting hella crunched for whatever reason

All oversaturated, over contrasted, and blegh

[tropic shell]

Is this from a model you made?

[smoky oak]

Nope, strangely its from models I have used for a while

ever since I started working on my upscaling process, I swear my SD has been getting worse somehow

[tropic shell]

Prolly the sunset; dunno what your keywords are

[smoky oak]

I wasn't even using sunset as a trigger word, which is what makes it weirder

for whatever reason my realistic vision models are just not working well for me. Everything out of them is full of problems suddenly-

but my slight mix with realistic vision seems alright

[tropic shell]

Well, beach scene = sunset. It may not necessarily need to be that. But you could try saturation or saturated as a negative, natural lighting, or sunset

Etc

[smoky oak]

Very weird behavior

Realistic vision 1.4

This is my mix with a little real vis in it

much better

[ripe cedar]

Let's say I want to use instruct pix2pix to change an image. Should I first generate an image I'm happy with, and then input it as a new ControlNet image to change it with words? Ie change hair color of a character?

[glacial storm]

Rest

[wispy ether]

[ripe cedar]

The best deal for the money is a 3060 12gb

Anyone know how to inpaint based on a selected segmentation? Is it possible?

[wispy ether]

I already have that one lol

But i upgraded resolution and now i got 25% gpu bottleneck

Is rather getting this

Because this is not good

Bottleneck calculation

[hasty nova]

what on earth did I find https://civitai.com/models/48893/the-best-model-ever?modelVersionId=53472

I really want to download it but I am scared especially because it doesn't seem like a safetensor

[hasty nova]

...curiosity?

I just want to know what the model does

if anything

or if its a reupload of some random model

could be a dangerous file :(

[ripe cedar]

I upgraded to Pytorch 2.0 and CUDA 11.8 and now I tried what I used to but got the message: torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 62.57 GiB (GPU 0; 12.00 GiB total capacity; 5.06 GiB already allocated; 4.52 GiB free; 5.41 GiB reserved in total by PyTorch) I thought the upgrade was supposed to be much faster but something isn't right

[tropic shell]

I'd recommend checking #🤝｜tech-support

As always, too, be safe when downloading

[hasty nova]

Yep

[hasty nova]

Is it cheap anywhere?

Yeah,

But for someone my age who has no current income source I don't think I would be able to afford a new pc, especially since I won mine for free

Yeah I wasn't planning on it

[blissful wren]

https://civitai.com/models/46294/diffusion-brush-everything-sfw-nsfw-all-purpose-checkpoint-nuclear-diffusion-anime-hybrid

[faint locust]

why my pictures look like this? like blurry

[open sun]

Quick AI Concept Art tips video: https://youtu.be/0Zl_J0JKCOk

[smoky oak]

If you are using high res fix, you need to turn your denoise higher

@dense tapir Did you see that 4070's are already below MSRP cause of how trash it is as a card?

Some retailers in Canada are selling the 4070 for as much as 10% below MSRP in hopes to move the cards, while other retailers like microcenter are bundling the 4070 with 16 GB DDR4 at MSRP, and the RAM is completely free. On top of that, microcenter is also giving out $100 steam gift cards to those who buy 4000 series at all

[faint locust]

im using endless reality v11

i dont know why does it look textured

[smoky oak]

Are you using high res fix?

[faint locust]

idk what that is

[smoky oak]

Take a picture of your stable diffusion

[dense tapir]

Yes, and it is so bad Japan they give a free A750 and Microcenter is giving away a $100 Steam card for any 4k series purchase.

[smoky oak]

Yeah, so weird. But they give the A750 away with 4090's, not 4070's so I don't even know what to make of that

[dense tapir]

Well, I wish that kind of deal was around me.

[smoky oak]

Yeah, a free A750 is pretty cool haha

[dense tapir]

All 4k cards are not selling as they should as the 4060 is going to be crowned as the worst if they stick to their plans

[smoky oak]

What GPU are you running?

[faint locust]

3060ti

[smoky oak]

Don't up your batch count, up your batch size

set batch count to 1, set batch size to 5

instead of doing 1 image 5 times, it does 5 images 1 time

[dense tapir]

Well, I did the math this morning and about Christmas, or in Jan and grab a 4090.

[smoky oak]

7950XTX will probably be the move by then at least

[smoky oak]

I am not sure why your image looks so blurry without high res fix

[faint locust]

okey thanks but the result is the same

[smoky oak]

I am assuming its a mix of bad prompting with a bad model

But that is weird, never seen results like that without high res fix

[dense tapir]

The one thing I am very sad about is the loss of Tensor Cores but supposedly AMD has something like them now just no one is using them yet.

[smoky oak]

Yup

RDNA 3 has its own system of AI acceleration cores

[dense tapir]

I have seen that before

I know that fuzz

first, what is the model?

