[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: tmp.jpg (1.26 MB, 3264x3264)
1.26 MB
1.26 MB JPG
Discussion of free and open source text-to-image models

Previous /ldg/ bread : >>101908455

>Beginner UI
EasyDiffusion: https://easydiffusion.github.io
Fooocus: https://github.com/lllyasviel/fooocus
Metastable: https://metastable.studio

>Advanced UI
Automatic1111: https://github.com/automatic1111/stable-diffusion-webui
ComfyUI: https://github.com/comfyanonymous/ComfyUI
Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge
InvokeAI: https://github.com/invoke-ai/InvokeAI
SD.Next: https://github.com/vladmandic/automatic
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI

>Use a VAE if your images look washed out
https://rentry.org/sdvae

>Model Ranking
https://imgsys.org/rankings

>Models, LoRAs & training
https://civitai.com
https://huggingface.co
https://aitracker.art
https://github.com/Nerogar/OneTrainer
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts

>Flux
https://huggingface.co/spaces/black-forest-labs/FLUX.1-schnell
https://comfyanonymous.github.io/ComfyUI_examples/flux

>Pixart Sigma & Hunyuan DIT
https://huggingface.co/spaces/PixArt-alpha/PixArt-Sigma
https://huggingface.co/spaces/Tencent-Hunyuan/HunyuanDiT
https://huggingface.co/comfyanonymous/hunyuan_dit_comfyui
Nodes: https://github.com/city96/ComfyUI_ExtraModels

>Index of guides and other tools
https://rentry.org/sdg-link
https://rentry.org/rentrysd

>GPU performance
https://vladmandic.github.io/sd-extension-system-info/pages/benchmark.html
https://docs.getgrist.com/3mjouqRSdkBY/sdperformance

>Try online without registration
txt2img: https://www.mage.space
img2img: https://huggingface.co/spaces/huggingface/diffuse-the-rest
sd3: https://huggingface.co/spaces/stabilityai/stable-diffusion-3-medium

>Maintain thread quality
https://rentry.org/debo

>Related boards
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/b/degen
>>>/vt/vtai
>>>/aco/sdg
>>>/trash/sdg
>>
>>101911047

https://huggingface.co/spaces/fancyfeast/joy-caption-pre-alpha/tree/main

>git clone, create a new venv for it, activate venv, pip install the requirements then run the file and it will launch the gradio. For batching you will need to edit the file.
>>
I wish purple witch would post in the containment thread so the schizo would stop spamming
>>
File: ComfyUI_Flux_8475.jpg (205 KB, 768x1344)
205 KB
205 KB JPG
>>
>>101911094
Thanks anon, it seems to use non local models so this needs to be adapted if that's even possible :

CLIP_PATH = "google/siglip-so400m-patch14-384"
VLM_PROMPT = "A descriptive caption for this image:\n"
MODEL_PATH = "meta-llama/Meta-Llama-3.1-8B"
>>
Blessed thread of frenship
>>
File: 00097-3227912660.png (703 KB, 1152x648)
703 KB
703 KB PNG
Here's your Elle Fanning picture, saar bee bop be bop
>>
File: 1719723531301757.png (2.61 MB, 1024x1024)
2.61 MB
2.61 MB PNG
>>
File: file.png (7 KB, 976x23)
7 KB
7 KB PNG
Sooner
>>
File: 00104-1974336298.png (1022 KB, 896x1152)
1022 KB
1022 KB PNG
>>101911145
now with lora

