[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 (1.21 MB, 3264x3264)
1.21 MB
1.21 MB JPG
Discussion of free and open source text-to-image models

Previous /ldg/ bread : >>101784055

>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

>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

>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

>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: 2024-08-08_00540_.png (1.8 MB, 1024x1280)
1.8 MB
1.8 MB PNG
>>101787894
thank you baker
>>
File: ComfyUI_00155_.png (1.26 MB, 1024x1024)
1.26 MB
1.26 MB PNG
>>
File: delux_flebo_00003_.jpg (300 KB, 768x768)
300 KB
300 KB JPG
>mfw
>>
>>101787881

Yeah I've seen it, though you can't do that with the image model, only vae and clip.
And I wish you could simply pool the memory instead and use 2x24 or 3x24GB...
>>
File: ComfyUI_00206_.png (1.36 MB, 1024x1024)
1.36 MB
1.36 MB PNG
>>
>>101787966
>Yeah I've seen it, though you can't do that with the image model, only vae and clip.
you can on this fork of the script, found it on the comments
https://www.reddit.com/r/StableDiffusion/comments/1el79h3/comment/lgtnw8m/?utm_source=share&utm_medium=web2x&context=3
>>
which model makes the best feet?
>>
>>101787998
local? SDXL+feet lora or FLUX when asking for feet .. depends how goot the feet lora is
>>
>>101787992
Oh, maybe useful for sdxl model loading at the same time, thanks anon.
>>
File: 11.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>
>>101788016
loras reduce flexibility on what you can gen and still have good feet
the best foot model is DALL-E 3 and this alone makes it the best model of them all
>>
File: ComfyUI_00182_.png (912 KB, 1024x1024)
912 KB
912 KB PNG
>>
File: Capture.jpg (1.1 MB, 2654x3321)
1.1 MB
1.1 MB JPG
>>101787853
>>even on high resolutions like 2048x2048
>How much vram does that eat?
Miku has the answer
>>
File: 2024-08-08_00519_.png (1.7 MB, 1024x1280)
1.7 MB
1.7 MB PNG
>>101788039
delish gen
>>
File: 2024-08-08_00005_.png (1.31 MB, 1280x720)
1.31 MB
1.31 MB PNG
>>
File: 13.png (964 KB, 1024x1024)
964 KB
964 KB PNG
>>101788058
Same to yours, the heart looks incredible
>>
File: ComfyUI_00120_.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>>
>>101787998
Flux by far. See
https://desuarchive.org/g/thread/101689050/#q101693628
And https://desuarchive.org/g/thread/101694073/#101694586
(I made several posts on those threads)
>>
File: ComfyUI_01718_.png (949 KB, 1024x1024)
949 KB
949 KB PNG
>>101788024
>>
File: 14.png (1.09 MB, 1024x1024)
1.09 MB
1.09 MB PNG
>>
File: Capture.jpg (59 KB, 650x665)
59 KB
59 KB JPG
>>101788068
the lightning is a bit saturated, do you use the new settings on DynamicThresholdingFull?
>>
>>101788051
Thanks miku, so that would mean 36GB for fp16 if it scales linearly, which would be nice if only 2x3090/4090 was possible.
>>
>>101788095
not quite DALL-E 3 level, also those backwards knees
>>
File: file.png (2.4 MB, 1152x896)
2.4 MB
2.4 MB PNG
>>
>>101788117
the problem is that you can't separate the image model into two gpu's yet
>>
>>101788115
Also using Euler, and Simple, in the sampler.
>>
>>101788048
>best foot model is dalle
I'm the only footfag who has tested and pushed Flux to its limits. See >>101788095
I'm not too impressed with Dalle since it's censored, but Flux gives me hope we'll get there.
>>
>>101788149
yeah, those are the old settings, I found better with that, use it -> >>101788115
And beta is the better scheduler desu
>>
>>101788129
Yes, which is weird to me since I can do that with text models.
I guess the need for images wasn't there outside of gigantic models like DALLE that have their stuff behind closed doors.
>>
>>101788165
What's the scheduler even do?
>>
>>101788153
I've tried using a slightly uncensored dalle and most of the time it gave me ugly/dirty feet.
None of these beautiful smooth feet, even less with cum on them obviously.
>>
File: ComfyUI_00038_.png (1.01 MB, 1024x1024)
1.01 MB
1.01 MB PNG
is there been any sucessful flux loras besides the realism one.