[smoky oak]

and if there is one thing we know, AMD has been consistently beating NVIDIA in terms of hardware perf for cost/watt/size, so I wouldn't be surprised if RDNA came out swinging with some nasty AI acceleration cores

Also, that opens the very real possibility of a proper AI assisted alternative to DLSS on AMD

[dense tapir]

RDNA4 is when a lot of tables will turn. 5k will beat it at what cost? 3x? No thanks

[smoky oak]

"endless reality V11" apparently

No idea what that is

[dense tapir]

I have never heard of that model

Come get in this ignorance boat with me as there is plenty of room. reality V11? wth?

endless no less too

[smoky oak]

Yeah I have 0 clue

[dense tapir]

SS Clueless

[smoky oak]

Also looks like it is in desperate need of a VAE

[dense tapir]

Agreed

You know I was reading up on AMD and Chiplets and watched a few vids. I love it

A HUGE cost cutting measure as things that need smaller ARE while the things that are slow and don't need it can get larger. Saves on costs, energy, and die stuff.

Nvidia is monolithic still which is why a 5090 is going to be around 3k (will be about 2x the price)

@smoky oak I tried Shark on Windows and Linux

[smoky oak]

Being monolithic is why intel was getting their ass handed to them by Ryzen. Now Intel is chiplet based

[dense tapir]

Them too? Oh, Farnsworth "Good news everyone".

5k is still monolithic.

[smoky oak]

Or well, intel is in some strange middle ground

They have multiple core complex's that are connected

[dense tapir]

SHARK is not for my card. They added it but not really as I could not get it to work on Windows or Linux. I asked for help and oh, well.

[smoky oak]

Just took out my earrings for the first time in a long time and man, its nice to have a break

[dense tapir]

I know on Windows it had made over 20 gigs of files so everytime I change to a different rez and a different type (vulkan, cuda, cpu) it has to compile another 1.6-2g file.

[smoky oak]

just can't have them out too long or they will close

[smoky oak]

blegh

[dense tapir]

They are supposedly lightning fast but have a spare TB.

I couldn't get any help from them so I thanked them and deleted from both OSes.

[smoky oak]

hmm

[dense tapir]

Supposedly it is the fastest

[smoky oak]

Dondal Tump

https://www.youtube.com/shorts/ZExFN4NchiU

Excuse the strong language, I just thought this was funny lol

[dense tapir]

Ahhh, I have shorts blocked

https://www.youtube.com/watch?v=Y2l0aFYysVg

[smoky oak]

What?

[dense tapir]

I was wondering why I could play it as it just went to the main screen. I do not do #shorts

btw, Nvidia's response to the 4070 bad sales is to put all production of it on a freeze for a month instead of lowering the price.

That came from China I saw last week. what a crock of shit.

I am 100% in a fog as to what he is doing, and why as none of this is how a normal company acts, or responds to their market.

[smoky oak]

Wait, I am so confused. So you have shorts blocked? What is even the point of that

[dense tapir]

Because I don't watch them and I don't get them offered to clutter up my suggestions.

[smoky oak]

I mean, I guess I get it

You really are an enigma sometimes lol

[dense tapir]

hehehe

Ahhh, that video I linked above even says Jensen is all about high prices now not cheap cards and AMD is the same only they were trying to gain market share. I firmly believe that in the near future AMD will be the Intel of the gfx market and will see Nvidia's prices and offer theirs for 5-50 USD less. Don't like it? Tough.

We will either mortgage our homes, do without, or sell it on Fifth and Main, or just do with a console. You will own nothing...and be happy says the Bond Arch-Villain wannabe.

[dense tapir]

Wen for 160 USD more than I was willing to give for it

3060 12GB

[smoky oak]

You ordered a 3060?

[dense tapir]

No, that was ebay so was already used

[bronze wave]

Make any pic

[smoky oak]

ah

I get what you mean

[dense tapir]

I bud 100 and it went for 260

*bid

market is insane with no relief in sight

[bronze wave]

Vhbf

[dense tapir]

Here we go again where a prompt I just did I reloaded via pnginfo and it is 100% different

saved the first then loaded it back up to regen for it to become the second one.

I am stumped

@smoky oak where is the option at for pnginfo txt2txt it loads the model? I no longer find that in settings?

[smoky oak]

I don't recall that ever being a setting

personally

[dense tapir]

Yep, it was because I turned it off

was so maddening when I loaded it and my model would change

[smoky oak]

not sure, it may have been removed at some point

[dense tapir]

I guess so. Well, why is this so different?

same model same all from pnginfop so wtf?

[smoky oak]

Welp, looks like several people have already beat me in terms of making decent quality upscale guides. I won't say they are of exactly the same quality as what I've been working on, but they seem to work consistently which isn't something that I can say for myself

I had a huge feeling my reluctance / struggles with ironing out this process would lead to a lack of interest in my process down the line as things got over saturated

[dense tapir]

That got me

so I understand

[smoky oak]

I put in a total of what I would estimate at least 400 hours dedicated into trying to figure out this process and get it published, only for people to achieve decent quality results with simpler processes that kind of remove the impact of what I was able to accomplish

I would have shared my progress weeks ago if it wasn't for the fact that the more you use it, the worse it works

I've still yet to figure out what exactly is ruining my stable diffusion. As of late no matter what I'm generating, the result seem drastically reduced for some reason

[dense tapir]

The one thing I realized last year was that stable diffusion is anything but stable.

[smoky oak]

Yeah true

All right then, I guess that means that I'm just going to throw in the towel on my guide and move on to something else

Sorry to everybody here who was looking forward to my process, I would have shared it if I deemed it reliable enough

[dense tapir]

Reliability is the issue with this. Look at my prompt I have tried everything and I cannot get it to repeat (not even close)

This is why I will gladly trade a few it/s if I can use the new SDP optimization for deterministic

[boreal falcon]

Are you going to switch to Dreamstudio?