i guess we can call this overCAKED
>>
>>101911093
Those times will never return. Make new memories, or transcend all attachment while you wait for death.
>>
>>101911196
>>
>using joy caption
>it randomly cuts off mid sentence at the end of the paragraph
uhhh
>>
File: 103158-tmp.png (2.79 MB, 1536x1728)
2.79 MB
2.79 MB PNG
>>
>>101911250
/plap/
>>
>>101911250
its the underwear print that makes it for me
>>
>>101911234
>catie
https://www.youtube.com/watch?v=out-etvwT8A
Boxxy is an eternal archetype. The human who channeled her is irrelevant.
>>
File: ComfyUI_00509_.png (1.33 MB, 1024x1024)
1.33 MB
1.33 MB PNG
trying out using Anynode to have a basic prompt go through an LLM (gemini atm) to enhance/embelish the prompt so I can be lazy. Pretty cool desu, out of free tokenz for OpenAI tho. Gemini is hitting me with content filters a lot
>>
>>101911250
go back to /sdg/
>>
File: 63678064.png (1.38 MB, 1344x768)
1.38 MB
1.38 MB PNG
>>
>>101911260
don't use that node it creates mustard gas
>>
>>101911256
>>
File: 1710595210204345.jpg (17 KB, 350x508)
17 KB
17 KB JPG
>>
>>101911249
it has a 300 token max output
>>
>>101911289
I like how you can begin with Taylor Swift at a political rally and end up generating /x/ boxxies four hours later.
>>
>>101911310
RIP
>>
>>101911310
it should be plenty for most images, stop giving it collages
but I'm running it locally and can use whatever limit I want, it actually describes nudity and uses the correct pronouns so that's nice
>>
noob here, i installed automatic1111 webui last year and it worked okay i have a shitty amd 8gb gpu but it did work okay somehow) but months ago i changed cpu (a cheap ryzen 5 , before i had fx 6300 lol) and mobo and now webui sometimes restarts the pc when generating pics.
do u think it would help reinstalling the whole thing or the problem is somewhere else?
>>
>>101911325
Still not as good results as comfyui
>>
File: ComfyUI_02392_.png (1.95 MB, 1024x1024)
1.95 MB
1.95 MB PNG
>>
File: ComfyUI_00511_1.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>101911286
this is a different node from the compromised llm one
https://github.com/lks-ai/anynode
>>
>>101911347
This happens to me when I do VR, too. I think I fried my GPU generating porn.
>>
File: 3418281226.png (1.66 MB, 896x1152)
1.66 MB
1.66 MB PNG
>>
File: file.png (637 KB, 823x1018)
637 KB
637 KB PNG
kek
>>
>>101911183
>>101911145
>>
File: ComfyUI_02002_.png (1.43 MB, 896x1152)
1.43 MB
1.43 MB PNG
>>101911356
i know im just being a goofy lil guy :3
>>
>>101911347
maybe check if things go out of memory with the task manager
>>
File: ComfyUI_00508_.png (1.01 MB, 1024x1024)
1.01 MB
1.01 MB PNG
>>101911384
you silly goose
>>
>>101911356
Prompt?
>>
>>101911345
>uses the correct pronouns so that's nice
huh, does the local model work better at detecting whether the subject is male/female?
unironically just had it spit out
'the figure is androgynous and it shows their bare chest'
as if you couldn't discern sex from that bit
>>
>>101904568
so what, this model works both as a direct clip-l replacement and as long-clip?
>>
File: 00029-2388015797.png (1.17 MB, 832x1216)
1.17 MB
1.17 MB PNG
>>
>somebody has been generating busty nuns with horns for literal hours
>>
>>101911411
I used some Pony images and it did okay, changed the prompt for it to describe only sexual content and there were no refusals
>>
>>101911397
yeah it goes oftenly out of memory but it used to just crash or not generate anything, now it suddenly restarts everything. gpu temperature stays lower than when i play vidya so i don't think it's that
>>
how the hell do I make flux generate something without a head, it's giving the thing a head or a helmet like 50% of the time even with the descriptor
>he is headless, there is a small stump at the end of his neck instead of a head
I guess that applies to diffusion models and general prompting as well, I'm a bit new to this stuff
>>
File: ifx40.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>
File: 103163-tmp.png (2.95 MB, 1536x1824)
2.95 MB
2.95 MB PNG
>>101911262
I can post in both threads
>>
>>101911445
Can you make it not look like a professional shot?
>>
File: 1177985049.png (978 KB, 768x1344)
978 KB
978 KB PNG
>>
>>101911442
>instead of a head
that doesn't work, it can't handle negatives at all
enormous LLMs like GPT only started handling them reliably enough with GPT-4 so our little text encoders stand no chance
>>
>>101911452
sexo! goblin sexo!
>>
>>101911442
if dullahan doesn't work you have no chance.
>>
So how much vram are the 50xx cards going to have?
>>
>>101911442
Deformities and missing limbs were impossible for me to gen.
You might be able to fake it with something like "head out of frame" or "neck down"
>>
File: 00031-4139821224.png (2.62 MB, 1280x1920)
2.62 MB
2.62 MB PNG
>>101911490
5090 rumored to have 28GB vram. Thankfully I don't have a reason to upgrade then.
>>
>>101911490
48gb minimum for the 5080/90
>>
File: ComfyUI_02400_.png (666 KB, 1216x832)
666 KB
666 KB PNG
>>101911514
>mfw develop cubital tunnel syndrome
>stop playing videogames completely because of the pain
>only joy in life comes from genning

