#🏞|general-with-images
1 messages · Page 157 of 1
u was there for gamescom and cs major or only one of them?
ik a fren who went to Cologne for cs major recently
Today I visited the city and GamesCon tomorrow
W have fun
Thank you. The city is pretty nice
cant believe NaVi lost Cologne in finals
I am not really a gamer. But I am ASUS guest. And the technology will be interesting for sure
indeed it will
Waiting for 5090 😂
i may get refurb 3090
Yeah.... Expensive hobby...
indeed it is but £580 GBP for 3090 isnt bad i will say
How much vram?
24gb
24 sounds good
ye
proving that if people aren't told, they really can't tell the difference any more
Text might be a bit to perfect and photo to bad for A. I.
the image quality is a bit too high for it to be AI
but its confusing cos the camera artifacts gave some people AI-face
Have to see it on a big display🤣
I feel like a massive idiot when bro said real images from Cologne 😭
its a really good thing cos it shows the tech is getting good
don't. it's just an example of how good AI has gotten that people automatically assume a photo is generated, or automatically assume a generated image is a photo. and a way to prove to people that say they can tell that no, they can't. not if they assume it's something else for some reason
That is true
do you guys know https://thispersondoesnotexist.com/
Its Stylegan 2 from the year 2020
you have to refresh to get a new image
I still prefer it to this day to most stable diffusion photo portraits
love that site. pulled a LOT Of images down off it, and used them to train a lora with
Yh it's actually crazy
giving up on GANs may have been the biggest mistake
no one's given up on gans - they just got sidetracked.
yeah
Adobe still does them
they dropped videogigagan this year
I actually think within the next 2-3 years diffusion will be dropped for autoregressive
stuff like Lumina-mGPT
the LLM simply writes out the pixels one by one
or writes out functions that produce the pixels
others are still doing them too.
ah okay maybe I have underestimated the supply of new GANs
there was one EMAD mentioned, not released, that SAI made
every now and then a new ERSGAN model shows up on openmodeldb too
oh I forgot stuff like insightface as well
they're being used for all sorts of stuff, just not a lot of image gens right now
ThisPersonDoesNotExist had some interesting failure cases.
I called these ones "shoulder demons".
I don't have any saved, but there was another whole category that I called "is it a hat or is it hair?"
And there were others with ambiguous gender or even ambiguous age (have you ever seen someone who simultaneously looks like a young child and an old woman?)
yeah, and it love to make messes with glasses
I assume that stable diffusion?
workflow is in the image. click the image, click open in browser, right click on the large image, save it, load it into comfy. it's flux
How it feels when there was a long line at the store
cinematic ultrarealistic grey alien, found footage, film grain, desaturated, glitch, blurry, alien, woods, grey alien
it's schnell, so it's going to do the poorest job. but it's not bad. my prompt is that string of terms
you can get the dev model for free right
and everythig is in the specific order on purpsoe
good morning coffee with a buzz
yes. providing your machine will run it
you're not running it
do you have a car or anything like that
cinematic ultrarealistic grey alien xenomorph, found footage, film grain, desaturated, glitch, blurry, alien, woods, grey alien
you could sell the passenger doors
and the stereo
that and maybe the passenger seat is def enough for a 4090
and saran wrap for the holes
you will have to educate me more on the importance of prompt order
the stuff that's most important to your image put at the start
you're telling me I can get a weight deduction for increased performance AND be able to run Flux as result of the same action?
brilliant
yes lol
can't get your car totalled because of airbag replacement if there are no airbags
same prompt, but with flux dev on my machine inside comfyUI
it's going in the right direction but I want it to look more like..
can I post a youtube video for reference
shit if you commit to selling the seats hubcaps and stereo that's prolly enough for 2x4090
and airbags
just chill in a lawn chair and hold a beach ball in your lap in case you crash
holy fuck that's the funniest thing I've heard in like a week
and how did you know I have an Accord
it was obvious
he is everywhere
Swapped over to Miniforge.
Now I'm using phi 3.5 3.8b for prompt enhancement.
It's not bad.
If there wasn't any stupid censcorship that won't even let me type "breast pocket", then I think imagen 3 would be the best image generator that's free, doesnt require a good pc, is fast, understands prompts very well, has extra ai features
chest pocket
I had to write patch pocket
I'm gonna put flux up to the challenge with the same prompt
This was heavily edited by me in photoshop but the text image and background image were both flux
jreg tech support
Is ThIs FLuX?
Real photos 😂
It seems controlnet stuff doesn't work with the gguf models for now
At least from what I can tell.
Loras do though.
this is one issue with going the gguf route, stuff may need re-doing
As an AI artist I am their giest
I'm literally no longer making my prompts. Phi 3.5 is making them for me.