[dense tapir]

Only if it allows me to run it locally.

[boreal falcon]

Oh, well its the web interface from Stability AI

And you pay

[dense tapir]

then no, and no, and a third no on principle alone.

[boreal falcon]

😄

I dont care about that kind of process tbh

[willow field]

tired swapping negative prompts and positive prompts

[boreal falcon]

Ideally and luckily that is going to be the case i will use generative AI directly in CC apps

[dense tapir]

for all of the idiots out there that are either really an idiot, or are simply lazy I LOL. I want as much control as I can get over anything on a computer. The more I can tweak means the more I am in control and the less others are in control. If people just want to press a button then they must lead rather simian like lives.

Their money, if they have any soon, so they can do them.

[boreal falcon]

I like control too

In my case i dont use SD but a different one but the controlability will be the same reason

[dense tapir]

Yeah, rather have control than be controlled

[boreal falcon]

My future workflow for art will be a long one tho

And expensive as it can get

[dense tapir]

Aye, it gets expensive even for analog canvas art.

[boreal falcon]

I mean it doesnt have to be that expensive

[dense tapir]

I about dropped dead over a decade ago when my step son would buy canvases, or his mom did for him. I was like WHUT? shit they were expensive.

I can only imagine now as they would be way more expensive and be made from left over rubbish that someone put together with spit.

[boreal falcon]

Can be a very expensive hobby depending on some factors

[dense tapir]

a tube of pain was expensive too

*paint

[boreal falcon]

Im fully into digital art tho, 2D as well as 3D

For both there are free apps

[dense tapir]

Yeah, but let's be honest Blender sucks in comparison to a for pay 3d program

[boreal falcon]

Well i want to get into that expensive pipeline for a reason

[dense tapir]

damn, it keeps changing

[boreal falcon]

Right now i only subscribed to Adobe CC, but very soon it will get much more and expensive

[dense tapir]

I have to admit the AI coming for CC does make my jaw drop

I am glad it is coming as it will let AI become more acceptable then the anti AI crowd will fuck off.

[boreal falcon]

3ds Max and Maya, let alone their expensive but top notch plugins, Zbrush, Marvelous Designer, Substance 3D and eventually some more

[boreal falcon]

Im using it

[dense tapir]

Yeah, it is in beta still is really damn good

[boreal falcon]

They will eventually get rid of that web thing and integrate all the features into CC apps

Or maybe they keep the web UI

[dense tapir]

I do not like Maya since Autodesk took them over which is 3dsmax too. I always preferred Cinema4D once AD took them over.

[boreal falcon]

Cinema seems to be very good too

From Maxon ofc

[dense tapir]

Yes

[boreal falcon]

Autodesk is kinda the Adobe of 3D

[dense tapir]

AD is just as much of a dick company, yes

[smoky oak]

Autodesk is a bad company, but nobody is worse than adobe

[boreal falcon]

Luckily they now have those indie licenses

[smoky oak]

At least auto desk has free versions of their software

[dense tapir]

well, not really truly free but simulated free

[boreal falcon]

So its much easier for me to afford Maya and 3ds Max and pour some big amount of money into their plugins

[smoky oak]

Fusion 360 is free at least

[dense tapir]

as long as you remain glued to us it is free. I loved the backlash when they stripped it to the bone

[smoky oak]

It's what I use

[boreal falcon]

That was expected, even before they announced that

[dense tapir]

I never like AD, nor its products. I tried F360 and saw where it was headed and it never worked with me.

[smoky oak]

I use F360 a lot as it works exactly as I need it, and its free, so I have been happy

[boreal falcon]

You know what i dont like? The annoying cult of the Blender community

Generally the open source cult

[dense tapir]

I like products that work with me, and for me. I prefer Solidworks

[boreal falcon]

But Blender is the most prominent

[smoky oak]

Although I just started testing a brand new alternative to blender and fusion 360 called plasticity, and its got my interest.

[boreal falcon]

Oh heard about that one

[dense tapir]

I heard of that one last year

[smoky oak]

As somebody who was part of that cult and used blender for 5 years, its really not nearly as good as people think it is lol

[smoky oak]

It just fully launched, and its damn good

its not full CAD, and its not full Poly, but its like the best of both worlds IMO

[dense tapir]

I just couldn't get with the workflow of F360 it was just too foreign for me.

[boreal falcon]

Not that its not good, but man those claims are so annoying

[dense tapir]

Like RHINO3d?

[smoky oak]

and its complexity performance blows blender and fusion out of the damn water lmao

its also just insanely intuitive

[boreal falcon]

As if Blender beats Maya, 3ds Max, C4D, ZBrush in their fields

[smoky oak]

for most things

[dense tapir]

Blender was never ever made for what it is being used as which is why its UI is so convoluted.

[boreal falcon]

Also lets not mention the plugins

[smoky oak]

I will say out of all of those, blenders biggest strength is sculpting

[boreal falcon]

Blender plugins vs Maya and 3ds Max plugins in comparison?