it's over
>>
File: JOY.png (647 KB, 1024x1024)
647 KB
647 KB PNG
>>101910731
Okay, so here goes Joy's attempt!
>This is a digital illustration in a cartoon style featuring two anthropomorphic characters, a hedgehog and a blue hedgehog. The hedgehog on the left has a brown and cream-colored spiky back, large black eyes, and a small nose. It is holding a large, bright yellow sunflower with a dark brown center, and is smiling with a sense of joy and wonder. The hedgehog on the right is Sonic the Hedgehog, known for his blue fur, large green eyes, and white gloves. He is standing with a slightly surprised expression, looking at the sunflower. His shoes are red with white soles and he wears white gloves. The background is a simple, gradient yellow with small, glowing stars scattered throughout, creating a warm and magical atmosphere. The ground is a light brown with patches of green grass, adding to the natural setting. The overall mood of the image is friendly and whimsical, with a touch of magic and wonder. The textures are smooth and vibrant, typical of digital illustrations.
Oh well...
>>
anything new or notable in terms of controlnet?
in the timeframe before flux i mean. especially for pony. (i've been on a break for a while)
>>
>>101911531
i don't know how you're all getting carpal tunnel
maybe it's all those wrists rests and titty mice and badly aligned tables and chairs
>>
for anyone using the XLabs realism and the Topless lora beware they interact badly and make the image very fuzzy with some prompts
>>
>>101911478
I didn't even know that was a thing, neat
the model doesn't seem to understand it though
>>
Anyone got a Kohya config file for low vram loras? Got 16gb
>>
>>101911553
yeah, it seems to have all semblance of violence, nudity, and sexuality removed from it, so little wonder it doesn't work
you may be boned.
>>
>>101911548
I played MORDHAU for 700 hours and I think that was the game that ultimately fucked me up
>>
>>101911560
Finetunes soon(TM)
>>
File: 103166-tmp.png (3.06 MB, 1536x1824)
3.06 MB
3.06 MB PNG
>>
>>101911558
>16gb
>low vram
>>
>>101911586
For training flux loras it is
>>
File: file.jpg (30 KB, 1280x720)
30 KB
30 KB JPG
Can you guys recreate something close to the classic boxxy image without using img2img? I tried and it seems Flux has very little understanding of a completely black background in a photo.
>>
File: 3091693059.png (900 KB, 1152x896)
900 KB
900 KB PNG
>>
>>101911607
that's shoe0nhead
>>
File: ComfyUI_00512_.png (1.11 MB, 1024x1024)
1.11 MB
1.11 MB PNG
>>
>>101911615
i think i know what boxxy looks like, anon

don't compare her to that slut
>>
>>101911615
Hurry up and die
>>
File: file.png (3 KB, 130x38)
3 KB
3 KB PNG
>>101911524
There is no such nightly. But I will have my local build soon.
>>
File: ComfyUI_00997_.png (503 KB, 1024x1024)
503 KB
503 KB PNG
>>101911607
>>
>>101911665
What do you mean there is no such nightly? Just use it as index url or click on the link, it exist, just not public yet.
>>
File: file.png (3.79 MB, 2668x1505)
3.79 MB
3.79 MB PNG
Fooocus
>>
File: file.png (10 KB, 476x78)
10 KB
10 KB PNG
>>101911681
>>
>>101911442
Doesn't even work when you try genning the headless horseman. Pretty gay.
>>
File: 00043-2702452189.png (439 KB, 1024x1024)
439 KB
439 KB PNG
>>101911607
>Low res Youtube screencap of a 20 year old woman her head is tilted slightly to the side, she has heavy eyeliner and a mid 2000s emo haircut, only her face is illuminated the background is completely dark
>>
>>101911453
DIY
>myspace photograph grainy 1999 > an awkward normal young woman's low quality image on social media > so random XD jess omgg ur so EMO!! neon black t-shirt
>>
>>
File: file.png (3.6 MB, 2634x1337)
3.6 MB
3.6 MB PNG
>>
File: ComfyUI_00516_.png (1.39 MB, 1024x1024)
1.39 MB
1.39 MB PNG
>>
File: ComfyUI_02251_.png (1.85 MB, 1344x768)
1.85 MB
1.85 MB PNG
>>101911679
>>101911702

nice one anons, my chatgpt prompt kinda sucked at getting it done right i guess
>>
>>101911490

Considering they're planning a "Titan AI" card, I'd venture a guess that the 5090 model will be somewhat cucked to 28GB of memory, to drive people towards the Titan.
Then we'll have something like 48GB on the Titan that goes for $3000 and it's basically the only card that makes sense to buy if you want to upgrade or to buy a high end card for AI.
That's my guess how this will play out.
>>
>>101911706
Owlle Fanning
>>
>>101911722
Now share the prompt you used for that pic
>>
File: 2447584693.png (1.34 MB, 896x1152)
1.34 MB
1.34 MB PNG
>>
>>101911681
Try to post a direct link to a rocm6.2 wheel. I'll wait.
>>
File: ComfyUI_02158_.png (1.48 MB, 768x1344)
1.48 MB
1.48 MB PNG
>>101911722
>>101911731

lemme just share a catbox because it has some really silly non-standard settings from when I was messing around with nodes

