[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: 1709735436775199.png (1.94 MB, 1504x768)
1.94 MB
1.94 MB PNG
Discussion of free and open source text-to-image models

Previous /ldg/ bread : >>101685374

>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
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

>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

>Kolors
https://gokaygokay-kolors.hf.space
Nodes: https://github.com/kijai/ComfyUI-KwaiKolorsWrapper

>AuraFlow
https://fal.ai/models/fal-ai/aura-flow
https://huggingface.co/fal/AuraFlows

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

>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

>Related boards
>>>/h/hdg
>>>/e/edg
>>>/d/ddg
>>>/b/degen
>>>/vt/vtai
>>>/aco/sdg
>>>/trash/sdg
>>
File: ComfyUI_00322_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>101693731
>>
>no collage
still blessed
>>
>>101693452 non-local local models
>>
>>101694073
Stop changing the names of your 50 trillion AI generals. I'm tired of adding new rules to my filter.
>>
File: FD_00149_.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>
>>101694106
Are you logging my prompts?
>>
File: ComfyUI_01001_.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>>
>>101694137
Even if he told you no, would you believe him?
>>
>not filtering all generals
>>
>>101694073
I haven't played with local models for like a year. I was really good with A111 and all the tools of the time.

Do we have local models now that can do text? Also, can the new models that do text, can they be trained? Or is there a LoRA for existing models to do text?
>>
>>101694073
Imagine if we had a collage thodesu
>>
File: FD_00159_.png (1.4 MB, 1024x1024)
1.4 MB
1.4 MB PNG
>>
File: ComfyUI_01004_.png (1021 KB, 1024x1024)
1021 KB
1021 KB PNG
>>101694128
Move on sir, this is an official thread split
>>
File: FD_00166_.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>
>>101694162
only one bred lets be real
>>
File: ComfyUI_Flux_1153.jpg (127 KB, 1152x864)
127 KB
127 KB JPG
>>
File: file.png (1.7 MB, 1024x1024)
1.7 MB
1.7 MB PNG
>>101694162
gotta catch me first anon
>>
File: FD_00173_.png (1.1 MB, 1024x1024)
1.1 MB
1.1 MB PNG
>>101694128
>>
File: FD_00174_.png (1.24 MB, 1024x1024)
1.24 MB
1.24 MB PNG
>>
>>101694162
Why the hell do you need a thread split? This is just /b/ for jpegs.
>>101694199
>>101694223
I look exactly like this and say these exact things.
>>
File: ComfyUI_01009_.png (1.01 MB, 1024x1024)
1.01 MB
1.01 MB PNG
>>101694266
Actual tech discussion and not obsessing over avatars
>>
>>101694266
i know sdg is getting desperate because of the possibility of this general growing bigger than theirs, but you don't have to be so obvious about it. What did you expect? That SAI was going to be a key piece forever in img gen? or that you can just kill ldg whenever we get a wave of new anons?
>>
>>101694307
>Actual tech discussion
Where?
>>
>>101694073
Retard OP

Previous thread
>>101689050
>>
>>101694320
fuck off back to sdg
>>
>>101694316
>>101694341
For fuck's sake, you people have warring factions now?
>>
File: ComfyUI_Flux_1145.jpg (99 KB, 1152x864)
99 KB
99 KB JPG
>>
>>101694174
Is this flux? I can't for the life of me get anything but realism. What is your prompt?
>>
>>101694358
Not really it's just a couple of schizos. Same shit every general deals with. Stable Diffusion devs shit the bed though and there's significantly better alternatives so /sdg/ doesn't make sense anymore. The only ones left there are avatarfags who have some weird friendships
>>
File: ComfyUI_Flux_01888_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
>>
File: ComfyUI_Flux_01890_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
>>
>>101694383
Comic book illustration of xyz
>>
File: ComfyUI_Flux_01892_.png (1.24 MB, 1024x1024)
1.24 MB
1.24 MB PNG
>>
File: 1722644278194909.jpg (235 KB, 1600x1600)
235 KB
235 KB JPG
>>
>>101694501
https://files.catbox.moe/rn4ede.png
>>
>>101694528
wait cum isn't allowed?
>>
File: ComfyUI_Flux_01896_.png (1.08 MB, 1024x1024)
1.08 MB
1.08 MB PNG
>>
>>101694543
Not on a blue board I don't think
https://files.catbox.moe/f9qssj.png
>>
File: flux1-dev-273816418.png (1.54 MB, 1024x1024)
1.54 MB
1.54 MB PNG
>>101694486
Thanks. I'm going for a more realistic fine art style but even if I exaggerate and add powerful words like impressionism I end up getting photorealism instead. I'm just throwing whatever I can think of at it and seeing what sticks. I'm getting there, I guess.
>>
>>101694374
Nice
>>
File: dungeon2.png (1.65 MB, 1024x1024)
1.65 MB
1.65 MB PNG
>>101694606
I get the exact same style with "oil painting by Frank Frazetta" (doesn't seem to recognize him or adds too much photorealism).
>>
>>101694931
common issue, seen people mention it all over the place
>>
File: flux1-dev-131709732.png (1.47 MB, 1024x1024)
1.47 MB
1.47 MB PNG
>>101694931
Pretty good. It's irritating trying to get the vectors to point the right way. I guess I can see how that style is midway between Frazetta and the default realism. I've actually got my environments down pretty good finally, but now the characters are a bit too stylized.
>>
File: ComfyUI_Flux_01901_.png (1.15 MB, 1024x1024)
1.15 MB
1.15 MB PNG
>>
cozy thread
>>
>>101694412
Almost looking ok
>>
File: ComfyUI_Flux_01905_.png (987 KB, 1024x1024)
987 KB
987 KB PNG
>>
>>101695096
They look ok check gens from last thread
>>101693689
https://files.catbox.moe/jyi4yy.png
>>
>>101695097
the body colors are a bit off but the lighting on the faces is very realistic, impressive model
>>
>>
>>101694586
How do you get such a good result?
>>
File: ComfyUI_01579_.png (778 KB, 1152x832)
778 KB
778 KB PNG
>>
>>101695143
This is the most fun I've had since Hunyuan, perfect soles, realistic girls are possible, etc...
https://files.catbox.moe/9i007g.png