[smoky oak]

sculpting is the only area where I see professionals try it out and go "wow"

Like full on Zbrush users are like "Damn, I wish Zbrush had this" to some of the features

[boreal falcon]

@smoky oak and even there zBrush is the king

[smoky oak]

Zbrush is good for some parts of sculpting, bad for others

for example, Zbrush isn't a 3D sculpt when you are working on it, so you can't utilize realistic lighting or rendering to help with volume

Whereas in blender you can drag lights where you want them and have them interact with the sculpt

Thats usually the biggest thing I see Zbrush users praise when they use blender sculpting

[boreal falcon]

ZBrush is anyway a pipeline tool

I wont use it alone

[smoky oak]

yeah

I'm just saying out of all teh things blender does, it does sculpting the best compared to the paid alternatives

[boreal falcon]

Blender is jack of all, master of none

Which is okay

Especially for amateurs

Its free too

And its developing and having a big community

[smoky oak]

Blender is really good for some things, like rendering performance, and sculpting, but also really bad at some things like simulations and optimizations for scenes

[boreal falcon]

A big plus

[smoky oak]

The reason i stopped using blender is cause I realized that the devs don't care about the users like they used to. They don't fix blenders glaring bugs, and instead add new half baked features nobody wanted

[boreal falcon]

I will just split my workflow into several apps and plugins incorporated into them

[smoky oak]

Plasticity 3D has to easily be the most impressive next gen 3D software I have seen in years

[boreal falcon]

Will it be expensive? Oh damn it will be.

[dense tapir]

Why is it so many companies do that? I bring that up one XYZ company and program and get from people that is the best way or it stagnates, yadda yadda

[smoky oak]

The problem with blender is how massively fragmented it is

[dense tapir]

fix the damn bugs first then worry about adding new bugs

[boreal falcon]

Its community is really cult like

[smoky oak]

Its not a team, its a ton of small cliques of people good at one thing working to add what they do good at the expense of everybody else

[dense tapir]

Ahhh, like the automatic1111 devs?

[smoky oak]

For example, the render pipeline which got overhaulded a couple years ago (its really damn good now), had to be postponed for 10 years because the lead dev left, and he was the only one that knew how to work with his own code

[boreal falcon]

Which render do you guys useM

@dense tapir i guess you will try the 3D to image thing from Firefly? ^^

[smoky oak]

For me it was always Cycles X once it dropped. I tried redshift, luxcore, octane, and a couple smaller ones like apple seed, but Cycles X always just did it so damn good

[boreal falcon]

Aah okay 😄

Didnt touch Blender for centuries xD

[smoky oak]

Blenders new Cycles X render pipeline is a big deal

or at least it was when it dropped

[dense tapir]

Not sure but I know two things I want in PS that is coming. AI remove (far better than what we have now) and the change scenes.

[boreal falcon]

I guess i will use the native one from Maya/3ds Max, Arnold and for specific things Krakatoa (for particles for example)

[smoky oak]

CPU/GPU parallelization for rendering. Much less VRAM used, on average over 4x as fast, new features like grouped lights, and basic level support for new systems like advanced caustic and diffraction engines in the future. A whole dedicated API to running on AMD GPU's and CPU's with iGPU's

[boreal falcon]

The issue is my laptop isnt the strongest

But im primarly into static artworks right niw

Now

[smoky oak]

Cycles X shook up the industry big time when it dropped

[boreal falcon]

Not AAA animating and films

[smoky oak]

Oh, another thing that blender is actually top dog in is geometry nodes

nobody else has anything like it, and it shows

[boreal falcon]

Wdym AI remove? Aaah the beta tool

Kinda the better version of content aware fill

[dense tapir]

I hate this new gradio as every app that uses it I get weird clicks loading random shit

way better

side by side no comparison

[boreal falcon]

Didnt try it yet although i have the beta app installed

[smoky oak]

@boreal falconIf you wanna do some trippy stuff and not worry about your PC's performance, blenders gemoetry nodes is unrivaled in performance and capability for abstract creation

[dense tapir]

You should as I use content aware fill all the time and really the only thing I do use in PS 99% of the time

[boreal falcon]

Didnt look up for that before tbh

[smoky oak]

Billions of faces like its nothing, real time responsivity, unrivaled creative possibilities

[smoky oak]

if you know anything about houdini, think that on crack mixed withe very aspect of modeling, lighting, and simulations

[boreal falcon]

Dont use it often

But could change

[dense tapir]

Sytan, you notice that even the 2k series cards are hellishly expensive now?

[boreal falcon]

Houdini is some end boss software

[dense tapir]

always was

[smoky oak]

Yeah, now imagine that with what I just said

The stuff that geometry nodes can do makes houdini look like a joke honest;ly 😅

[boreal falcon]

Im a beginner in coding and math is something i used to hate but it seems like i need to start loving math

If i want to do some stuff

[smoky oak]

Its all GPU processed with instancing and parallelized processing acrosss CPU to tear it apart for teh GPU to rip through

[boreal falcon]

@smoky oak will look up on that one today after the work 😄

[smoky oak]

I wasn't kidding when I said BILLION of objects/faces/animation paths

[boreal falcon]

Damn i have to wake up on 4 hours lol

In

[smoky oak]

https://www.youtube.com/shorts/GZgoNMzZCgA

Somebody made a circl packing algorithm using geo Nodes in blender

https://www.youtube.com/watch?v=WYxzPpx_qlI

Everything in this video was done with Geo Nodes

[dense tapir]

Does anyone have a different checkpoint merger program as auto's is now beyond frustrating? I load merge and can't merge a second time I have to merge something else then come back to it. Not sure what they fucked up a couple of months ago.