https://files.catbox.moe/4vq1hj.png
>>
>>101911702
>>101911679
>>
>>101911781
Ty based challengefag
>>
>>101911795
It's not doing the dark background
>>
File: grid-0153.jpg (738 KB, 2688x1536)
738 KB
738 KB JPG
>>
Should this fucking shit be out of memory all the time with 24gb? what the fuck. it out of memories the graphics card or it out of memories the 32gb ram. fucking shit
>>
>>101911795
>>101911813
>1.5
You'll likely have to adapt the prompt to a booru tag style
>>
File: ComfyUI_01001_.png (498 KB, 1024x1024)
498 KB
498 KB PNG
>>101911607
Combining these two prompts
>>101911679 (Me)
>>101911702
>>
File: file.png (15 KB, 435x150)
15 KB
15 KB PNG
>>101911837
make sure you have the correct weight_dtype set
>>
>>101911858
with 24GB of VRAM he only needs to pin T5 to the CPU to avoid OOM
>>
>>101911858
Using forge right now. It out of memories whenever I load a LORA, even the nf4 one.
>>
>>101911837
>24GB vramlet
>>
>>101911884
this is not funny i did the right thing i did my research i bought the right graphics card only to get cucked
>>
>>101911702
Wife
>>
File: ComfyUI_00520_.png (2.32 MB, 1536x1152)
2.32 MB
2.32 MB PNG
>>
>>
>>
File: ComfyUI_02422_.png (1.05 MB, 1216x832)
1.05 MB
1.05 MB PNG
>>101911722
>mfw lowering the interpolate_phi on dynamic thresholding just makes miku progressively more retarded
>>
>>
>>101911945
>>101911958
Please share proompt
>>
>>
>>101911837
I also have 24GB VRAM and 32GB RAM and never crash
skill issue
>>
>>101911702
>20 year old woman
>looks 35
maybe next time don't specify white woman
>>
File: 103176-tmp.png (3 MB, 1536x1824)
3 MB
3 MB PNG
>>
File: 977243966.png (793 KB, 1344x768)
793 KB
793 KB PNG
>>
>>101912014
You're underage aren't you?
>>
>>101912014
this is what the average college girl looks like
>>
>>101911693
>>101911761
https://hub.docker.com/r/rocm/pytorch
>>
File: MM.jpg (244 KB, 1024x1024)
244 KB
244 KB JPG
>>
>>101911976
Basically the same one I've kept using since last time but with Nihei Tsutomu replaced with Ayami Kojima.
https://files.catbox.moe/gf1572.png
>>
>>101911988
reminds me of land of the lustrous
>>
File: MM2.jpg (200 KB, 1024x1024)
200 KB
200 KB JPG
>>
>>101912070
Thanks, but I need gfx1010, which is never included by default.
>>
>>101912097
Holy shit I know this girl.
>>
>>
>>101912097
gross
>>
File: MM3.jpg (223 KB, 1024x1024)
223 KB
223 KB JPG
>>101912122
dubs checked
>>
>>
>>101912080
NTA, but thanks.
are you enforcing a colour pallete by loading in an image with a high denoise like that?
>>
>>101912080
So it's img2img with very high denoise, you just use like a paper texture for the input or something?
>>
>>101912065
>the average college girl looks halfway to bogdanov
you can't fool me I went to college 10 years ago
>>
>>101912170
if you went to college 10 years ago then you should've seen girls like this everywhere lol
>>
>>101912185
bro I think you're just face-blind
this one's got wrinkles
>>
>>101912185
you must have some kind of autistic face blindness that causes some next level horror vision if thats what any girl looks like to you, what in the uncanny fuck
>>
>>101912185
>>101912197
I went to college 15 years ago and a couple of the pics look like my gf back then.
>>
>>101912197
>>101912200

to clarify: i live in bongland
>>
>>101912222
oh
carry on
>>
File: 1708048385782557.png (2.29 MB, 720x1280)
2.29 MB
2.29 MB PNG
>>
File: RanColor 1 720p.png (23 KB, 1280x720)
23 KB
23 KB PNG
>>101912168
I use either solid colors, or random large blocks of color. This just gives me interesting gens sometimes.
>>
>>101912231
"Artists" and morons don't understand. AI makes you fucking dream when you see stuff like this. Directly downloaded from the astral planes and parallel universes.
>>
Any lora that fixes the butt chins yet
>>
>>101912222
nevermind, it all makes sense now
>>
>>101912222
I'm so sorry
>>
>>101912161
Sometimes I do but it doesn't always work without changing the prompt, and you need to modify the prompt. For instance, this one used a neon green color as the base image I think. Without prompt modification, it's still mostly muted colors
>>
>>101912260
>parallel universes
I like to think some of the stuff I gen is existing in parallel universes, books, characters, whatever.
>>
File: fp112.jpg (213 KB, 1024x1024)
213 KB
213 KB JPG
>>
File: 1707491851186904.jpg (101 KB, 720x1280)
101 KB
101 KB JPG
>>
>>101912299
Take the schizo pill. Generative AI densifies thoughtforms by pulling them out of the quantum foam. Augment your low VRAM gens by the use of meditation and focused intent. Apply minimal thermal paste to your CPU for increased entropy.