Like one that is either style or character based lora.
>>
>>101788123
Dalle was never perfect with feet anon. It was slightly better on average sure, but not special in this case. With good gens you can get Dalle level with Flux...
>>
File: ComfyUI_00138_.png (953 KB, 1024x1024)
953 KB
953 KB PNG
>>
>>101788204
nice
>>
File: Capture.jpg (171 KB, 1912x1466)
171 KB
171 KB JPG
>>101788180
https://stable-diffusion-art.com/samplers/
Basically a scheduler is something that will chose how much noise it's gonna be for the next step
>>
>>101788176
yeah, I'm pretty sure it will be implemented at some point now that we have a huge model, it wasn't really needed before with little toys like SD1.5 (0.75b) and SDXL (3.5b)
>>
>>101788216
Is the noise the same noise, or is it also different noise?
>>
File: 15.png (1.44 MB, 1024x1024)
1.44 MB
1.44 MB PNG
>>
is there any free ai video makers similar to klingai and luma ai. One that can could be local or online but allows you to make unlimited videos.
>>
>>101788251
it has to be the same noise I guess, if it was random then we wouldn't have deterministic outputs, and that would be uncool as fuck
>>
File: file.png (2.1 MB, 1152x896)
2.1 MB
2.1 MB PNG
>>
>>101788260
SVD but it is even worse than Luma, no text conditioning, you just gen and pray you get something worthwhile
>>
>>101788196
the great shrekening is upon us...
>>
>>101788216
Karras makes the image oversatured white.
Weird.
>>
>>101788260
>>101788326
there's CogVideoX-2B now
https://github.com/THUDM/CogVideo
>>
>>101788326
yeah, i remember using SVD awhile back. Has there been any improvements from when it was first released?

Just like you mentioned, I would hope for the best random generating any images i put but it was like a 50/50 chance of getting something worthwhile.

Has that been the only open source ai video or are there other ones?
>>
>>101788359
Karras works at really high steps, like 50, I hate that one, you can try it out it gives really unique outputs though
>>
File: 18.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>
File: 2024-08-08_00006_.png (1.02 MB, 1280x720)
1.02 MB
1.02 MB PNG
>>101788280
I mean vs each other.
>>
>>101788397
Oh that's why.
OK.
>>
>>101788414
>I mean vs each other.
I don't know that much about schedulers, what I know is that beta is the best of all of them in my testing kek

Btw, nice gen, the new settings really makes the picture less fried, now we can go on even higher CFGs
>>
>>101788397
>it gives really unique outputs though
Then samplers DO change it up.
>>
>>101788397
ya.. but DDIM works at high CFG .. while it shits your bed at low CFG
>>
>>101788365
thanks. From your experience with this tool would you say its worth trying out or is it a hit or miss with generation
>>
File: ComfyUI_01721_.png (1.59 MB, 1024x1024)
1.59 MB
1.59 MB PNG
>>
>>101788465
never tested it, but it looks fine from what I've seen on leddit
https://reddit.com/r/StableDiffusion/comments/1el2eib/cogvideox2b_results/
>>
File: 1698281143209347.jpg (264 KB, 1024x1024)
264 KB
264 KB JPG
How did they manage to make the model so stupidly good at text
>>
>>101788497
Guess: Use OCR and/or a vision model with good text understanding to caption every image in the pre-training dataset.
>>
>>101788534
So they injected an ungodly amount of OCRed text. Makes sense.
>>
File: ComfyUI_01722_.png (1.48 MB, 1024x1024)
1.48 MB
1.48 MB PNG
>>101788497
I don't know, but it's easily the best at text, even better than dalle3, it just works!
>>
File: Flux_00346_.png (1.3 MB, 1024x1024)
1.3 MB
1.3 MB PNG
>>101788497
>>
File: tmp1fs93t85.png (1 MB, 1024x1024)
1 MB
1 MB PNG
>>
>>101788551
Doubt it. Text can be in widely different styles. Best vision models can read text pretty well.
>>
whats the difference between this thread and the other stable diffusion thread? im trying to make porn locally with ponyxl, is this the right spot?
>>
>>101788611
>>>/h/hdg
>>>/b/degen
>>
File: 2024-08-08_00557_.png (1.85 MB, 1024x1280)
1.85 MB
1.85 MB PNG
>>101788497
easy answer 12b lok at text
>>
File: Flux_00471_.png (1.36 MB, 1024x1024)
1.36 MB
1.36 MB PNG
>>101788611
This Thread is where the Flux chads hang out.
the other thread is for the SD homos.
You have 10 seconds to fuck off with your pony shit before Rape man spawns inside your house.
>>
The flux devs knew what they're doing with this model, they knew that Miku was the mascot of /sdg/, /ldg/ and /lmg/, they made sure she would be rendered perfectly. I wish it would be the same for other animes characters though.
>>
>>101788635
Pony is local, rape man, nothing you can do about it.
>>
>>101788663
go back to /hdg/
>>
>>101788115
>been trying and failing for days to get floox to follow prompt for motion blur, off-center, out of frame shot
>just werks after using these settings with higher cfg
I kneel
>>
>>101788635
>Flux
can this make futa on femboy images?
>>
>>101788727
What did it cost you (in speed)?
>>
>>101788727
>I kneel
it was a shit lot of work, but yeah, I'm glad I finally found the sweet sport for those settings, I had to share it to everyone