I gave it a sys prompt and a usr prompt
🤷♂️
TBH using LLMs to write prompt is better
ComfyUI Ollama-Describer i2i
?
beautiful murals
Flux. Most of this batch weren't really what I was prompting for, but this is hilarious!
the faces look quite strange though 😄
nahh I think that could pass as street art
one thing AI will probably always be better at is random/glitchy textures
the human mind can't really come up with randomness on par with a computer
out of curiosity, what on earth did you prompt for?
Should be in the metadata, but it was along the lines of “action GoPro photograph of Will Smith standing on a shark (punching a venous membranous flesh sack:1.5) in San Francisco Bay”
I had to up the weight in the middle to get anything other than a black guy standing on a shark.
flux doesn't know most names... you did well to get what you got
This was an older prompt of mine from SDXL and I think it got the idea a bit better than flux.
Yeah, I know. But there is barely any flesh sack, let alone punching.
what do you get if you just run: venous membranous flesh sack
as the only prompt
I’m at work right now, but might try it later.
serious sam?
Has anyone gotten ipadapter/controlnet flux models working with GGUF?
I really don't want to have to download the base/fp8 variant of the model just to use them
Was that supposed to be Fallout?
bruh didn't read the words literally right above the image
Thas' Serious Sam, Chief.
serious sam 1
Swapped back to Llama 3.1 8b 8_0 for prompt enhancement
It just knows context better lmao
I never heard of serious Sam
Forgive me 
it means you didn't have a childhood😭
Or he's younger than us.
I'm 24 and I know of it
🤷♂️
is legend
What's the best current 13b and under model for prompting anyways?
That's rude bro I know hella games 
Specifically tells who wears what, adds "mixed bodysuit colors" as negative, well, it does absolutely opposite lol
Oh, holy tuba Shub-Niggurath! Alien Fanky is wearing a t-shirt that says, 'tickED BEANS ALL SINK to A B free'! This is ridiculous; you've notitizens in costume? You look at the clownfist costume you're wearing! You look at the (chef's pizza)tingdog costume you're using! You're turning around and aremans are crunching!
An ordinary day at the office
Oh, holy tuba, pop out the links, popeye's face, profile, angle, dreamlike, depth of field, best quality, masterpiece jacuzzingoblue, dreamlike, spooky crobypusk, wrinkly, covered in mr. pibb
stole your prompt
Flux vs SDXL:
Oh, holy tuba, the jumping spider is wearing a lab coat sitting at a booth in a metal room, horror movie still, scene, entire frame, from below, overhead, film grain, (sloppy:1.2) apefighter, flicks, boomer style iddling Jack, (cropped:0.8), (blurry:0.8)
prompt: 'a vibrant, sun-dappled scene of viscous, glistening fluids intertwining in a sensual, playful dance
Oh, holy tuba, professional photograph, wide angle, intricate, (leather:1.2) personalodome, intricate, elegant, high quality, renoir,avaitor, pumping, dreamlike
stole your prompt
Help me design a door head, the font is: Chen's professional massage
hi
are you trying to prompt that or asking the community to help you with a prompt?
a airplane in the water
Here is the image you requested.
this might be my next desktop
Here is the image you requested.
Flux.Dev and Blue Future LoRA
a greyhound running on the sky with donuts
/help command
Hey @subtle depot Here is basically how you do it with comfy
Random workflow injecting noise :
- "На холсте изображена девушка с мечтательным взглядом, стоящая у окна с видом на осенний парк. В ее руках книга, а за окном опавшие листья и дождливое небо."
Flux.Dev and rbcharc0al and Blue Future LoRAs
why need OH in promt?😃
It’s not “needed”, just what I used. These are prompts from Invoke. There is a custom node for Invoke called the nightmare prompt node that uses a weird, broken custom llm to generate random prompts. You can optionally specify some text that will appear at the beginning of the output and may or may not affect the text that is generated after. I gave the node “oh, holy tuba” for these.
oh ok
Like using the phrase at the beginning of each prompt "custard hegemony", or even "once a fish forgotten" etc. et sim
I didn't use Oh holy tuba here 🙂
Nor here
too scary
Flux.Dev and rbcharc0al and Blue Future LoRAs
Good morning!
Tonight I'll be joined by my friends @MachineDelusions and @BinxNet as we discuss Flux LoRA Training.
- How to assemble a data set
- Training Settings
- Where to do it
- How much does it cost?
- Best practices
- Cool Tricks
Black Forest Labs - Flux
https://blackforestlabs.ai/announcements/
Machine Delusions / @MachineDelusions
X - https://x...
Flux.Dev and rbcharc0al and Blue Future LoRAs
this combo is great
So cute 🤗
"nah bro this weed isn't very strong"
Flux.Dev + LoRA-fest 1) rbcharc0al and Blue Future; 2) rbcharcoal and Art Nouveau; 3) rbcharc0al and Pop Art
best style of motorbike
Good morning coffee!