The model is very fun to play with, we have not seen its full capabilities yet, with LoRAs I can only imagine what will be possible.

>>101695236
Drag and drop into comfyUI, prompt is in file

I'm still experimenting with key words, but I found that a combination of Normal photo, an iPhone photo of X (X being your prompt), soles towards the camera, (optional: oily), foot, feet, soles, fetish, toes,

works well for soles. Possibly transfers over to different styles and keyword combinations.
>>
>>101694316
I'm a simple man. I see discord trannies I leave. So I refuse to visit the other thread anymore. I couldn't care less about the name
>>
File: ComfyUI_01580_.png (703 KB, 1152x832)
703 KB
703 KB PNG
>iphone photo of a penis
I shit you not it just makes women in lingerie
>>
File: ComfyUI_01570_.png (1.22 MB, 896x1152)
1.22 MB
1.22 MB PNG
>>101695401
welcome /ldg/
>>
>>101695404
Nice gen desu
>>
File: ComfyUI_01584_.png (815 KB, 832x1152)
815 KB
815 KB PNG
>>101695501
bro I'm telling you just prompt for dick pics and you receive women
>>
>>101695512
That's a problem. Even if you ask for nude men?
Do we have image to image for flux yet?
>>
>>
>>101695348
Another simple but effective idea
https://files.catbox.moe/exhyfc.png
>>
>>101695554
>Even if you ask for nude men?
not rolling the dice again

>Do we have image to image for flux yet?
yeah
>>
>>101695564
but that doesn't look like cum
>>
>>101695348
>This is the most fun I've had since Hunyuan, perfect soles, realistic girls are possible, etc...
the only downside i'm finding is that some keywords can change the output dramatically just because, adding bangs will instantly make most images about asian women
>>
>>101695334
gud
>>
File: ComfyUI_01016_.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
>>
File: ComfyUI_01018_.png (891 KB, 1024x1024)
891 KB
891 KB PNG
>>101695598
Wow you're right.. but it can be corrected
>A European woman with bangs stands in front of a white wall with a black sweater outfit
>>
>>101695564
>Got exact same photo twice with this prompt
Weird, what are chances of converging?
https://files.catbox.moe/9cu059.png
>>
>>101695584
Refer to >>101695659
It was the second (third) gen with this prompt.
>>
File: ComfyUI_Flux_01923_.png (1.15 MB, 1024x1024)
1.15 MB
1.15 MB PNG
Alright enough feet girls for me today, time for anime
>>
File: Image.jpg (1.61 MB, 2688x1152)
1.61 MB
1.61 MB JPG
>bizzare, acid trip,
>>
File: ComfyUI_Flux_01925_.png (1.18 MB, 1024x1024)
1.18 MB
1.18 MB PNG
>>
>>
>>
So... what was SAI spending 8 million dollars a month on?
>>
>>101695554
Yeah, just use it the same way for normal SD in comfy
The issue is that there's a very fine line between getting it to not strictly adhere to the input, and not even using the input at 0.8 denoise.
It just seems like it's absolutely unusable for img2img
>>
>>101696002
safety
>>
File: ComfyUI_13122_.jpg (603 KB, 1280x768)
603 KB
603 KB JPG
>>
>>101695860
>The real people in the background.
>>
>>101694073
What's the word on a Flux NSFW finetune?
Is it happening?
>>
>>101696088
we cant even tune it yet
>>
>>101696002
SAI walked so FAL could run
>>
>>101696088
It's been out for a day dude. I think people are still collecting their spaghetti.
>>
>>101696088
lodestone furries are on it
>>
>>101695348
Thanks anon, I'm new to Comfy so I didn't know you could do that.
>>
File: Flux_00003_.png (1.15 MB, 1024x1024)
1.15 MB
1.15 MB PNG
Looks like another model that doesn't understand that within a banana peel is in fact not another banana.
>>
File: ComfyUI_00905.png (1.28 MB, 1024x1024)
1.28 MB
1.28 MB PNG
just throwing paint at the wall here
>>
>>101696117
Very nice
>>
>>101696154
for future reference: the workflow gets stored in the metadata. so converting formats or uploading to 4chan wipes away this ability
>>
File: Flux_00004_.png (1.21 MB, 1024x1024)
1.21 MB
1.21 MB PNG
>>
>>101696117
there's a lot of protected IP's and celebrity likenesses in flux, I don't think they were too concerned with getting permission before scraping
hope that doesn't come back to haunt them, for real
>>
File: ComfyUI_01010_.png (1.04 MB, 1024x1024)
1.04 MB
1.04 MB PNG
>>101696117
FAL just hosts it. Made by https://blackforestlabs.ai/
>>
>>101696215
I'd like to think that taking a copyrighted work, converting it to a 12 billion dimensional vector, then chewing it up and spitting it out n-billion times (and then not even releasing the vector just a derivative of the vector) counts for something
>>
File: ComfyUI_13133_.jpg (315 KB, 1280x768)
315 KB
315 KB JPG
>>101696048
>>
>>101696117
FAL and Replicate are just their partners
>>
File: ComfyUI_Flux_01932.png (1.06 MB, 1024x1024)
1.06 MB
1.06 MB PNG
>>101696087
>The real people in the background.