>>101788740
activating CFG is always a 2x speed decrease, like on SD models previously
>>
File: lqGKGuT.png (1.89 MB, 1920x1080)
1.89 MB
1.89 MB PNG
Sorry if there is an obvious answer but I'm having trouble finding any info on this so thank you in advance. I'm currently running Debian with an AMD RX 7900 XTX GPU. Would I be able to run ComfyUI without installing ROCm on my distribution? If I understand the process correctly the Python script installs it's own version of ROCm so please correct me if I'm wrong about that. And if it doesn't would you recommend I switch over to a newer distro like Fedora or Mint. Thank you again for you help kind anon.
>>
File: ComfyUI_02480_.png (1.93 MB, 1280x1024)
1.93 MB
1.93 MB PNG
interesting detail: when i lower guidance to really low values, the artstyle i request (in this case oil painting and impressionism) pops out, but i lose fine details like the "harp sign" and extra characters
>>
File: everything.jpg (5 KB, 294x171)
5 KB
5 KB JPG
>>101788740
>>
File: ComfyUI_02501_.png (1.56 MB, 1280x1024)
1.56 MB
1.56 MB PNG
If guidance is high, it sticks to prompt, but ignores the artstyle
>>
>>101788798
yeah, but low guidance also means that it won't follow the prompt as good and the text and hands are getting bad
>>
File: AllHeilNegativePrompt.png (2.46 MB, 3562x1690)
2.46 MB
2.46 MB PNG
>>101788727
you can even use negative prompts with higher cfg now, so if there's something you don't like, just put this shit on
>>
File: Flux_00442_.png (1.07 MB, 1024x1024)
1.07 MB
1.07 MB PNG
>>
File: Flux_00407_.png (1.45 MB, 1024x1024)
1.45 MB
1.45 MB PNG
>>101788856
how come all the flux woman have this Butt chin?
whats causing this?
>>
>>101788635
damn he is hot, how to summon 2?
>>
File: Flux_00486_.png (1.14 MB, 1024x1024)
1.14 MB
1.14 MB PNG
>>
File: Flux_00485_.png (1.27 MB, 1024x1024)
1.27 MB
1.27 MB PNG
>>101788903
>>
File: ugly.png (123 KB, 303x282)
123 KB
123 KB PNG
>>101788856
This is incredibly ugly, image models should stop adding that.
>>
So a bit of backstory, way back in September of 22, I managed to get SD running after a week of figuring things out. Back then, Auto was the only option and since I have an AyyyMD GPU, I had to dual-boot Linux to get things to work. Half the problems I had were from ROCm not playing well with Ubuntu. I had to fix the ROCm installer to get things working. Now that Flux is out, I decided to switch to Comfy, but I'm now getting "HIP" errors when I try to run it with the new ROCm version. I'm not opposed to just nuking my Ubuntu partition and starting from scratch. Any advice on what distro to switch to?
>>
>>101788962
windows
>>
File: Flux_00493_.png (1.32 MB, 1024x1024)
1.32 MB
1.32 MB PNG
>>
>>101788933
older people effect?
>>
>>101789125
kek
>>
blessed thread of frenship
>>
File: 00001-2024-05-17-.jpg (2.51 MB, 2048x2480)
2.51 MB
2.51 MB JPG
>>101787954
Not enough attention retard?
How can you still struggle making gens with all of these advancements thread schizo?
>>
>>101789185
which one is the cursed thread of hatred?
>>
File: Flux_00494_.png (1.26 MB, 1024x1024)
1.26 MB
1.26 MB PNG
>>101789125
>>
File: Flux_00495_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>
>>
File: ComfyUI_00036_.png (1.17 MB, 1024x1024)
1.17 MB
1.17 MB PNG
>>101789281
>>
File: Flux_00242_.png (1.34 MB, 1024x1024)
1.34 MB
1.34 MB PNG
>>101789277
>>
File: CO3CE9~1.png (1.38 MB, 1024x1024)
1.38 MB
1.38 MB PNG
>>101789311
>>
>>101789211
I'm 40 and I don't have that particularly ugly wrinkle.
>>
>>101788153
>Multiple posts, none actually showing the prompt that will give me some nice feet