Highway to hell ...
Waiting for Dorothy ...
is it Tin Man or is it C3PO 🤔
Prompted C3PO 😄
sometimes if you prompt R2D2, C3PO will show up also
Yes ... they belong together 🙂
good day guys
sup
Oh, holy tuba, intubateing, warm milkshake warehouse, squishy, sewage-covered, covered in strong ketchup, super nintendo, ektar 100 pushed to 400, zenza bronica professional photograph, archs and wonderzoans Chudmancer: what? i thought sewer pig was a baby bear... youre telling me hes some kind of famous ape? colon-collapsing, dumpster, celebrity...
Oh, holy tuba, the sloppy one next to it, nonerize,eanooky, covered in bang energy, miller high life, qinqin baobao jugaogao , giaggost , detailed masterpiece photograph , for some reason, they're not making any sense
Oh, holy tuba,detailed high res, intricate, comfy, covered in gween tea, freegans, wuice, professional photography,!!!!!!!!!!, puddi-dome,gae,darling,slime, potato time,potato~"
OH 😃
Why does flux do this? The images though different prompts are very similar in the composition / background, same angle, same plate, same background
yeah flux has really low image diversity
not a big fan of flux
You are the dancing queen. Young and sweet only seventeen...
some serious dance moves coming from C3PO there
tested this flux.1-dev ascii art lora
Oh, holy tuba! wet smellio between megalovania andchromy with detail on wizzman's brand mustard, goth hospital, sad, at taco bell endorsements, HD, cinestill, variety ofapes graverobber, professional photography, enigmatic, high resolution
I have been told to thank you for the Flux version. My friends love it!
Flux.Dev and LoRAs Blue Future, Metallic and Novel Illustration
thats great news. And your welcome mate
@wispy nest https://arxiv.org/pdf/2403.03206
The flux guys wrote that paper and they built flux on a very similar architecture
promt?
So cute 🥺
thx
Make sure you´re not using the same seed, I noticed that forge only increased the seed number 1 by 1 instead of randomizing it , maybe its happening the same thing here
I'd need to look it up but my A.I. is rendering ...
That lighting damn, nice gen
Deforum?
Morph 2 vid
Neverending game ...
just some test videos i run a while back
I don't have an idea how to do that 🙂
true, that was the case. The seed was the same. When using from UI it changes after each generation, but i had saved the workflow and loaded each time to make calls via api (so seed in the json workflow was the same each time).
Bonjour comment générer des images ici ?
this went a bit wrong but was so hard getting ai to make a guy break dance lol
heybro
Hey
Ok merci je vais lire tout ça 🙂
that will do for now gets boring after a while lol
imake some totally random weird stuff when the jack daniels kicks in lol
this is what happens when you mix me with Jack Daniels and my checkpoints lol
Pretty cool!
I bet I would need a new PC for that 🙂
And some more knowledge ...
its easier than youd think bro , but yeah you need a big vram to tinker around with all that crazyness
I know there are technologies to do the one or other thing but in this mix I think it's a mix of different technologies ...
Couldn't find a better notebook at the GamesCon 🙂
Next one will be a Desktop when nvidia comes out with the 50's
Youll still be better off of running 2 duel RTX 3090s putting out 48GB Vram than remorgaging your house for 1 RTX 50 series lol
I have a coupon for ASUS 🙂
He only has to manipulate bitcoin ... ... ...
Uhhh... they really try to push Alife Virtual ...
hremlins
Bro your stuff is unbelievable!
yeah very consistent
bizarre unbelievable action photograph of a scifiender cyborg post apocalyptic entity screaming in front of a chaoszone, kewpie mayos photo, zigzong lighting, scientist, scientific instrumentation, millipede, human skeletons, etc. detailed intricate (cyberpunk:1.2), (metallic:1.1), strange hottened slay, deep dish, brackin', (cold dark:1.2)
bizarre unbelievable action photograph of a sea urchin made out of paella and meatballs,legraph technical cross section, artstation DJ is CBT, moody lighting photograph, blurry photoreal Michael B. Jordan logo, alien bagpipes JUBA decane baton quality, chaotic dra tablespoons scattering
help what happen to forge? i just updated from the update file and when i tried 2 start, i got this error
there's been a lot of discussion about forge in #💬|general-chat today
can you help crystal?