Well, it was a very long prompt and it had implications of realism, even with word camera, etc...
>>
>>101696286
Thank you. you are correct. I incorrectly thought Black forest labs was the development team and FAL was the company.
>>
>>101696350
>FAL was the company.
Pretty sure FAL is just some guy. He made AuraFlow
>>
File: Flux_00015_.png (887 KB, 1024x768)
887 KB
887 KB PNG
When I ask it to generate a world of warcraft screenshot it usually just generates boring epic concept art.
This is the only one that came kind of close. How can I push it more in this direction?
>>
File: ComfyUI_temp_riblx_00039_.jpg (1.42 MB, 1792x2304)
1.42 MB
1.42 MB JPG
>>
File: Flux_00029_.png (1.05 MB, 1024x768)
1.05 MB
1.05 MB PNG
It's fucking over for /sdg/
>>
>>101696472
flux is still diffusion isn't it?
>>
File: ComfyUI_01020_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
>>101696363
https://x.com/cloneofsimo/status/1811562996541624830
>>
File: FD_00180_.png (944 KB, 1024x1024)
944 KB
944 KB PNG
>>101696374
idk but in trying I got this aesthetic as fuck model
>>
>>101696514
thats what the page says
>>
>>101696514
It's not Stable Diffusion or SAI
3.1 desparately needs to be able to gen porn out of the box or it literally is over for them
>>
File: Flux_00033_.png (1.15 MB, 1024x768)
1.15 MB
1.15 MB PNG
>>
>>101696514
/sdg/ is more for running things in the cloud but why do that when flux works out of the box without censorships and control from horrifically ran companies?
>>
So that dead baby pic was fake, right
>>
>>101696472
Time marches on.
>Don't cry because it's over, smile because it happened
>>
>>101696530
they just found a new investor, they'll be fine
>>
File: ComfyUI_Flux_01936_.png (1.44 MB, 1024x1024)
1.44 MB
1.44 MB PNG
>>101696215
Don't think anything'll happen. ClosedAI has done it with GPT 4, Dalle, etc... Judges have dismissed such lawsuits

https://arstechnica.com/tech-policy/2024/02/judge-sides-with-openai-dismisses-bulk-of-book-authors-copyright-claims/
https://www.legal.io/articles/5516216/Judge-Throws-Out-Majority-of-Claims-in-GitHub-Copilot-Lawsuit

If AI is censored in the States, China will get ahead, stifling innovation, this is a legitimate argument that is brought up to the judges
>>
>>101696550
it wasn't, the prompt was literally shared a few posts below, did you check it?
>>
File: ComfyUI_00035_.png (839 KB, 1024x1024)
839 KB
839 KB PNG
>>101696546
>without censorships
It's just less censored than SD but it's still pretty cucked. Unless women are supposed to have buttons for nipples
Still, at least it draws them, unlike SD3 which just makes picrel.
>>
>>101696561
yeah but that's microsoft
I know it's not fair, but that's how things are
>>
>>101696571
Nah, that was completely fake. Take a look at >>101694183
>>
>>101696614
He used dev, try schnell, it gets much closer results to the picrel
>>
>>101696554
>they just found a new investor
Imagine still getting scammed by SAI
>>
>>101696601
>yeah but that's microsoft