fuck you
>>
File: Flux_00230_.png (1.22 MB, 1024x1024)
1.22 MB
1.22 MB PNG
>>101789341
>1080ti
when will you upgrade bro?
>>
>>101789356
>>101789341
it's even worse he has a 108Ti
>>
File: ComfyUI-Flux1__00010_.png (1 MB, 800x1024)
1 MB
1 MB PNG
>>101789356
when you get a new video card and give me your old one
>>
>>101789363
is this flux?
>>
File: file.png (1.47 MB, 1152x896)
1.47 MB
1.47 MB PNG
>>
>>101789383
don't reply to pedos
>>
>>101789363
die in a fire
>>
File: Flux_00497_.png (1.08 MB, 1024x1024)
1.08 MB
1.08 MB PNG
>>101789371
kek
>>101789374
I have a busted 3090 that I still havent send to repairs.
>>
>>101789393
I'll reply to whomever I want, grandma.
>>
File: Flux_00462_.png (1.46 MB, 1024x1024)
1.46 MB
1.46 MB PNG
>>101789385
RapeMan would agree
>>
>>101788882
>how come all the flux woman have this Butt chin?
>whats causing this?
Idk but I've wasted my time trying to get rid of that with a negative prompt, without much success
>>
File: Sigma_12508_.png (3.14 MB, 2048x2048)
3.14 MB
3.14 MB PNG
>>
>>101789412
This guy needs to be a thread staple
>>
not sure what happened here but it's kinda cool
>>
File: CatJak_00035_.png (1.95 MB, 1024x1024)
1.95 MB
1.95 MB PNG
I want to make things look less cooked I might need to mess with the dynamic cfg more or something
>>
File: file.png (1.52 MB, 1152x896)
1.52 MB
1.52 MB PNG
>>101789412
>>
File: 1723009242804038.png (1.19 MB, 1024x1024)
1.19 MB
1.19 MB PNG
>>
File: CatJak_00036_.png (2.06 MB, 1024x1024)
2.06 MB
2.06 MB PNG
Some progress
>>
File: prompt_in_file_name.png (41 KB, 892x360)
41 KB
41 KB PNG
if you wanna be cool and share your prompts to maybe help others or help yourself remember what worked heres a comfyui method
>>
So now that we know that flux can't be trained, what do we do now? Is it completely over?
>>
>>101789499
It's only over if you let the brain worms get to you
>>
File: Flux_00502_.png (1.3 MB, 1024x1024)
1.3 MB
1.3 MB PNG
>>101789466
>>101789424
>>
>>101789536
Chef's kiss would be a look of glee in his eyes
>>
>>101788861
>>101788903
>>101788919
How did you make them realistic? Just adding Miku to the prompt immediately makes an image anime, adding stuff like photo, realistic etc doesn't help
>>
give it to me straight (or gay) guys, will Flux EVER be capable of full NSFW?
>>
>>101789576
He's using the negative prompt trick
>>
>>101789424
Enjoy your new avatarfag
>>
File: file.png (2.18 MB, 1152x896)
2.18 MB
2.18 MB PNG
>>101789536
>>
File: Flux_00441_.png (1.16 MB, 1024x1024)
1.16 MB
1.16 MB PNG
>>101789576
>>101789589
you can also try
>photo of a woman dressed like Hatsune Miku
>>
File: Flux_00505_.png (1.45 MB, 1024x1024)
1.45 MB
1.45 MB PNG
>>101789618
>>
>>101789452
Man, if only most risqué poses weren't scrambled, I would prompt stuff like this non stop.
>>
File: 20.png (1.97 MB, 1024x1024)
1.97 MB
1.97 MB PNG
>>
>>101789720
>uses a cross for a capital T
nice
>>
File: dwarf.png (665 KB, 1024x1024)
665 KB
665 KB PNG
>>101789731
To make the gen blessed twice
>>
File: file.png (2.2 MB, 1152x896)
2.2 MB
2.2 MB PNG
>>101789651
>>
File: dwarves.png (2.25 MB, 1024x1024)
2.25 MB
2.25 MB PNG
>>
>>101789424
Definitely not, take your avatarfagging to /sdg/.
>>