THE FUTURE IS NOW
>>
>>101912310
I'm serious. Look at this. This stuff stimulates creativity and fires up the imagination.
This thing exists.
>>
>>
File: file.png (6 KB, 998x24)
6 KB
6 KB PNG
Kitta kitta!!
>>
File: 2548324703.png (1.56 MB, 768x1344)
1.56 MB
1.56 MB PNG
>>
File: ComfyUI_00533_.png (1.8 MB, 1152x1536)
1.8 MB
1.8 MB PNG
>>
>>101911615
holy shit you are horrible
>>
File: COMICFAIL.png (1.31 MB, 768x1536)
1.31 MB
1.31 MB PNG
And then I had this idea: I make FLUX draw each style in its own panel like in a comic, and then I make the characters join at the end so both styles are shown! Like this:
>3 Panel comic. The top one contains children card illustration vintage style page scan of cute sunflower hedgehog at the park. kindergarden eyes. The middle one depicts Sonic the hedgehog. The bottom one has both of them sit with each other, and she offers the flower to Sonic.
Except it didn't work...
Man, FLUX looks waaaaayyy smarter when other people use it...
>>
File: file.png (219 KB, 857x1650)
219 KB
219 KB PNG
>>101912310
>>101912359
it's pretty cool that all this can come from nothing
>>
>>101912448
See, the really cool thing is that it does not come "from nothing". It comes from a large part of our culture, which is encoded in these models. In a way, it comes from the physicalized, form of our collective unconscious. Statistically manifested.
>>
>>101912231
How did you prompt that?
>>
>>101912253
Oh like this one I just got.
>>
File: ComfyUI_02446_.png (1.44 MB, 1216x832)
1.44 MB
1.44 MB PNG
>>
>>101912448
pretty nice, is that GPT-4o?
>>
File: ComfyUI_00534_.png (1.7 MB, 1152x1536)
1.7 MB
1.7 MB PNG
>>
File: torch-egg_.jpg (289 KB, 1024x1024)
289 KB
289 KB JPG
>>101912379
>>
>>101912490
This image is a photograph of a vintage paperback book cover titled "Robotic future" by 4CHAN Press. The cover art features a surrealistic illustration of a robot with a human-like face and long, flowing, golden-brown hair. The robot's face has a blank, expressionless look, and it is wearing a red robe with a pattern of intricate, dark designs.
The background of the cover is a stylized depiction of a building with turrets and a large, domed structure reminiscent of the U.S. Capitol Building. The building is rendered in muted tones of brown, gray, and beige, giving it an aged, historical feel. The sky above the building is a deep blue, and the horizon is slightly hazy, suggesting a dreamlike or fantastical setting.
The title of the book is prominently displayed in bold, green text at the top of the cover, while the author's name and publication number are listed in smaller, green text in the upper right corner. The cover has a slightly worn and aged appearance, with visible creases and a slightly yellowed hue, typical of vintage paperback covers.
>>
>>101912418
nice
>>
File: file.png (1.11 MB, 896x1152)
1.11 MB
1.11 MB PNG
>>
>>101912310
>>101912231
These gens made me think of Children of the Whales and Gargantia on the Verdurous Planet for some reason.
>>
>>101912594
Thanks, FLUX really rules at things like that!
>>
>>101912657
Yeah it's astonishingly good.
I give random idea to an LLM, it fluffs it for me, then I give the fluff to Flux, which outputs amazing results.
Too bad it's shit at poses and nsfw, it would the perfect model.
>>
So, which quant should I use now? I am currently using fp8 (have 12GB of vram)
>>
File: file.png (736 KB, 512x512)
736 KB
736 KB PNG
HOLY SHIT
It's schnell q4_0 at 512x512, and it took 3 minutes, but I'm running Flux on an AMD RX 5700 XT with locally built torch and torchvision using rocm 6.2, which means I can run accelerated textgen with the same libraries and not have to rely on weird hacks anymore.
It's probably useless and I will go back to SD 1.5, but I'm so fucking happy. I thought this would be impossible.
>>
>>101912688
Q1_0
>>
>>101912695
0_0
>>
>>101912694
congrats anon!
>>
>>101912694
Good job!
>>
>>101912708
uwu
>>
>>101912694
You know that you could have just built it with patch or use a distro packages that build with those. But, it's nice they finally fixed it upstream.
>>
File: fp114.jpg (262 KB, 1344x768)
262 KB
262 KB JPG
>>101912688
QQ
>>
where the fuck do you put the vae in forge
>>
>>101912694
very nice
>>
>>101912750
The vae folder
>>
>>101912750
youre probably in the right directory but rename it .safetensor instead of .sft
>>
>>101912750
the square hole
>>
File: file.png (1.24 MB, 896x1152)
1.24 MB
1.24 MB PNG
It keeps on trying to merge multiple jets into one, but a few are singular.
>>
File: ComfyUI_02463_.png (1.39 MB, 1216x832)
1.39 MB
1.39 MB PNG
>>
>>101912769
.safetensors with an s i mean
>>
>>101912750
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050
>>
File: ComfyUI_02469_.png (1.67 MB, 768x1344)
1.67 MB
1.67 MB PNG
>>
>>101912694
Just gen at 1024x1024 and depict 4 panels and what is in them, whatever time it takes it's 1/4 of it.
>>
>>101912777
>>101912831
very gay
>>
>>101912777
>>101912831
very chiny
>>
File: 00076-3093794071.png (1.22 MB, 1280x720)
1.22 MB
1.22 MB PNG
>>
File: 00016-3501321742.png (1.04 MB, 1024x1024)
1.04 MB
1.04 MB PNG
Im going to hunt down that anon from last thread who insulted me for wanting to get a workflow which he referred to as "childplay".