If judges clamp down on the smaller guy then they will 100% clamp down on the bigger guy. Billions of dollars would be lost, that means ClosedAI can no longer use the GPT 4 weights without owing a ton of money themselves as well. There has never been such a thing as paying to use a dataset or owing anyone anything for machine generated outputs.
>>
>>101696639
yesterday they were top of the open source class
>>
File: FD_00184_.png (1.26 MB, 1024x1024)
1.26 MB
1.26 MB PNG
>>101696554
How many chromosomes does the investor have?
>>
I haven't had this much fun since SDXL 1.0 release a year ago
I really feel like FLUX is the first big spark of a new paradigm
>>
File: FD_00029_.png (1.36 MB, 1024x1024)
1.36 MB
1.36 MB PNG
>>101696650
Well now it's today and they are fucked.
>>
File: output_0000000009.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
Cool stuff. I have this model cloned to a local folder, and I have all the Python requirements in a venv. No one can take this from me.
>>
>>101696650
SD3 was an embarrassment, no question
whether flux dropped or not, they'd not be in a good spot now
>>
File: ComfyUI_Flux_01926_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>101696649
Now, realistically speaking there might be a law that bans you from using it to reproduce copyrighted characters or citing copyrighted work without asking for permission, but that's as far as that might go.
>>
does /ldg/ have a mascot?
>>
>>101696702
SD3 was by all accounts a self inflicted wound.
>>
>>101696719
Probably Miku, but I like ldg because it hasn't turned into a tranny chatroom for avatar fags yet.
>>
https://files.catbox.moe/bhzckn.png
>>
File: file.png (1.58 MB, 1280x768)
1.58 MB
1.58 MB PNG
Share your favorite prompts!
>>
>>101696702
isn't BFL made of the SAI team? or the ones who knew what they were doing anyway
SAI now is probably only middle management and diversity hires
>>
File: ComfyUI_temp_riblx_00045_.jpg (1.54 MB, 1792x2304)
1.54 MB
1.54 MB JPG
>>
File: Flux_00047_.png (967 KB, 1024x1024)
967 KB
967 KB PNG
>>
File: FD_00203_.png (1.26 MB, 1024x1024)
1.26 MB
1.26 MB PNG
We truly live in a society
>>
File: Flux_00048_.png (1.03 MB, 1024x1024)
1.03 MB
1.03 MB PNG
>>
official pixart bigma, lumina 2 and hunyuan finetune waiting room, now with flux 12b
>>
>>101696295
that's cute
>>
>>101696836
My only issue with this is that now we're getting some actually decent alternatives to SAI, we don't have all the useful tools that SAI models had built up over time and they can't be adapted to every new model. At some point, the community has to choose.
>>
>>101696852
image gen will split apart into little tribes based on their model of choice. so far we'll have the fluxian fatties, little pixshartists and the hunyawns. i'm not sure what to call the lumina people.
>>
File: Flux_00051_.png (1.32 MB, 1024x1024)
1.32 MB
1.32 MB PNG
pepperoni nipples
>>
File: FD_00210_.png (1.53 MB, 1024x1024)
1.53 MB
1.53 MB PNG
Still can't get Jesus to surf on his crucifix. One of these days one of these models will get it.
>>
>>101696885
lumatics
>>
>>101696907
lumaniacs
>>
>>101696852
Every other ai general is talking about Flux. It's over for the competition.
>>
>>101696929
watch bigma shake heaven and earth
>>
Trying to get flux to generate N64 game footage, any suggestions?
>>
>>101696929
>very other ai general is talking about Flux.
Only really matter if reddit talk about it.
>>
>>101696946
just wait a year and it will start actually producing N64 game footage
>>
>>101696619
Interesting, perhaps schnell is the less censored model then, too bad it kind of sucks compared to dev at generating consistent feet and hands, or maybe it's prompted differently?
>>
File: 71335180734.png (1.28 MB, 1544x352)
1.28 MB
1.28 MB PNG
>>
File: Flux_00055_.png (1.65 MB, 1024x1024)
1.65 MB
1.65 MB PNG
>>101696900
I thought "Pfft I can do that" but nope. Really hard. No idea where the horse came from.
>>
File: flux1-schnell_00007_.jpg (415 KB, 768x1024)
415 KB
415 KB JPG
>>101696738
>>
>>101696950
It's all reddit is talking about
>>
>>101696960