File: 26.png (1.66 MB, 1024x1024)
1.66 MB
1.66 MB PNG
>>101788058
I'm offering a heart for a heart
>>
>>101789412
you got a good thing going there, just dont ruin it with le spam. just have him pop up every now and then to remind everyone.
>>
File: Stache.png (1.91 MB, 1024x1024)
1.91 MB
1.91 MB PNG
>>
File: CatJak_00041_.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>Try karras
>get muddy mess unless high steps
>go high steps
>instead of cooked too dim
I don't know why he has a watch I didn't add it to my prompt but it looks like my watch
>>
File: 2024-08-08_00520_.png (1.61 MB, 1024x1280)
1.61 MB
1.61 MB PNG
>>101789926
>>
>>101790142
>it looks like my watch
go to /sdg/ aka /soc/ catfagman you can blog there
>>
>>101790169
You should go back to your containment thread, I'm a foundational poster here.
>>
>>101790142
How I imagined Gor as a teen.
>>
File: ComfyUI_00190_.png (1014 KB, 1024x1024)
1014 KB
1014 KB PNG
>>
>>101790183
*poser
>>
>>101790215
*sight*
*unzips*
>>
File: rapeman.jpg (435 KB, 1080x1080)
435 KB
435 KB JPG
>>101789815
>>
>>101790216
Nothing to provide?
Really you should be shouting in the void in your main thread unless you got something to show. If you don't post a image we're done here
>>
File: ComfyUI_Flux_9.png (1.21 MB, 832x1216)
1.21 MB
1.21 MB PNG
>>101789589
>>101789620
Damn, even with the threshold guidance and (anime:2.0) in the negative I still get either flat 2d anime or 3dslop
>>
>>101790250
i doesn't take much to rile you, stay classy ;^)
>>
File: fire.png (2.04 MB, 1024x1024)
2.04 MB
2.04 MB PNG
>>
>>101790273
you're a disgrace debo
>>
File: 00060-jak.png (3.92 MB, 2048x2432)
3.92 MB
3.92 MB PNG
>>101790300
seeing how he's trying to post in this thread I think he desperately needs a community. Too bad nobody wants him and now that he has his own thread it's just too slow for him to be happy. Seen this happen many times on /vg/ he's in the bargaining phase
>>
>>101790268
>(anime:2.0)
Do these brackets even work for flux? In both positive and negative
>>
File: album.png (1.62 MB, 1024x1024)
1.62 MB
1.62 MB PNG
>>
both the threads are the same speed which do I post innnnn
>>
What's currently the best version of pony for general anime proompting?
>>
File: 1706149237671797.png (1.47 MB, 1024x1024)
1.47 MB
1.47 MB PNG
>>
File: tokyo album.png (2.21 MB, 1024x1024)
2.21 MB
2.21 MB PNG
>>
flux nipples when
>>
>>101790028
Rape Man gotta make his presence know tho
>>
>>101790160
>>101790215
>>101790250
nice
>>
File: Flux_00514_.png (1.75 MB, 1024x1024)
1.75 MB
1.75 MB PNG
>>
File: 0.jpg (154 KB, 1024x1024)
154 KB
154 KB JPG
>>
File: Flux_00516_.png (1.71 MB, 1024x1024)
1.71 MB
1.71 MB PNG
>>
File: 1719309699511514.png (1.31 MB, 1024x1024)
1.31 MB
1.31 MB PNG
>>
File: 0.jpg (102 KB, 1024x1024)
102 KB
102 KB JPG
>>
>>101790396
No, it's basically noise.
>>
>>101790691
Have the same but white instead.
>>
File: 0.jpg (266 KB, 1024x1024)
266 KB
266 KB JPG
>>
>>101790699
Cool
>>
File: FD_00007_.png (785 KB, 768x1024)
785 KB
785 KB PNG
>>
>>101790699
could be an album cover
>>
File: heresy.png (1.97 MB, 1024x1024)
1.97 MB
1.97 MB PNG
>>
>>101790513
don't you worry about that son.
>>101790244
ha ha
>>
>>101790831