i don't use forge, i use comfy. You might post in #🤝|tech-support and see if @dry crow can assist
Good morning
thanks bud. Means a lot
bizarre unbelievable action photograph of a final beanmancking! elapinetful slam god finder plobbin', cgi render, high detail, fashion, clothing, detail on shirt
bizarre unbelievable action photograph of a large magical glowing sahelond sausage flying through a magical storage room,ocalypticeprojections, hazy fenex control device, ominousiectess subdued lighting, cinematic
A moring frost on a concept hyper car parked in a busy city street .... NOOICE
Very welcome! I'd love to learn some of your techniques and how you make some of that stuff some day
Yes im alaways up for showing how to's mate
Hi mate yes ill make some telsa now for you
does he have a spacific model Tesla
here is my prompt i use if you want to test your own to spectacular digital rendering of a (transparent:1.8) Tesla car, revealing internal mechanical components such as engine, car chassis, suspension, and internal wiring, detailed textures, detailed machinery, accurate lighting and shadows, 8k quality, intricate patterns, high-definition, glossy finish, vivid reflections, perfect lighting, showroom BREAK,
that was rendered on my Flux hyper checkpoint
spectacular digital rendering of a (transparent:1.8) Tesla car with the texr above the car, the text says "TESLA"., revealing internal mechanical components such as engine, car chassis, suspension, and internal wiring, detailed textures, detailed machinery, accurate lighting and shadows, 8k quality, intricate patterns, high-definition, glossy finish, vivid reflections, perfect lighting, showroom BREAK,
if he likes tesla i have a tesla on fire lora
Our hub provides members with exclusive access to an elite selection of AI image generation models, designed to produce superior quality images that stand out in any creative project
They look sweet mate. What checkpoint did you use?
Yes - when the wheel arch and the tyres are also see-thru it looks wrong alright!
FYI I use dev-fp16 and fp8_e5m2
Flux.DreamDiffusion
I use flux not SD3
I rarely use SD3 - (except on SwarmUI)
I even hyper trained sd3 and i still dont even use it now days lol
lol
i stopped too
flux is amazing at tiny details
Good evening 🙂
lol it gave it a motor
LOL thats the prototype they used a ford as the base mode 😆
Ahoi Dice 🙂
Howdy Mr MOJO
Everything OK?
No wonder it has to get fixed or repaired daily hhahahahhahahahha classic
Fkd on race day
boom
all good bud. Hows tricks?
They took about 0,5 L of blood from me. But everything fine 😄
pinto edition
Had to let ChatGPT translate "Hows tricks?" 😄
was your doctor count dracula
Looks like ... DNA analysis ...
In TV they only need a drop ^^
lol tv medical stuff makes it all look so expedient and efficient and specific.
CSI is a good one
Yes ... I have been at every place where is a CSI series 🙂
Miami, Las Vegas, New York ... it makes watching it even better 😄
What about Washington DC
i've never really cared about set locations tbh. I enjoyed all the CSI even without going to the cities.
so many shows are filmed in vancouver and it has never enhanced it for me
wll a little. i get tickeld whenever stargate sg1 goes to a new alien world and it's just a spot down a logging road in BC
or that time muldor chased a guy into a BC transit bus graveyard
I haven't been in Washington and don't know that part of CSI ... I'll have to search. The fun is more to see a place where you have been ... I liked StarGate Atlantis more ... but that doesn't make SG1 worse ... and the movie ... you can still watch it today and it's great!
CSI Cybercrimes i think it was called
it bombed
SG:A is amazing in it's own right. It's the same thing though and was filmed in BC. Every alien world is just some landscape outside of Van.
It's all been good ... the rest is just a matter of taste 🙂
The original movie holds up so hard. I've always wanted to rotoscop macguyver's young face onto Kurt's face in the movie
Hell, i loved universe , but it had issues and they wanted to make it all sexy and soap opera like battlestar reboot
Same thing there too, every alien world was just a gravel pit in BC
Something for the Dice 😄
You thought of that cause the name Daniel Jackson went through your brain a moment ago huh
I remember him saying it sometimes let him do crazy stuff 😄
Mojo
dicordos!
I eat fruit marshmallows, do they eat this in Germany?
I don't do that often ...
i too
If you need it ... do it!
thx!
I think it's a bad idea to say: "I'm not allowed to do it" I think it's the better idea to say: "Let me think whether I really want it now."
you wanna it now?
Most of the times I think I don't really need it now ...
Took me two years to try a new beer mix ... it's been interesting but not really the best time for it ...
you need to try new bear mix
The point is not to consume cause it's available ... but if you really really want it ...
To much is available ...
sleep dont work and play games😃
If you need that you should be able to do so ...
food patrol👮♂️
Well the point is .... maybe you have better 60 years ... than not so happy 90 years ....