while that is true, im just gonna mess around with it, see how
>>
>>101697040
closer
>>
File: 148181311.jpg (435 KB, 1472x1472)
435 KB
435 KB JPG
>>
>>101696946
i thought this was a screenshot of a gen with an emulator window open to compare kek
>>
File: FD_00230_.png (1.55 MB, 1024x1024)
1.55 MB
1.55 MB PNG
>>
Oh man, flux actually has me excited. It just seems to "get" the prompt.
>>
>>101697092
i keep thinking theyre real for a second too
>>
File: 00035-2679405515.jpg (786 KB, 1489x2303)
786 KB
786 KB JPG
>stuck on finetunes
I wish flux knew asuka
>>
File: ComfyUI_00938.png (480 KB, 1024x512)
480 KB
480 KB PNG
>>101696946
I had put "twitch stream", and it started adding screenshots of tweets to the image, i did get one image that looked like it was from twitch.
In that Previous one, i had put emulator, Game Ui, but that was 2 broad.
>>
>>101697128
gameplay, in-game footage?
>>
File: FD_00237_.png (547 KB, 1024x1024)
547 KB
547 KB PNG
>>
>>101697128
try screenshot of game, or in-game footage
>>
>>101697128
>You're Gay
anon, it's trying to tell you something...
>>
File: 3.jpg (1 MB, 2048x2048)
1 MB
1 MB JPG
a green marble obelisk with golden veins on the left, on the center a ebony fabergƩ egg with glowing pink neon trims, on the right a metal pyramid with a cyan and red checkerboard pattern, river pebbles backdrop
>>
File: FD_00242_.png (536 KB, 1024x1024)
536 KB
536 KB PNG
>>
File: 84.png (1.84 MB, 1024x1024)
1.84 MB
1.84 MB PNG
prompt:"your mom"
>>
File: Flux_00066_.png (921 KB, 1024x1024)
921 KB
921 KB PNG
>>
I'm going to killmyself
>>
File: ComfyUI_00083_.png (1 MB, 1024x1024)
1 MB
1 MB PNG
>>
>>101697214
livestream it post link you won't do it faggot
>>
File: Flux_00072_.png (647 KB, 1024x1024)
647 KB
647 KB PNG
>>
>doesn't know mainstream characters
>doesn't know artists
>struggle hard with nudity
Yeah I'm thinking the flux honey moon phase is over for me. Generating memes gets old after a day.
>>
>>101697199
>>101697316
kek
>>
>>101697338
>doesn't know mainstream characters
it does
>doesn't know artists
prompt clip separately
>>
File: FD_00261_.png (578 KB, 1024x1024)
578 KB
578 KB PNG
>>
>>101697350
>prompt clip separately
what do you mean by this, have an example
>>
File: FD_00266_.png (1.23 MB, 1024x1024)
1.23 MB
1.23 MB PNG
>>
>>101697350
>sub(g)ect and sty(l)e making a comeback
no way
>>
File: Flux_00076_.png (1.25 MB, 1024x1024)
1.25 MB
1.25 MB PNG
>>
>>101697352
lol
>>
File: FD_00271_.png (1.23 MB, 1024x1024)
1.23 MB
1.23 MB PNG
I'm having so much fun.
>>
File: ComfyUI_00082_.png (840 KB, 1024x1024)
840 KB
840 KB PNG
>>
File: 408031551020.png (3.93 MB, 1919x1048)
3.93 MB
3.93 MB PNG
noob here. what's the easiest way to train a lora for in game character? do i just take bunch of screenshot
>>
File: Flux_00079_.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
>>
File: CumfyUI_00298_.png (805 KB, 768x1024)
805 KB
805 KB PNG
>>
>>101697350
It doesn't even know what 2B looks like, no matter how carefully I describe her in order to jog it's latent space memory
>>
>>101697479
>read the OP

if you just want a lora to use it, then check to see if one has already been made here:
https://civitai.com
if you want to train one for the experience, read these:
https://github.com/Nerogar/OneTrainer
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
>>
File: FD_00289_.png (1.61 MB, 1024x1024)
1.61 MB
1.61 MB PNG
>>
slutty robes vs robes
>>
>>101697597
Hunyuan knows, guess we'll have to wait for v2 of Hunyuan for it to follow prompts as good as Flux and truly know as much mainstream as possible
>>
>>101696852
One exception, Hunyuan has controlnet, kohya training, comfyui support, soon IPAdapter support, soon auto1111 support. We should soon have all but IPAdapter for Flux.
>>
File: ComfyUI_temp_riblx_00075_.jpg (3.6 MB, 1792x2304)
3.6 MB
3.6 MB JPG
>>
File: FD_00338_.png (1.41 MB, 1024x1024)
1.41 MB
1.41 MB PNG
>>
>>101694073
Flux is fucking good as fuck holy shit lol.
>>
File: ComfyUI_00107_.png (952 KB, 1024x1024)
952 KB
952 KB PNG
>>101697983
It's capable of a lot of stuff. Turns out that more parameters and not cucking the anatomy is all you need.
>>
File: ComfyUI_00072_.png (2.58 MB, 1920x1024)
2.58 MB
2.58 MB PNG
>>
>>101696002
DEI safety officers
>>
File: ComfyUI_00109_ (1).png (977 KB, 1024x1024)
977 KB
977 KB PNG
kek
>>
File: file.png (69 KB, 1093x355)
69 KB
69 KB PNG
I know i'm retarded, but