>>101790884
Thanks . This flux is great, I'm going to have to check it out more.
>>
Challenge: recreate this on Flux without concept bleed (without making the modern solider wear knight armor pieces or the other way around)
>>
File: Flux_00511_.png (1.55 MB, 1024x1024)
1.55 MB
1.55 MB PNG
>>101790918
>Because he's the hero /ldg/ deserves, but not the one it needs right now. So janitors hunt him. Because he can take it. Because he's not our hero. He's a silent guardian, a watchful protector. A rapist.
>>
File: 1710691821322192.png (1.53 MB, 1024x1024)
1.53 MB
1.53 MB PNG
>>
File: fungi.png (2.2 MB, 1024x1024)
2.2 MB
2.2 MB PNG
>>
File: fsc001.jpg (128 KB, 1024x1024)
128 KB
128 KB JPG
>>
>>101790974
cant be done
>>
>>101790974
why would I want to waste my precious gpu on this shite?
>>101791129
: D
>>
>>101790974
If that were the cover of an Isekai manga, I would read it.
>>
>>101791226
That's already the plot of GATE, anon.
>>
>>101791237
Yeah, but Gate sucks. The moment they bring Earth politics and countries from Earth into the Isekai, it ruins the story.
>>
File: FD_00009_.png (1.02 MB, 768x1024)
1.02 MB
1.02 MB PNG
>>
File: fac010.jpg (204 KB, 1024x1024)
204 KB
204 KB JPG
>>
File: 0.jpg (374 KB, 1024x1024)
374 KB
374 KB JPG
>>
File: seg4.jpg (684 KB, 1024x1024)
684 KB
684 KB JPG
>>101790979
kek
>>
File: 48.png (1.96 MB, 1024x1024)
1.96 MB
1.96 MB PNG
>>
File: FD_00071_.png (1.42 MB, 768x1024)
1.42 MB
1.42 MB PNG
>>
File: 1719070080505847.png (1.78 MB, 1024x1024)
1.78 MB
1.78 MB PNG
>>
>>101791725
OH! How did you prompt this style? I have been trying to get this but I can't remember the name of it.
>>
File: 88.png (831 KB, 1024x1024)
831 KB
831 KB PNG
>>101791735
I would guess that >>101791725 is probably Ukiyo-e?
>>
File: 1692318810397980.png (1.75 MB, 1024x1024)
1.75 MB
1.75 MB PNG
>>101791735
ukiyo-e, yeah
>>
File: 89.png (1.88 MB, 1024x1024)
1.88 MB
1.88 MB PNG
>>101791791
nice, cozy samurai pig
>>
>>101791589
>>101791725
https://www.youtube.com/watch?v=7U6L3nSlaWY
>>
>>101788095
Do you have a catbox link for post 101694441? I'm loving the pose.
>>
File: FD_00332_.png (1.67 MB, 1024x1024)
1.67 MB
1.67 MB PNG
>>101791750
>>101791791
>ukiyo-e,
Hmm, I tried the other day and got picrel, not quite what I wanted
May I have your full prompt, please?
>>
File: 91.png (1.46 MB, 1024x1024)
1.46 MB
1.46 MB PNG
>>101791856
no the anon that posted >>101791791 but some models do it better than others, try a combination of this, what model you are using?
>Edo-period, fashion, female-figures, everyday-life, Japanese, nature, portraits, Ukiyo-e, woodblock
>>
File: 1709875332669956.jpg (326 KB, 1536x1536)
326 KB
326 KB JPG
>>101791856
>a weathered and dated ukiyo-e featuring a fat pig in samurai armor sleeping near a dirt mound and a rising sun flag on the background
>>
Can you create a sexy image through sdxl and detail it through flux?
Or the opposite?
Maybe there is a way to combine both
>>
File: The sun.png (2.34 MB, 1024x1024)
2.34 MB
2.34 MB PNG
>>101791941
She is a cutie anon, very nice gen
>>
File: out-2.jpg (88 KB, 1024x1024)
88 KB
88 KB JPG
>>
File: FLUX_00003_.png (1.04 MB, 1024x1024)
1.04 MB
1.04 MB PNG
>>
File: file.png (3.02 MB, 1664x1216)