👏
@dry blaze here bro
Damnit, can't use my fancy self hosted image sharing 
@wispy nest
Any idea what could chug away the memory like that?
try this workflow bro
download that to your workflows file
its a slightly simpler flow that runs fast
Got a link to the model you used? As it's not the "stock" one :P
you can choose the stock one from that drop down in the node
but if you want my flux checkpoint its here https://tensor.art/models/759856135286068673/FLUX-HYPER-TRAINED-DREAM-DIFFUSION-BY-DICE-V-1
save the flux checkpoints to the unet folder in comfyui
Aye, but then i get the memory ramp up again
Hence i wanted to try the model you used to see if it'd help
Nope. too slow
should be faster than your curent set up
Or i use it rarely, but mostly just to prune models and fetch images.jsons from civit
Aye, i might try it for flux. Cause comfy is miles faster than forge with regular SDXL/1.5 models.
forge 35 sec to render
Ain't so much time taken, more memory used for flux.
its your memory thats causing the time taken
i find forge to use less and you can can the set the vram in forge
I haven't named any time due to knowing how much slower model mush into swap is vs just vram :P
Just need to move the models over to my other nvme once the damn dram cache doesn't shit itself lol
what file name do you have the T5 and clip_L saved to?
they should be in the model/clip or copy paste them also into model/clip_vision folders
Oh, name as in directory/path
Aye. Also, why 2 copies of them?
some comfy ui like mine as i run comfy as backend in swarm ui work for me in clip_vision and comfy ui main work in clip folder . so i put in both
Ah, so i just need one of them :P
With swarm, iirc it's a collective performance boost for everyone, by everyone. Is there a limit to vram usage on that one? As iirc it has quite the queue
swarm is local
its very fast for me
my swarm comfy in real time generation
I've no idea why comfy decides to peak out vram for me.
And forge seems to only spit out black images
Throw me one of your gens in forge, and i'll check what i might be doing wrong there
forge settings ....
top bar use T5 not vae
CFG 1 Distilled CFG 4
shedular type simple
sampling steps 20
low bits automatic
my forge settings
Ah, there we go. Didn't know the T5 WAS the VAE lol. What's the "ae" file for then? 
Wasn't about speed, but memory usage as i said
As comfy seems to use all 24GB plus swap
As comfy didn't have this
happy days
Aye.
also, could you fire up task manager and check gpu usage while comfy runs flux for you?
try this prompt Title: Display the title "SpiderMan" in large playful text at the top or center of the poster.
a photo of a SpiderMan , surfing a big wave
yes i can i have 3 monitors
You can do that on just one too 
Now that i've gotten it to work, i'm curious to try the larger flux models :P
Though, i don't know if the full fp32 will make much difference though
Aye. fp8 can already eat all vram anyways lol
Aye. Super curious how this one will turn out. Might crank up steps though
Thankfully as nvidia has nvenc, i can just watch plex in the meantime 
Tiny bit dissatisfied with the lack of detail/crispness 
150 steps next
gees the steps are great between 20 and 30
cfg upto 4
you can always use hires fix , real x4 hires steps 10, denoise 0.6, upscale by 2,
Kohya HRFix Integrated can somtimes cause the image to not render
or use DynamicThresholding (CFG-Fix) Integrated
moving convo here since i need to explain better. This "Show NSFW" button does absolutely nothing.
nope
where is that settings page? you have a different node setup i think
bizarre unbelievable action photograph of a bizarre creature peeking out of a colon-collapsing pill, spaceship meal hall detailed sharp focus intricate best resolution (tangy chunkybutt:1.2), chaotic intricate, realistic lighting, intricate screaming hobbit female scientist cyborg, pronounced feminine features, pronounced emotion, chromatic aberration, bizarre costume, ominous low angle shot, dim scene lighting, spooky glowing meat chunks
what interface is this?
okay, you should probably open an issue on there about this then
bizarre unbelievable action photograph of a scifiorm fishfangler playing Dino Crisis Xtreme Rivertides with husband cody, a local bartender, hot peanut butter and qinqin baobao jugaogao!!! quality masterpiece intricate detailed marinara quality!!! huge salamiarro delicious cheese sauce masterpiece
bizarre unbelievable action photograph of a chainmur gone-for-a-walk on an elementary street in schween'srekland inglyrics marinara, wind chimes and papaya marigolds well-lit intricate to best quality render ownsales photograph of a high tech combinationilla device
same prompt
bizarre unbelievable action photograph of a weapon�, disassembled components running amoketimely background, aged cinematic lighting, sharp focus, ominous cyth drusk bananas party shot high popeye's Fruit Juice Release, sloppy like 4 the Incur 1953 0, skilled resolution,abyss shot, macro shot, perfect focus, variety of flavors, snacks, greebles, garmonbozia
done, but also looking for a better node if this one is broken
the 4 variation outputs are with and without LORA, and with and without Baked/External VAE
looking for help in optimizing this workflow:
@wispy nest Eyy! Found someone' s custom node to offload clip text encoders to cpu, and now i can even load in fp16 text encoder without issues!
fp8 vs fp16 text encoders
awesome its going well then?