I'm trying to look up guides for hand detailers in comfyui, and they all have workflows withnodes that don't exist and trying to find them brings me to an online version page with no downloads. Where do I get this for the local version?
>>
So apparently Flux license doesn't allow finetuning for commercial purposes without permission. Sad if true.
>>
TensorRT Flux wen? It's so slow on 3090
>>
>>101698527
I was never gonna spend money on it anyway
>>
>>101698527
Don't most if not all current gen models have a similar stipulation?
>>
File: FD_00374_.png (1.11 MB, 1024x1024)
1.11 MB
1.11 MB PNG
>>
>>>/gif/27637708
>>
File: ComfyUI_temp_riblx_00099_.jpg (1.26 MB, 1792x2304)
1.26 MB
1.26 MB JPG
>>
So flux is basically the thing we should have gotten with SD3, right? Only question is, how do we train these things or make loras locally when it's so big? Can loras be trained on FP8?
>>
>>101698864
With text models people make 4-bit loras all the time.
>>
Can Flux do 2.5D graphics ala Doom/Duke 3D?
>>
>>101696809
kek
>>
>>101698931
>screenshot from a Build engine shooter. the player is shooting at a pixelated Donald Trump
>>
>>101698982
>>
>>101698982
Shiet that looks good. DALL-E 3 was the only one that really worked for me for that particular style, but this looks like a potential improvement, and presumably not as badly dogged as well. Guess it's time for me to finally learn Comfy
>>
Anyone figured out how to get the cli app from flux repo to load using the fp8 weights for tokenizet and encoder? I've tried swapping the torch code for safetensor code but it keeps trying to download the tokenizer and encoder from hf
>>
File: FLUX__00089_.png (852 KB, 768x1024)
852 KB
852 KB PNG
>>
>>101698982
Biden version
likeness isn't nearly as good sadly, probably way fewer pictures in the TD
>>
File: FLUX__00091_.png (1.08 MB, 768x1024)
1.08 MB
1.08 MB PNG
>>
File: 00014-4102347810.png (2.57 MB, 1344x1728)
2.57 MB
2.57 MB PNG
yes, they're sharing a drink they call loneliness, but it's better than drinkin' alone
>>
File: 1715624422499926.jpg (216 KB, 1024x1024)
216 KB
216 KB JPG
>>101699148
What happens with a prompt like