I have a couple of words for him.
>>
File: ComfyUI_02475_.png (1.45 MB, 832x1216)
1.45 MB
1.45 MB PNG
>>101912861
>>101912868

shut up chud!
>>
>>
>>101912942
Are you ever going to post your workflow?
>>
>>101912949
You are not entitled to a catbox
>>
File: 00017-1801798314.png (770 KB, 1024x1024)
770 KB
770 KB PNG
>>
>>101912949
Are you asking specifically about the 3D conversion?
>>
>>101913037
Yeah. The results seem pretty decent. Better than what I've seen in the past.
>>
>>101912222
Oh yeah, that makes sense
I thought the first girl looked like Lauren Mayberry
>>
The day flux can make pron, I can feel it!
>>
>>101913072
Talk to you in 2025
>>
File: Capture.png (34 KB, 1030x444)
34 KB
34 KB PNG
>>101913060
I use iw3 with pic related settings:
https://github.com/nagadomi/nunif
>>
>>101911066
lust provoking image it up
>>
>>101913112
>I want to watch any 2D video as 3D video on my VR device, so I developed this very personal tool.
>iw3 provides the ability to convert any 2D image/video into side-by-side 3D image/video.
Oh cool. Man I can't wait to try all these nifty things after the Valve Deckard comes out. It's going to be great.
Thanks.
>>
File: fp115.jpg (223 KB, 1344x768)
223 KB
223 KB JPG
>>
File: grid-0161.jpg (485 KB, 1536x2688)
485 KB
485 KB JPG
>>
File: 1704617548160251.jpg (159 KB, 1024x1024)
159 KB
159 KB JPG
>>
File: ComfyUI_20_.jpg (1.47 MB, 2048x2048)
1.47 MB
1.47 MB JPG
>>101913137
>>
>>101912996
Looks like a legit screenshot, what was the prompt?
>>
File: 00114-3574905596.png (1.35 MB, 1152x896)
1.35 MB
1.35 MB PNG
>>101913154
>>
File: ComfyUI_00024_.png (1.8 MB, 1824x1248)
1.8 MB
1.8 MB PNG
flux won
>>
Flux GGUF 4_1 and 5_1 got added (also they improved the 8.0)
>>
>>101913144
It really is neato.
>>
File: Chen sketch.jpg (362 KB, 1280x1856)
362 KB
362 KB JPG
>>
>>101913205
Sauce?
>>
Oh fug I missed a whole ass thread while I was asleep, that doesn't normally happen.
I guess we back
>>
>>101913247
https://github.com/city96/ComfyUI-GGUF/commit/88fb6fa0014850615ca5b3e0ec1c018f67319237
>Use FP32 to do the calculations while dequantizing to improve existing Q8_0 quants and to avoid double dtype conversion.
>Add Q4_1 and Q5_1 dequant kernels
Can find Q4_1 and Q5_1 quant on the same HF repo.
>>
>>101913172
"a still frame from a 90s anime tv show of a girl screaming captioned in the bottom "my love has died""
>>
>>101913200
Where's her other leg?
>>
File: file.png (747 KB, 488x600)
747 KB
747 KB PNG
>>101912844
>just gen at 1024x1024
Any side being above 512 is out of the question right now. Both forge and comfy (yes, I moved clip and vae to cpu) run out of memory if you look them sideways.
Can anybody share memory optimization tricks for forge? I'm working with 8 GB here.
>>
File: 1711732093717735.png (1023 KB, 1024x768)
1023 KB
1023 KB PNG
>>101912776
it's not bad. got the rafale's fuel probe but the lines are mostly right.
>>
File: forge.jpg (287 KB, 1853x913)
287 KB
287 KB JPG
>>101913352
1 minute 30 second gens with this setup, 8gb vram
>>
>>101913261
Seems like the Q8_0 is the same
>>
>>101913413
Mine looks the same except I'm using schnell so that I only do 4 steps. But with the dimensions you're using, I'm guaranteed to OOM. Do you have anything else behind the scenes?
>>
File: fp118.jpg (215 KB, 1344x768)
215 KB
215 KB JPG
>>
File: 1710642674077054.png (2.07 MB, 1024x1024)
2.07 MB
2.07 MB PNG
>>
>>101913439
Mmmmm, Olives
>>
>>101913448
nice
>>
File: fp119.jpg (172 KB, 1344x768)
172 KB
172 KB JPG
>>
Erm am I supposed to believe that a computer made all this
>>
File: ComfyUI_02515_.png (1.34 MB, 832x1216)
1.34 MB
1.34 MB PNG
>>
>>101913428
not that i know of.
fresh install of forge today though.
>>
guys i have an urgent question about forge, everything is working but
how do i put it in dark mode???
>>
I'm gonna kill myself by the time I'm done captioning this dataset with joycaption