Now it is, as the old workflow didn't properly offload the text encoders as everything was packing into the vram :P
Sadly, as apparently this one can speed up flux by nearly 2x, it just errors out on my end, so need to do further googling tomorrow lol https://www.youtube.com/watch?v=AzeZkosyqp4
A guide through installing the most recent quantized models and the GGUF loader to speed up your FLUX generations using comfyui—even on low-end GPUs to maximize the performance of your VRAM and save you time.
GGUF workflow: https://goshnii.gumroad.com/
GGUF - Github Repo
https://github.com/city96/ComfyUI-GGUF
FLUX.1-dev-gguf models:
https://h...
holy shit, is that sdxl, sd3? mind sharing the json?
thats my Flux hyper workflow for upscaling
ah, i havent got flux installed in my env yet
do you want my flux hyper?
Oh lol! Tested 2048x2048 flux with just 13 steps, and he went completely derp xD
i'll take anything to help learn
read the model discription
FLUX DREAM DIFFUSION BY DICEModel can be found on Tensor Art https://tensor.art/models/759856135286068673/FLUX-DREAM-DIFFUSION-BY-DICE-V-1or all my models are also over on Shakker.aihttps://www.shakker.ai/userpage/8b0d2aadaa2a4f2592cbb367c329ea51/publishStart of with these settings in comfy to get a feel for how it runs ....Simple Prompt : a jet...
thanks, model downloading
enjoy bro
Flux.Dev+LoRAs
nice renders up there🔥
Good morning coffee!
is this the charcoal lora
rbcharc0al, blue future (and the elephant adds wraith_bw)
Wraith_BW makes great moody pics
ah yeah the blue one
Blue Future and Wraith BW are LoRAs by @vagrant birch ⭐
😄
bizarre unbelievable action photograph of a fight-at the ball Game 5 with the rest of your life as curtainsooks : D menacing fantasy humaninteractivesoundball eralblewichcasting vest???s, biscotti werevs, (wide angle:1.1), chaotic views apediric disc repiteral engineer birdpoop asymmetrical plyombie lighting asymmetricalizonsur highchass
bizarre unbelievable action photograph of a (collapsing dongrel) mindfreakeraunting primordial jungle faygo, zos gibs hoard of the same barnaba in background _______________ ________________________________________________________________ roczy etards, ills discretion and usage ________________________________ icaefeyondbandface armor _________________!!!!
Imagine
Can anyone show me a simple video with videofx? I finally joined the waitlist so I wanna see if it’s actually good
Really cool, could almost be a promo shot
And now we wait. Moving most of my SDXL's to ny server to free up space as i just figured out how to properly use flux, which is my new luuv lol
I got dream diff flux though, but i wanna try his vae too :P
Don't know if it's "his", but he used one different from ae.safetensors :P
Also, by the way, is there a nifty way to make comfyui completely contained within it's venv? As for instance now, i'm on my emergency ssd to backup my windows drive as bootloader went to shit, and even this ssd is near deathbed lol. So as i don't wanna fill this decade old 120GB with A.I jargin and python stuff, i wanna make a comfyui "portable edition"
I think he only renamed it ....
@wispy nest Was it just renamed? Or did you have a fancy self tuned VAE as well? :P
In his article about his dream flux he refers to the original vae
Gotcha
🖐️
(I miss |Dice| - he might be a bit Walter Mitty - but he was always good to help!)
|Dicey|
I think he's cool ... but don't know him that good ...
Hoi, any idea why just LLM's being large ass models only transfering at 15MB's with 100% active time on a gen 4 nvme? 
Not enough RAM?
@wispy nestI have to ask, what do you use for your videos? Do you have a variety of different workflows or one specific one?
cpu bottleneck? ¯_(ツ)_/¯
Cpu was barely touched. And that was with robocopy set to use all 12 cores
Here however, a faster cpu will literally help loading flux faster as i set comfy to load text encoders in cpu to offload gpu there 
woot, finally got my SD3 workflow spitting out images how i like it:
Are there standalone one click tools like automatic's "civitai helper" that i can simply just add paths to each model folder, click it once, and it goes through all my new models to add missing metadata, links and pics? As it's annoying to have to quit comfy just to open up automatic to update models, then quit it to fire up comfy again
@wispy nestHoi, do you know of a python program that can fetch images and metadatas for models on tensor.art? As i use civitai helper in automatic for fetching for civitai, but not one for tensor.art
Im losing hard an argument and im about to cry
this artist is generating Ai art and then tracing over it right?