>screenshot of fps doom 1993 1994 on PS1 and N64, 3D pixel art sprite graphics, side view of parked 80s sedan in front of sidewalk, shuttered buildings behind, overcast grey sky, hud and gun pov, rust motifs, procedural textures, Phong shading, crt picture, ingame, screenshot, high resolution pixel art, polygon, HUD
>>
File: image (4).jpg (159 KB, 1024x768)
159 KB
159 KB JPG
Yesterdays thread was wild.
Not even a day after the stars aligned perfectly to give us a cool model like flux:
>some "random" anon comes in here. u-uhm guys i cant even post it here but have this link, flux made a picture of a little naked girl fucked up in a dumpster. this is loli nercophilia!
>immediately "another" anon comes in and says this needs to be regulated. baby violence is the line that shall not be crossed!
>"uhm, what would the police think if you showed them the picture?!?!"
crazy. i really hope we make it over the finish line with AI.
>>
>>101699228
Prime 1.5
>>
File: ComfyUI_temp_riblx_00107_.jpg (3.63 MB, 1792x2304)
3.63 MB
3.63 MB JPG
>>
File: Untitled.jpg (492 KB, 2048x768)
492 KB
492 KB JPG
tried with both schnell (left) and dev (right)
looks like neither of them wanted to do the HUD, could be too far back in the prompt.
>>
>>101699307
meant for >>101699236
>>
>>101695139
>>101694441
>>101694501
the fuck model is this
>>
>>101699322
you living under a rock? FLUX!
>>
>>101699307
>>101699321
Thanks, especially love the one on the left, would love to wonder around that place
>>
>>101699236
>>101699307
>screenshot of fps doom 1993 1994 on PS1 and N64, 3D pixel art sprite graphics, side view of parked 80s sedan in front of sidewalk, shuttered buildings behind, overcast grey sky, hud and gun pov, rust motifs, procedural textures, Phong shading, crt picture, ingame, screenshot, high resolution pixel art, polygon. at the bottom of the screen, the game HUD displays remaining health and other stats.
got it to add the HUD by being more verbose about it. not very coherent though
>>
>>101699322
>Literally everyone talking about flux on every possible location that discusses image generation
>WhAt MoDeL Is ThIs?!?
>>
File: ComfyUI_00261_.png (3.36 MB, 1024x1600)
3.36 MB
3.36 MB PNG
>>101699307
How do you use the dev one, I tried swapping it in comfy UI and it didn't work.
>>
>>101699361
nice
>>
>>101699351
improved the result by increasing the resolution from 1024x768 to 1280x960
looks like the lower res was too far out of distribution for the model and hurting its thinking a bit
>>
>>101699339
>>101699354
I am still using comfy UI with SDXL or whatever Ihavent prompted in a bit.
>>
>>101699361
shouldn't need to change your workflow at all to switch to dev
it just needs way more steps to converge, at least 20 steps. schnell is a turbo model and dev isn't
>>
File: Flux_00096_.png (896 KB, 1024x1024)
896 KB
896 KB PNG
>>
>>101699361
shouldn't need to change your workflow at all to switch to dev
it just needs way more steps to converge, at least 20 steps. schnell is a turbo model and dev isn't slower
>>
File: Flux_00097_.png (812 KB, 1024x1024)
812 KB
812 KB PNG
>>
>>101699454
4chanx users are such retards about deleted posts
just because your gay script lets you see deleted posts that doesn't mean I actually double posted
>>
>>101699477
>>101699454
>>101699440
What
>>
>>101696472
It was already over with SD2
>>
>>101697479
There's no easy way. It's going to take a lot of trial and error.
>>
>>101699497
my first attempt to reply had an accidental extra word that made the sentence confusing, so I nuked it and reposted a corrected version
the second poster is running a tampermonkey script that allows him to see deleted posts, so to him it appeared as if I double posted, and he is being a fag by reposting what I deleted
>>
File: ComfyUI_temp_riblx_00111_.jpg (3.2 MB, 1792x2304)
3.2 MB
3.2 MB JPG
>>
File: Flux_00103_.png (1.21 MB, 1024x1024)
1.21 MB
1.21 MB PNG
>>
File: ComfyUI_00126_.png (1.68 MB, 1024x1024)
1.68 MB
1.68 MB PNG
>>
>>101699529
I guess even Flux has limits
>>
>>101699570
The book burning going on beneath the swastika was a nice touch.
>>
File: 3.jpg (390 KB, 1344x1344)
390 KB
390 KB JPG
Sleep
Eludes
Me
>>
File: ComfyUI_00231_.png (2.43 MB, 1024x1600)
2.43 MB
2.43 MB PNG
>>
File: tall4.jpg (314 KB, 1624x1120)
314 KB
314 KB JPG
What's the current best SD3 finetune?
>>
>>101699642
there aren't any
>>
>>101699642
Haven't you heard that SD3 is basically useless trash yet?
>>
>>101699649
Theres a bunch
>>
>>101699657
where? they're not on civitai, are they hiding away on huggingface or something
>>
File: ComfyUI_00331_.png (1.76 MB, 1024x1600)
1.76 MB
1.76 MB PNG
>>
What downloaders with resume work with huggingface?
>>
File: Flux_00121_.png (623 KB, 1024x768)
623 KB
623 KB PNG
>>101699657
>>101699642
>>101699679
>>
>>101699744
axel
>>
File: FD_00380_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>
>>101699405
Awesome, love the repeated texture on those garage doors, looks so legit
>>
File: Pixart_ComfyUI_00102_.png (929 KB, 1024x1024)
929 KB
929 KB PNG
>>101699579
A little bit of inpainting should fix this. Where are inpainting scripts? Where is TensorRT? Any place to keep track of updates/ETA?
>>
>>101699642
Catbox?
>>
>>101699867
https://litter.catbox.moe/dmqu3c.jpg
>>
File: 00103-2979083462.jpg (236 KB, 1552x1200)
236 KB
236 KB JPG
>>101699405
This is dope as hell
>>
>>101699866
is this bunline?
>>
>>101699935
Nah, base model
>>
>>101699960
oh wtf that's pretty impressive, base is usually really bad with anime
>>
File: FD_00443_.png (1.13 MB, 1024x1024)
1.13 MB
1.13 MB PNG
he has a moustache because he is undercover.
>>
>>101699762
Nevermind, I found out the problem was Chrome removing the download link address. Have to use another browser.
>>
hello /g/ would you like to know things about flux

there aren't many i can share mind u
>>
File: ComfyUI_Flux_1269.jpg (125 KB, 1024x1024)
125 KB
125 KB JPG
>>
>>101700810
Is it trainable?
>>
>>101700829
>>101700810
Also who are you?
>>
>>101700829
-dev ought to be, you can definitely train a LoRA on it at least, idk how full scale finetuning would go but it should be fine

>>101700837
one of the people who made flux*

*i mostly keep the GPUs going brr but I moonlight as a researcher and sometimes have good ideas
>>
File: ComfyUI_Flux_1275.jpg (165 KB, 1024x1024)
165 KB
165 KB JPG
>>101700810
sure go ahead
>>
>>101700136
>seƱor robot, no robocop here
>>
>>101700861
write me a story about flux
>>
>>101700861
should people use caption dropout with lora training and finetune?
>>
File: 2024-08-02_00294_.png (2.42 MB, 1920x1080)
2.42 MB
2.42 MB PNG
>>101700810
what data set is it trained on? if you cant say, atleast what resolutions were used? I saw discrepancy in quality of hires gens depending on subject matter