Ahhh, I think I see how to work it out. I have to hit the update button now even though there it is.

it's a bug

[boreal falcon]

@smoky oak it seems like Autodesk has something like that for Maya too (not sure about 3ds Max)

Looks very cool tho from what i see for Blender

[dense tapir]

3ds max I am surprised is still around as it was tried to be killed a few times over the decades.

[smoky oak]

Kinda, but also not really. This system allows for intercommunication between lighting, mathematical functions, sumilation calculations, and all sorts of other things. I can't really explain how next level geonodes are

[boreal falcon]

Its not going anywhere tbh and the 2024 updates are cementing it too

Plus those beasts of plugins available

beasts*

[dense tapir]

yeah. What beats vray these days?

[boreal falcon]

i will use Arnold likely but lets see

[dense tapir]

Ahhh, did Arnold finally improve?

[smoky oak]

https://www.youtube.com/watch?v=xNMACKK8Nzs

[boreal falcon]

Its shipped in Maya and Max, so no cost

[smoky oak]

Not from anything I have seen

still slow and kinda outdated from last time I saw it

[boreal falcon]

they will update it soon i think

[dense tapir]

Damn

[smoky oak]

Most major companies use redshift or octane now it seems

[boreal falcon]

Maya and Max got the update

[dense tapir]

Arnold was always so slow and kinda buggy too

[smoky oak]

Yeah, its ancient

its almost as old as clockwork (I think that was its name?)

Blenders original cycles is over a decade old, but they brought the heat with Cycles X being at the cutting edge now

[boreal falcon]

V-Ray seems to be used often for Max

Especially architecture professionals

[smoky oak]

oh man, V-Ray is ancient too

[boreal falcon]

Corona catches up

No. Not that corona

Hehe

[smoky oak]

Corona render, I know

[boreal falcon]

Joke aside

[smoky oak]

Since when? Thats been Octane renders market for years

[boreal falcon]

For a bit longer with Max users

[smoky oak]

Octane render is prodviz archviz king for like 5 years

Octane can do diffraction and caustics, not sure fi V-Ray can do either

Ok, it looks like it can do caustics, but not diffraction

[boreal falcon]

I will be a generalist and specialise later on so i will go for Arnold first and then see

[smoky oak]

I don't know many people whoa re just starting out who decided to jump into basically one of the most expensive softwares available in the industry lol

Then again, I also bought FL studio when I was 11... Soooo

[boreal falcon]

Also im happy AWS made Thinkbox not only free but even open source now

[boreal falcon]

Im such a guy lol

[smoky oak]

Understandable lol

I dropped like $300 on FL studio when I was 11/12, so I get it lol

[boreal falcon]

What is FL

With 11?! XD

[smoky oak]

Fl studio, music production software suite

[smoky oak]

Yeah, I have always collected and saved large sums of money my whole life

I have few things, but the few things I have are pretty damn nice cause of it

[boreal falcon]

Lmao, damn

[smoky oak]

Music is still my passion I want to pursue professionally, so it was a good purchase in hindsight lol

[boreal falcon]

Thats good to hear

[smoky oak]

Nobody beats the value of FL studio when it comes to professional software

$300 for life

and I mean for life

[boreal falcon]

People dont understand it when we put that amount of money into such hobbies

And it annoys me

[smoky oak]

if you bought it back in 1998, you are still getting every single major release and feature to this day, and still will 50 years down the line

I don't know of any other professional grade software doing that

[boreal falcon]

Because they themselves are burning their money for cigarettes, weed, alcohol etc while looking weird at us when we spend a ton of money on art for example

Screw them

[boreal falcon]

Thats based

[smoky oak]

I have close friends who are industry professionals, and I grew up with them and learned from them, so I do have a bit of an advantage

I am a sound designer/synthesis engineer. Thats my focus at least

[boreal falcon]

Nice

[smoky oak]

I make the sounds lot of other people use

and I am working on an album, but recent life has not been too nice to my music side

[boreal falcon]

I wish you nothing but success there

[smoky oak]

I have been wondering what it would be like to use my sound design knowledge for things like games

Like I could imagine a super cool dark themed game with neon like enemies fueled by the sounds I make lol

[boreal falcon]

I wish i did pursue consequently what i wanted before, but i listened to other people who were not even skilled in those things

[smoky oak]

Weird sounds like this (on the melodic side)

Thats my magnum opus, even if it doesn't mean much to people who don't sound design lol

[boreal falcon]

But now I really gotta go to sleep, i have less than 3 hours time and then gotta work

Gotta talk today again with ya guys 🙂

[smoky oak]

3 tones, 4 notes, 41 hand crafted effects, and 14 real time automations

[smoky oak]

Sounds good

sleep well

[boreal falcon]

Thanks you too 😄

[smoky oak]

2 little narsty sound's I recently designed

They were based off David Productions sound design in the Anime's they produce

Just very abstract is all

[dense tapir]

@smoky oak 😦 ive had the rx 6700xt, 6600xt and the premium 6950xt. basically no matter which card you get they're all garbage for any type of Professional work. AMDs own " PRO RENDER " plugin doesn't even function properly in blender.

[smoky oak]

PEENG

[smoky oak]

Where did this come from?