https://www.phixiv.net/en/artworks/121612863
im not a sore loser, but it just doesnt make any sense, his artstyle its all over the place
probably not
im fucking losing it
of course its not 100% ai generated
he started drawing may 2023
but you can see he traces, and he actually learns to draw by tracing and copying the AI's artstyle
so my theory is that he actually is learning and tracing at the same time, but you can see that AI was involved somewhere in the whole process
but that doesn't mean that what he's posting is what he traced over. he could just be looking at it and using it as refrence - though i don't see anything about it that makes me think an AI generated even the image he's refrencing
Tracers have been around for eons in the digital art world, using AI to make a base image is just another reference tool for it. I don't know or really care if that guy is actually tracing or not, but as long as people keep liking his stuff, he'll probably keep doing whatever he's doing.
as is his right
Yeah, to each, their own and all that jazz
90% of what I do with diffusion is making concept art references that I then use for ideas for making 3D assets and environments. I understand that some might frown upon that because that bypasses needing a concept artist
Just means you are functioning as the concept artist yourself
GGUF
Flux. Careful, he's got the right to bear arms.
bizarre unbelievable action photograph of a verypping weapon hotted to endanger a delicateizon elegant bedroomorthyrah Teenaged viewer,re into photograph (poster) high res, extreme angle, intricate, from below (catbox et a rellabos), swirling colors, stripeQUESTmann bear wearing light hair, beard, clothing
bizarre unbelievable action photograph of a (cute bacon pizza with small details:1.1), detailed porcelain pot, (detailed diagrams macro and intricate component:1.2), ice cream, meatballs and beancrunch (demand the evil BEAN'S MEatsayette with tuna-free peener)!!!1 usagebanana barrier!!!
bizarre unbelievable action photograph of an evil grinning budgerigar holding two potatoes, intricate,indigo quality, rationing press footage of the ortolan crobypusk kicking them together graverobbing Rank Randy! taylor brand ham and the butthorse logo is whaturcup display A, with a short cyberticle heard it? urinal inU, urinum urgompetula view
I want the metadata like that provided by A1111 - but in ComfyUI? You have a w/flow for me? 🙂
do you mean you want all the png metadata like A1111 had by default?
I would like a wolf in a dark forest
All my comfy generations already has workflow baked in. Was more looking for pretty much a "standalone" civitai helper for the A.I site you use :P And for civitai, than having to boot up the heavy ass automatic1111 :P
from diffusers import StableDiffusionPipeline
import torch
加载Stable Diffusion模型
pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
pipe.to("cuda") # 如果您有GPU,使用cuda以加快生成速度
设置提示文本 (Prompt)
prompt = (
"A detailed and futuristic cover image representing the Integral-Omics workflow. "
"Show a scientific experiment setup involving the sequential extraction of omics layers "
"(metabolome, lipidome, genome, transcriptome, proteome, phosphoproteome) "
"from biopsy tissue. Use abstract, modern, and high-tech elements to represent each omics layer. "
"The design should be clean, minimalistic, and emphasize the flow of data and molecular interactions."
)
生成图片
image = pipe(prompt).images[0]
保存生成的图片
image.save("integral_omics_workflow_cover.png")
显示图片(可选,如果是在本地运行)
image.show()
@humble geode when pony video?
If v7 works out, then I'll start experimenting after it, I've been collecting datasets for a while, unfortunately it's even more expensive...
ye video models would be rough for pure open source to fund
Commercial photography, powerful yellow powder explosion, fried chicken, black background, bright environment, white lighting, studio lighting, OC rendering, super detail, solid color isolation platform, professional photography, color gradinging About Midjourney Parameters --ar 9:16 --v 5.2 --s 750 --c 0 --q 1
thought i don't see any of the ai image galleries here, i wonder if this project is interested in updating their tool for a few sites.
https://github.com/mikf/gallery-dl/blob/master/docs/supportedsites.md
what is this list
?
bro
what is the list?
i don't know where to get this sht from
lol
i dont think this is a horny server. you might want to ask where all the horny people would be. i'd look into animatediff with sparse ctrl. for automatic1111 you may need an older version of the contorlnet extension and it won't work for forge. I have no idea what nodes or workflows would support it on comfyui. sparse ctrl though its good. i've only ever seen people use it and ive not hooked it up yet.
alright this is what i would like to transform
from imgtovideo
wow nice
what do i do here?
any intros to thiss? would like to make a running video of naruto with that image, how do i do that ?
getting this error
where do i find that?
How do i download this lol
can't find download linki
i am sorry i have so many questions lol
i am trying my best
to understand this model
what do you think?
if its transparent how can we see it 🤔
someone should make a lora from Burnout game covers
Searching her sleeping room ... ^^
this is kolors
would need a second pass with realvisXL or something
kolors needs a refiner usually
Have you tried Cascade Ultrapixel?