I think I'm half way to becoming a purple prose loving fantasy author with the amount of shit I've had to correct
>>
>>101913535
--theme=dark
>>
File: ifx41.png (1.45 MB, 1024x1024)
1.45 MB
1.45 MB PNG
>>
>>101913557
British people eat this btw
>>
File: ComfyUI_Flux_9161.jpg (201 KB, 1024x1024)
201 KB
201 KB JPG
>>
Mmh... Forge is idiotically slow even on SD. sdwebui isn't like this.
>>
File: ifx42.png (1.44 MB, 1024x1024)
1.44 MB
1.44 MB PNG
>>101913565
Seuss was american, it's green ham
>>
>>101913535
settings --> user interface --> gradio theme drop down menu --> pick one --> apply changes --> reload ui
>>
File: ComfyUI_31479_.png (961 KB, 1280x720)
961 KB
961 KB PNG
>>101913500
I drew it all.
>>
File: 00028-3959659492.png (985 KB, 1280x720)
985 KB
985 KB PNG
>>101913413
lol this prompt is great. Love how well it captures the novel look
>>
>>101913565
Russians definitely eat this.
Picrel is хoлoдeц
>>
File: 1703376327829011.png (2.13 MB, 1024x1024)
2.13 MB
2.13 MB PNG
>>
>>101913565
It's basically aspic.
>>
File: kholodets.jpg (48 KB, 450x338)
48 KB
48 KB JPG
>>101913637
picrel
>>
File: 0.jpg (183 KB, 1024x1024)
183 KB
183 KB JPG
>>
File: 1714312999698391.png (2 MB, 720x1280)
2 MB
2 MB PNG
>>
File: flux_00001_.png (991 KB, 1160x896)
991 KB
991 KB PNG
Finally managed to coax flux into making a saint seiya armor. It REALLY wanted to make an anime picture, but halfway through it changed course and went for realism. Still not there yet.
>>
File: 00132-2049185604.jpg (681 KB, 1920x1080)
681 KB
681 KB JPG
>>
I can't try myself since I'm still downloading those models, is it possible to merge those gguf models in Comfy? I want to make schizo merge gguf model if you ask
>>
>>101913747
Almost a Moebius feel.
>>
>>101913751
why
>>
>>101913500
Have you gone to this link?:
https://huggingface.co/spaces/gokaygokay/FLUX.1-dev-with-Captioner
Uploaded a picture and hit generate image?
And gotten an image?
If so, what's your theory about from where it came from?
>>
File: file.png (1.11 MB, 1024x1024)
1.11 MB
1.11 MB PNG
Helloooo, still using Forge from February, anything new in speed or efficiency?
Also, if you're using a Pony-based model (or SDXL, I don't really know about this whole Flux thing or w/e or whether it does best doki), what're yer gee pee yuu and how does it work for you?
Been thinking of getting a used 2070 with 8GB or 3060 with 12GB (3060 because I have one, thus I know how it performs) so I can hook it up to my homeserver and share it over interwebz with a poor lad using a tater tot and mf Bing, so I wanna know how the 20-series performs but I don't really know where to find those stats
>>
File: 1715655880698166.png (2.08 MB, 1024x1024)
2.08 MB
2.08 MB PNG
>>
>>101913782
The cloud
>>
>>101913829
uuuhhh krill yourself?
>>
File: ComfyUI_02528_.png (1.39 MB, 1344x768)
1.39 MB
1.39 MB PNG
another "messing with all the options because i don't know what they do" gen before i go to bed
>>
>>101913747
prompt pls
>>
>>101913829
Go for 3xxx and 4xxx with at least 12GB of VRAM or better, 16+.
>>
>>101913870
"acoustic painter having a spasm"
>>
File: file.png (1.56 MB, 1280x720)
1.56 MB
1.56 MB PNG
>>101913555
lol
>>101913593
thanks and wow thats a lot of themes
>>
File: 00088-946292504.png (1001 KB, 1280x720)
1001 KB
1001 KB PNG
>>101913635
i'm enjoying all the fun ways it's failing to render "retarded"
>>
>>101913886
it was pruned right
>>
>>101913886
It seems quite easy to find the words it never trained on or avoided.
>>
>>101913878
Didn't seem to work for me
>>
File: ifx46w.jpg (283 KB, 1600x1600)
283 KB
283 KB JPG
>>
>>101913902
you are such a n-word individual
>>
File: file.png (1.34 MB, 896x896)
1.34 MB
1.34 MB PNG
This image took 4 minutes.
>>
>>101913912
thanks i saved it, to my pc
>>
>>101913854
You mean like, eat 10 kilos of krill?
>>101913873
so, I should hunt for a 3060 then?
Why, exactly? I'm not gonna let the dude load 2 gorillion loras, and unless told otherwise I'm gonna open up a Forge checkpoint which should run fine