[dense tapir]

Someone just responded to a comment I made on Youtube.

[smoky oak]

Pro render is an open source renderer that AMD themselves made. Its not the API I was talking about in relation to acceleration for Cycles X, which is the built in renderer

AMD used to use OpenCL, but their new one is a weird name

[dense tapir]

My response to this was that it has been over three months (since the OP posted) did it ever come? His post was "The reason the Radeon cards are so much slower than Nvidia cards in Blender, is because the Ray Tracing (RT) cores are not utilized at all. They said that support for the RT cores is potentially coming in Blender 3.5 in a couple of months."

OpenCL had promise but just petered out

[smoky oak]

But... the Whole point of the HIpPip API is to use those ray acceleration cores-

Thats what its called HIpPip, or HIP shortened in blenders API section

[dense tapir]

No idea just asked him and then this guy chimed in

[smoky oak]

See

Its made specifically for AMD GPU acceleration

Now i have not been able to use it, but I remember when it first came out, it was a solid improvement over the uselessness OpenCL brought

[dense tapir]

Yeah, but he is saying it is still shit

[smoky oak]

No idea

To be fair

[dense tapir]

the op did n't answer

[smoky oak]

most people in blender don't even know to switch to GPU acceleration in general

😅

[dense tapir]

oh, shit

lol

[smoky oak]

Yeah

[dense tapir]

that is weird and funny

[smoky oak]

I saw so many people in the blender community like "I got a 3090, and my renders are no faster"

And then I had to be like... My guy, you are rendering on your 2600x

Or they switch to cuda acceleration and get mad that their perf isn't that good

CUDA accel isn't made for RTX cards LOL

OptiX is what is made for RTX cards, and its way faster and way more power efficient than standard CUDA accel, as it uses the RT cores

[dense tapir]

RT cores hence Tensor Cores.

TC are just drop dead sexy

so fast they wait around over half the time for the internal memory to give it more data

[smoky oak]

Tensor cores and RT cores are very different, but while the tensor cores on RTX cards are useless for rendering, they do offer another massive bonus to blender

which is

just a sec

[dense tapir]

wouldn't know as I have neither

[smoky oak]

The sexiest thing lol

[dense tapir]

I would gladly give up 10k of the cuda cores for 5k Tensor cores on the 4090

[smoky oak]

5k tensor cores-

[dense tapir]

can you imagine?

[smoky oak]

The 4090 has like... 380 IIRC lmao

[dense tapir]

yep

[smoky oak]

I think the tensor cores are far too big to do that

[dense tapir]

here, take 10k of the 16384 cuda cores and slap me some luving 5k TCs

[smoky oak]

its not like 1 cuda core = 1 tensor core

[dense tapir]

I know

[smoky oak]

But yeah, I get you 😅

[dense tapir]

why I said 10k for 5k

[smoky oak]

that would be outrageously hot

[dense tapir]

hell, I will even sweeten the deal and 10k for 3k

[smoky oak]

Its honestly probably a lot closer to 1 tensor core to 100 cuda cores

[dense tapir]

roughly, yes

in speed

[smoky oak]

I don't really know, they are wayyyy different situations

[dense tapir]

1TC waits for data over half the time so aren't even fully utilized right now on even the 4090

[smoky oak]

Which is exactly why the 4070 sucks so bad in SD, cause its bus is so constricted that it can't even operate without being suffocated

[dense tapir]

we just can't give it enough data fast enough because they are so ripping fast

[smoky oak]

and also why VRAM overclocking for SD is all that matters

It was the same way with crypto mining

[dense tapir]

well, I have a 1060 with 192 bit bus and 6gb of ram and the 4060 with 8gb has 128 bit? wth?

[smoky oak]

you downclock the hell out of your core and overclock the hell out of your VRAM

[smoky oak]

Wait woah, only 128 bit? Thats criminally bad-

[dense tapir]

yes

[smoky oak]

That doesn't sound right...

[dense tapir]

it is right

4070 has 128 too

[smoky oak]

that RTX 3050 VRAM

[smoky oak]

no, its 192

[dense tapir]

you are right same as my 1060

4060 has less ram so they shoved it 128 bit bus

[smoky oak]

I was gonna say

so VRAM capacity and bus are directly linked, I am not sure if you knew that

[dense tapir]

it is BS and 4070 is DOA the 4060 is DOA before it has even been born

[smoky oak]

They can only be multiplicative

[dense tapir]

they are playing games now though with a buffer chip in it so "we don't need as wide of a bus". My ass

[smoky oak]

so for example, the 3060ti has 8GB VRAM and a 256 bit bus. It could run on a 128 bit bus, but that would be way too slow

Or it could run on a 512 bit bus, but that would be way overkill

Thats why the 3060 has 192 bit, cause its either half that, or double that

[dense tapir]

12gb should have 384

[smoky oak]

and then the 3090 has 384 bit, which is 2x 12GB or the same speed ratio per bit

[dense tapir]

48 gb cards have 4k bit bus I believe it was (which is insane).

[smoky oak]

So when you said 8GB 128 bit I was like- that 6GB VRAM bus width lol

[smoky oak]

Hell, the AMD Vega 64 has 24GB VRAM at 4096 bit bus

[dense tapir]

I know the 4060 is DOA before it has been born.