That would be something for Dice, too ... to render a picture like 9000x6000 Pix 🙂
not yet, I would like to
I actually got a server earlier today to try it but I ended up making some nice SD 1.5 images which I forgot to save
Awww... that's bad luck!
haha yeah
I wonder why so many workflows don't save the results ...
its ok though since SD 1.5 is so fast
u can just run through like 50 seeds
oh I didn't even put a save image node thats why
Preview only?
yeah just preview node
but I use that gallery thing at the bottom as well
I don't save images much cos I don't actually have a use case for this technology I'm just messing around
don't rly need images LOL
Spybot 🙂
He sees you.
would you let me know please?
There should be an upscale_models folder
where to find a controlnet model?
how to download it? really github is so confusing for me i swea
swear
I can't tell you, I use DrawThings.
For upscaling I use Chainner.
this is controlnet model
i am done upscaling
where do i find a reliable link
please i need it
it seems to can't go without
jeesus, this thing is so tought to use
I don't use comfyui, you could ask in https://discord.com/channels/1002292111942635562/1204675216773619752
I must admit it was more of a joke, when I saw that video, I thought about you... and I couldn't help myself...
Guys,
Which Gguf is supposed to be as good as nf4_v2?
what model are you trying to use controlnet with? we can't give you a link until we know if you're using sd1.5, sdxl, flux, or something else
you really need to be on the L2 discord, especially for questions about ipadapter and controlnet. I don't know if there is a good, working, controlnet for flux yet. this https://github.com/lllyasviel/ControlNet is illyasviel's github page
and you should consider going to youtube and watching purz's videos https://www.youtube.com/@PurzBeats
l2 discords?
i am using animatediff with sd 1.5 it aint working
let me show u the workflow
yes. it's matteo's - he writes nodes for comfy, you'll find better technical help there than here. his channel is here https://www.youtube.com/@latentvision and he always lists his discord in the description
go watch Purz's videos on animatediff, i don't use it but he's very good with it
wait. why are you calling qr code monster for control net. that's what you use to make illusions with . you can see how it works here https://huggingface.co/spaces/AP123/IllusionDiffusion with angry penguins demo space
i think it might be good if you started learning comfy with simple workflows and worked up too complicated ones like that one
comfy isn't a point and click interface, it's a programming interface. you do need to learn the technical stuff not just run workflows
😄 Worked here ....
i mean technically comfy is a point and click interface, but you mean "simple" like a point and click camera would be. /pedant
that's a huge workflow, but without the .json file, none of us can look at it
I don't really do many morphing Vids ...
can you upload that
/carly simon - you're so vain
and i gave you the link to where you can download the controlnet package
i'll see if i can get this to run, but go to the github link i gave you and you can download the controlnet package there. the install instructions should be on the main page
It worked here ....
it loaded. this looks like one of purz' workflows
I think I've had it before ....
it's the animated controlnet node... that's one thing he does a lot
For sure! Always trying to learn 😄
computer don't care if it's readable...
Yeah ... I think I need to test a bit to choose the right pictures ....
and it blew up immediately. this is too big a monster for me to want to trouble shoot it right now
The hardest part for me is where to put the files 😄
it's a massive workflow and not something that a new user should even consider running. there's a lot of very complicated nodes in this. why are you trying to run this?
It will not work the way you think ....
um - while animate diff does do animations, i don't think you understand just how bad it is for that sort of stuff
It's morphing 4 pictures with a little animation
the only good video animators that exist are kling, luma, runway turbo - and stuff in development
so nothing open source then?
please start with a simple workflow to just make one image and learn what all the nodes do as you slowly use more and more of them
not yet, now.
maybe soon, black forest has a video function they're working on
i still would like to try and use this
SVD could work but only 4-8 seconds ....
i might make slow animation
turning into quick ones
somehow
but this is why i would really like to learn it
as much as i can
what i would advise you doing in that case, is look at each node, one at a time, search google for that node, and read what it is and does if you find it. if not, ask about it on the L2 discord - matteo at the least should know
i am already so far with animatediff
L2 discord?
yes. i gave you the link to the Latent Vision youtube channel. that's matteo's channel. he always links his discord - the L2 discord - in the description of his videos
he writes nodes for comfy.
it's really hard to learn to swim by jumping into the deep end instead of starting out slowly
https://www.youtube.com/watch?v=AbB33AxrcZo&list=PLIF38owJLhR1EGDY4kOnsEnMyolZgza1x Is my fav to learn Comfy .... start is a bit outdated but what you learn you can use later, too
Try a black female Jesus surfing 😄
Processing A1111 photos (via metadata) in ComfyUI+LoRAs
BTW any news about the final SD3?
SD announced a better SD3 Model ... after the first one was released ...
Same here ... I wonder how they wanna keep up ...
Always good to have competitors ... but Crazy what Black Forrest did ...
what did black forest do?
Release Flux