Whats the compability issues with samplers? some just dont seem to work or produce horrible output
>>
File: ComfyUI_temp_kgudx_00012_.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>
>>101700909
tell em
>>
File: 197xn.jpg (125 KB, 1024x1024)
125 KB
125 KB JPG
>>
File: 2024-08-02_00167_.png (1.94 MB, 1280x1280)
1.94 MB
1.94 MB PNG
>>101700918
not sure yet if larp or real.. but atleast I can ask something
>>
>>101700810
What's the business model? How many jews are involved? What's the NSFW policy?
>>
>>101700939
>>
>>
>>101700810
Plans for controlnet? IPAdapter? TensorRT?
>>
File: 2024-08-02_00052_.png (1.05 MB, 1024x1024)
1.05 MB
1.05 MB PNG
>>101700939
bro you can read that up on their webpage: no nasty things with their models per tos (not like they can enforce that in apache 2.0 version tho) and also they are selling API access to pro model, maybe they have hidden plans to make coorperations on personlized models with big corpos/media but they would never say so here
>>
>>101700810
Flux lacks good stylization. Was it a conscious decision (not to antagonise artist) or is it a result of the training?
>>
>>101700927
prompt please
>>
>>101700861
How many backflips can you do.
>>
>>101699236
this looks sick
>>
File: ComfyUI_Flux_1301.jpg (144 KB, 1024x1024)
144 KB
144 KB JPG
>>
>>101701046
how'd you get that aesthetic? My humans always look like they have studio professional photos, even the selfies
>>
File: 1267262280626868329_1..jpg (313 KB, 1360x768)
313 KB
313 KB JPG
>>101700883
do i look like nous-hermes to you

>>101700892
probably, all the usual training/dataset tricks from other t2i models should be just as applicable here

>>101700909
Can't answer any questions around dataset or training resolutions because I literally don't know

as for the compatibility issue, some samplers just don't work properly with things that aren't eps-prediction or use an unusual noise schedule. I have the same issues with some stuff just Not Working with wdV (based on cosXL)

>>101700939
>What's the business model?
see image
>How many jews are involved?
bait harder
>What's the NSFW policy?
in what sense? our API for Pro has various safety filters (and an adjustable safety level!) but we don't control what people do with the models after we release them

>>101700990
>ControlNet, IP-Adapter, TensorRT
No plans to announce on these for the time being. You could easily export it to ONNX / TensorRT yourself, though, but you won't get meaningfully better performance than you'd get just doing torch.compile
>>
File: ComfyUI_Flux_1291.jpg (132 KB, 1024x1024)
132 KB
132 KB JPG
>>101701089
https://x.com/skirano/status/1819444885943914688
>>
>>101701099
where training code
>>
>>101701099
>but we don't control what people do with the models after we release them
godlike stance! lets gooooooo!
>>
>>101701099
>in what sense?
obviously the question is whether you're training on cunny
>Can't answer any questions around dataset
I guess that settles it
>>
>>101700810
Why can't it draw nipples
>>
File: image (5).jpg (202 KB, 1024x768)
202 KB
202 KB JPG
>>101699236
In my experience flux doesnt like the short booru-like only tag prompting.
I changed it to ((c) sonnet 3.5 ) and got better result
>Hyper-detailed screenshot of DOOM (1993-1994) ported to PS1 and N64, side view of urban scene, 3D low-poly graphics with pixelated textures, CRT screen effect, first-person shooter perspective with visible HUD and weapon. Foreground: '80s sedan parked along sidewalk, rust-covered. Background: row of dilapidated buildings with shuttered windows. Environment: overcast grey sky, gloomy atmosphere. Style: high-resolution pixel art, early 3D polygon aesthetics, Phong shading, procedural textures. Key elements: authentic DOOM UI, pixelated gore, retro FPS charm. Additional details: scanlines, color bleeding, dithering effects typical of '90s console graphics.
Best I could manage.
>>
File: image (6).jpg (211 KB, 1024x768)
211 KB
211 KB JPG
>>101701153
In comparison with the original prompt. Was the best out of 4.
>>
File: FD_00534_.png (868 KB, 1024x1024)
868 KB
868 KB PNG
>>101701101
Thanks
>>
Another one ready to roll...
>>101701058
>>101701058
>>101701058
>>
>>101701099
>as for the compatibility issue, some samplers just don't work properly with things that aren't eps-prediction or use an unusual noise schedule. I have the same issues with some stuff just Not Working with wdV (based on cosXL)
thanks that was very enlightening

another one if you are still around:

What are your plans on an apache2.0 version of dev? Will only schnell receive foss releases, or will we get a dev release on an open license to?
>>
>>101701168
MODS
>>
>>101701152
obviously to sell pro
>>
>>101701181
pro can't draw nipples either, I tried
>>
>>101701099
Can you please tell the folks at Replicate to add the ability to manually change resolution when generating images? They don't let us do more than 1MP with any of the FLUX.1 models right now, and we can't control step count for dev.
>>
>>101701186
even on safety tolerance 5? then thats pretty pointless to even have that slider
>>
>>101701205
i think safety tolerance is about the nsfw checker
>>
cough
>>
>>101701578
new bake, anon >>101701168
>>
>>101701596
oh i didn't even notice, ty
>>
>>101698213
>parameter
What else could it be?



[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.