499 too

maybe 449 thanks to 4070 but all bets are on 499

it is a shit card as is the 4070

I swear my 6gb is 192 bit bus so why is the 4060 only 128?

it has 2 gigs moreram

oh, ffs this is the ti 4060

Jensen is trolling us

[smoky oak]

The Vega 64 also has HBM2 memory, which makes the 3090/4090 look like snails

[dense tapir]

yes

[smoky oak]

at 24x GB it should be around 4tb/s

[dense tapir]

HBM is on their pro cards Ampere

[smoky oak]

HBM2 is hot stuff

[dense tapir]

yes

[smoky oak]

Man, if the Vega 64 wasn't a flop... Just imagine

[dense tapir]

one of the cards I was looking at for servers is DDR and HBM

So far most of these numbers have been confirmed but not all. Bit buss and cuda has been for sure.

[smoky oak]

That still sounds really off, 8B VRAM should be incompatible with 128 bit

[dense tapir]

No, it has been confirmed by Jensen

8 GB

[smoky oak]

Both 8GB and 128 bit were confirmed directly?

oh wait no, thats right

my bad

[dense tapir]

from vids I watched yes

[smoky oak]

128 is right

its either 128, 256, or 512

they can only stack, I was stuck on 192 for 6/12 GB

Yeah, my GPU is 256

[dense tapir]

that 4060ti (no less) is gonna be slow on transfers

Just sad for a Ti

[smoky oak]

if that really has a 128 bit bus, it is going to be constricted even harder than the 4070

jesus what the hell are they doing...

[dense tapir]

I agree

I do not know. It is almost as if Jensen is trying to destroy the gamer's market

say it was a flop and shut that division down to concentrate on the AI side only

This is why everyone has said the 4060ti is going to flop worse than the 4070. You seen the vids of all the Nvidia returns?

open box specials are like 50 less dollars. All BS

[smoky oak]

Yeaahhh

[dense tapir]

people returning 4090s boggles my mind but there were more than one too.

maybe buyer's remorse

Here we go

@smoky oak Dev for Shark mentioned - We just exposed a REST API end point for Blender. In the following months we will integrate it deeply into blender obviating the need for the cycles path.

I have no idea what any of that is.

[smoky oak]

No idea here either

I ahve been out of blender for about a year now

[dense tapir]

Aye

[split rover]

Sometimes I miss messing around with blender, fun times

[tame oxide]

When will stability.ai plugin be able to use depth to image or is that working now?
Also will it auto be able to project textures onto meshes?

[dense tapir]

I am not even sure what that is

[kind quartz]

heat waves?

[dense tapir]

Aye

Hell is hot

[kind quartz]

will freeze soon 🙂

rather spoiler it than having problems, but i think it is o.k. image.

[dense tapir]

Looks fine to me

What is CF Spark?

[kind quartz]

CF spark? I dont understand 😄

[dense tapir]

It is like MJ, or Dall-E or SD

[kind quartz]

SD

[dense tapir]

I just only now heard of it

[kind quartz]

umbrellas are difficult

[dense tapir]

Yeah

Well

[kind quartz]

it is great model for caves 🙂

[split rover]

[kind quartz]

i like those things, IF some of 8k mountain is pyramid

[split rover]

you found my house

[kind quartz]

🙂

[vocal karma]

beautiful pale cyberpunk female with heavy black eyeliner, blue eyes, shaved side haircut, hyper detail, cinematic lighting, magic neon, dark red city --ar 9:16 --testp --upbeta

[kind quartz]

😄

1024x768, just trying. But it seems far better than 768x512, and it is 1.5 version

[wispy nest]

yea I can see it

without searching too, it's on 2nd row

[glossy herald]

let's move NSFW talk and screenshots of filters to PMs though please 😉

yeah sorry, party pooper was my master degree

[wispy nest]

what's wrong with filters tho

[dry crow]

The cat ball

[glossy herald]

generaly, we aren't pushing or endorsing people using SD on NSFW. so we try to divert talks on it to somewhere more fitting, and not promote tips on how to NSFW

but yeah, the filter names are all clean, np

could use a Trigger warning/spoiler tag for the themes if I really wanted to partypoop

[wispy nest]

Anyone tried this thingy?
https://civitai.com/models/22220/cardos-animated

weird mix...kinda curious

[glossy herald]

it's clean though

nice renders

[wispy nest]

I don't understand why they mixed level4 in

everything else is cartoon-ish , level4 realistic

[glossy herald]

maybe they use it mostly as "regularization" ? I mean, training on only cartoon has effects on the rest of the model, and if you try to prompt photo realistic it gets a lot worse. they may have wanted some balance. not sure. strange choice

[wispy nest]

¯_(ツ)_/¯

New interesting thingy too
https://civitai.com/models/48895/revsoap28dvaebaked

based on rev_animated too

[tropic shell]

Skeletons enjoying their day together

Kinda whimsical

[kind quartz]

yes they looks like nothing can separate/divide them
*i am very bad in english 🙂

[willow elm]

how can i create pics?

im new 🙂

can u help me? which tool do you think is the best one?

[hazy kite]

Classic astronaut pic

[blissful wren]

https://civitai.com/models/49836 released a lycoris that enhances images

[dense tapir]

@smoky oak I hope people will continue to stand their ground about his tyranny.

[dense tapir]

Heh, he used a comment I commented on and I saw my response to it. People are wise to Jensen hating us.