Anything higher is so piss bloody expensive in this retarded country market and not worth it even if I used the supposedly superior card to replace my current 3060 in my daily driver
>>
>>101913912
And it looks like that because I used negative which you're not supposed to do on flux?
>>101913923
You owe me a million dollars
>>
File: 00089-966360576.png (1.06 MB, 1280x720)
1.06 MB
1.06 MB PNG
>>101913894
>>101913893
this one works though
>>
How in the actual ever loving fuck do you use Joy Caption locally? Huggingface's gpu limits are dicks and buying pro doesn't fix them..
>>
Open release of NAI v3 in 10 days kek
>>
>>101913923
"No, help yourself. 3600 a bottle. Please enjoy it"
>>
>>101913958
I'm running it and I'm retarded so I don't know
>>
>>101913932
seeing on nvidias site that there's a 12gb 2060 but I doubt I'll be able to find it for any decent price
>>
File: 1696143015932.png (12 KB, 411x137)
12 KB
12 KB PNG
>>101913959
>5 days
>>
File: 1705900264291421.png (2.56 MB, 1024x1024)
2.56 MB
2.56 MB PNG
>>
>>101913932
>Why, exactly
Options, it's the better cards for generation, well supported and working well by almost all tools and models with no fuss or endless tinkering.
>>
>>101913989
wait, what, why?
>>
>>101913870
I can't post it because I'm embarassed
>>
>>101914002
alr fair
I'll take a look at getting a 3060 then. Anything more (expensive), and it'd be better to swap it with my daily driver's 3060, and it's not really worth it anyway since I'm already going to be spending the rough equivalent on a better chair.
>>
File: 00036-1263942962.png (1.55 MB, 1632x1152)
1.55 MB
1.55 MB PNG
>>
File: fp117.jpg (234 KB, 1344x768)
234 KB
234 KB JPG
>>
File: 00090-3994146773.png (1.73 MB, 1280x720)
1.73 MB
1.73 MB PNG
1970s illustration by Ralph Bakshi of a large urban area with tall fantasy buildings, outside of the city are tiny houses and rivers and streams.
>>
>>101913958
I tried to make it work but utterly failed, I'll just wait for it to come out of pre-alpha, maybe someone will write an installation guide for complete retards by then.
>>
File: 1720965107764047.png (2.42 MB, 1024x1024)
2.42 MB
2.42 MB PNG
>>
>>101913934
>negative which you're not supposed to do on flux?
lol you're not? been using it this whole time, no wonder it didn' seem to do anything.
>>
>>101914131
I didn't use that prompt
>>
>try to load some random ass lora on flux
>gen speed drops insanely quickly
??? -1 i guess
>>
>>101914209
Probably filled your VRAM to 99.999%
>>
>>101914216
quite
>>
File: flux_l_3.jpg (479 KB, 1216x832)
479 KB
479 KB JPG
>>
File: cfg.jpg (965 KB, 2048x1536)
965 KB
965 KB JPG
Looks like you should put distilled CFG scale pretty low if you want to get styles like pencil/oil painting etc. because having it at the typical 3 or so will make the model prefer well defined images over stylistic output and throw all your fine art styles in the trash. However putting it too low might make the image poorly defined so that's when you can raise the regular CFG a little to both preserve rougher styles while making them more defined.
Probably common knowledge but I just figured this shit out now and feel dumb because I was living with the mindset of higher number = adheres to prompt better.
>>
>>
>>
File: ifx44.png (986 KB, 1024x1024)
986 KB
986 KB PNG
>>
File: 1717318355164294.png (2.39 MB, 1024x1024)
2.39 MB
2.39 MB PNG
>>
>>101911881
How come I never run out of memory? I thought it must just use ram or something, I only have 8gb.
>>
>>
>>
File: ComfyUI_00031_.png (2.27 MB, 1248x1824)
2.27 MB
2.27 MB PNG
>>
and with some low cfg so i can add neg prompt lol it takes 2 years to gen
>>
>>101914459
NTA, but I can confirm fucky things happen when I load LoRAs on comfy too. No idea why either, sometimes it ooms, others it doesnt
>>
Come and get it...
>>101914501
>>101914501
>>101914501
>>
>>101911352
nice
>>
>>101914272
>>101914317
>landscape mode
nigger i had to zoom that shit down to 6% to see it
can't nobody cross their eyes that much



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.