[ornate geyser]

Sun

[dry crow]

Nice ones!

[hasty nova]

This has really nice texture

[hallow plover]

anyone knows how to use LYCORIS?

[hasty nova]

pretty sure you just download the extension for it

then use it the same way as any other lora

[hasty nova]

parameters: aaaaaaaaahhhhhhhhhhhh Steps: 15, Sampler: UniPC, CFG scale: 7, Seed: 1756271263, Size: 512x768, Model hash: d0d0902731, Model: digitaldiffusion_v1, Denoising strength: 0.7, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+
yes the prompt was aaaaahhhhh

[glossy herald]

someone got ideas for false pokemons ?

I got a model that is trained on quite some creatures, and photo realism, that I'm testing

well on lots of other things too

but I wanted to test the "pokemon" idea

like this cat octopus in a suit x)

With your prompt :

Photography Cyborg Snake-lizard wiith many legs with metal grip claws, half metal half flesh, mythological creature, half metal half flesh, blood on ground, blood dripping from tongue, blood dripping from moon, blood splatter on sun, blood dripping from clouds, background blood flowing in red stream of blood

But I also trained this specific picture that you submited to the #1087493722645725184 in this very model I'm currently using, with a simpler caption fitting to the context I was in : "a painting of a snake with blood everywhere, dripping, menacing, horror, Unmythical Creatures" (I pinged you on it in #1100010386004131980 earlier today)
here are the corresponding results

let's try it as realistic

I'm doing other tests on the model, running lots of prompts to get some previews of where it needs to be upgraded still.
i'll be back to the snake soon, but the list of prompt I just sent should take a little while, and the creatures are towards the end of it

this is on the 1.5 base, I can't wait to see how it does on 2.1 🙂

PoW : "abstract realities"

PoW : Never Ending loop

Alice in diffusion land

the prompt for the robot dolls, inspired/trained on this one : #🏆｜winner-gallery message
is, for this model, "robotic marionette alice and rabbit on a couch, flowers, in a room, painting, Alice in Diffusion land"

sorry, I'm not sure witch one you were referring to maybe

this one ?

this one was "Alice girl with neon rabbit hears, Video game rendering, unreal engine, Alice in Diffusion land"

leaving the realm of Alice, next test phase is on the "catstravaganza" theme

[glossy herald]

yeah, the impact of "alice in diffusion land", or tokens I would make up, will be high in my trained model, and not work on SDXL. it's normal though.
And as a base for training, it's going to be the best no contest, for whom it's available

next series of test is "Chaos and Order"

next is quite dispersed themes. it's from the community participations in #1099245477007990825

little break:p sorry for spam

let's go on Cosmic Horrors. A black and white dataset mostly

[edgy shard]

lol

[glossy herald]

love the cats in suits

[dense tapir]

A hybrid beast

no

digital diffusion

[edgy shard]

@glossy herald can I message you man? I'm stuck and need some tips

[glossy herald]

it's 3AM for me, I'm about to log off, I'm not going to solve problems right now tbh ^^

[edgy shard]

hah alright. you know where i can go to get some suggestions?

[glossy herald]

depends on your type of problem

we have quite some channels

explain a little here, I'll redirect you to the good place

and maybe answer there when I wake up x)

peter to the rescue 🙂

[edgy shard]

trying to overlay a character onto a pose. i have a character subject and i have an idea of what pose i want them in

[glossy herald]

by the way peter, I have sent a contact towards you today, they were looking into training photo realistic models, I hope it's not a problem

well, good night everyone 🙂

(and best place for such question would be #1011634831467221033 )

I see ways to do it, check the pinned message for some ideas, but training may be needed depending on the quality level you need in the end

[edgy shard]

thanks dude

[glossy herald]

yeah, try your hand already here

peter is good on people pictures

[edgy shard]

sure. so i have a character, like this

and i want to put her and the gun into a pose similar to this

but i'm getting a lot of artifacts in images, this is the closest i've gotten using straight prompting

and it's not a very good character match

[edgy shard]

which one? the character herself or the pose?

woman, frizzy yellow hair, thick body, (heterochromia green and red), cosplaying spaceship, holding giant needle gun, fullbody, cape, galactic background

hm?

alright

[dense tapir]

Your prompt

[edgy shard]

Haha

[blissful wren]

https://civitai.com/models/46294/diffusion-brush-everything-sfw-nsfw-all-purpose-checkpoint-nuclear-diffusion-anime-hybrid

[edgy shard]

that is an actual image

this is what i used to get her kind of in that pose though: highly detailed woman, kneeling down on one leg, aiming down the sight of her "oversized needle cannon" preparing to fire, (((frizzy yellow hair))), thick body, large bust, (heterochromatic eyes, left green, right red), cosplaying spaceship, fullbody, cape, ((on top of a spaceship in outer space, nebula background)), high detail

[hasty nova]

you could try controlnet

[edgy shard]

i have never used controlnet but i have heard people talk about it

[hasty nova]

try looking up a quick tutorial, it's pretty simple and exactly what you need

[edgy shard]

okay well i can learn. is there a tutorial anyone here suggests for it? am still curious what @stone cipher is coming up with though

[glossy herald]

a turtle unicat

[edgy shard]

so i tried inpainting stuff earlier but was unsure what i was doing