3.02 MB
3.02 MB PNG
>>101791951
You can pipe an image generated in flux for an advanced composition/good prompting and then resample at a high denoise it using an sdxl-based model.
>>
File: the world.png (2.34 MB, 640x1536)
2.34 MB
2.34 MB PNG
>>
>>101792029
>the slit appearing
nice
>>
>>101792013
o rape man origin story. "rape poisoning".
>>101792029
nice enlargement there. flux does pretty well at an initial resample after an "upscale with model" node. sky is the limit tho
>>
File: the hermit.png (2.09 MB, 640x1536)
2.09 MB
2.09 MB PNG
>>
File: 1721280127686651.png (1.92 MB, 1024x1024)
1.92 MB
1.92 MB PNG
>true origin story of Yasuke, the first black samurai
>>
File: 1718955221070041.png (1.62 MB, 1024x1024)
1.62 MB
1.62 MB PNG
>>
File: FD_00095_.png (1.61 MB, 1024x1024)
1.61 MB
1.61 MB PNG
>>101791919
>>101791941
Got it, thanks Anons.
>>
File: file.png (3.14 MB, 1664x1216)
3.14 MB
3.14 MB PNG
>>101792029
probably a better example of the kind of interesting compositions you can get out of flux that wouldn't be possible in the pony-based model which I'm resampling with lol
>>
File: 1702181304445742.png (1.78 MB, 1024x1024)
1.78 MB
1.78 MB PNG
God damn why does it produces such smug crocodiles
>>
>>101792147
Flux has baked in tarting.
>>
Probally the wrong topic but the best fitting i think and don't want to bump off another thread from the board:
Whats the best tool (local if possible) to swap faces that look good between 2 images?
>>
>>101792147
damn the influencer smug face slop training data jumped to the crocodiles!
>>101792141
lost all of its crispness tho, nah
>>
>>101792188
gimp / photoshop / krita
>>
>>101792141
So you get composition with flux, then noise again at 0.6-0.7, then "lewdify" with sdxl/pony?
>>
>>101792147
In old Russian the word for Crocodile is змeй eхидный which translates to "sarcastic snake"
>>
File: CatJak_00076_.png (2.17 MB, 832x1280)
2.17 MB
2.17 MB PNG
Getting closer but still so far
>>
>>101792264
lol, good texture
>>
>>101792297
I think I'm close to solving the cfg issue with some tinkering
>>
>>101792264
Her hand looks older than her
>>
>>101792315
That's the joys of testing you find something good then explore from there
>>
File: 1722032646340840.png (1.8 MB, 1024x1024)
1.8 MB
1.8 MB PNG
>>
File: FD_00100_.png (2.4 MB, 896x1600)
2.4 MB
2.4 MB PNG
We ukyo-e today boys
>>
>>101792327
kek
>>
File: out-0.jpg (159 KB, 1024x1024)
159 KB
159 KB JPG
>>
>>101792315
:)

just finding out about makeup?
>>
File: out-1nogra.jpg (214 KB, 1024x1024)
214 KB
214 KB JPG
>>
I wish something else would release so there was only one flux general instead of two. There's not even a point to the split anymore, the avatarfags are even coming to ldg
>>
>>101792378
lol
I don't want such notions in image making ideal characters.
>>
>>101792409
>>101792378
I could have just fucked up
>>
>>101790979
Rape man fucking rules lol
>>
I only have 32gb.

How do I get the model unloaded before the next gen? I have to, because while there is a tiled vae, there is no tiled clip stage (not sure which eats so much ram, the dual clip loader, or the encoders)

For now I have to restart in the manager before each next gen.
>>
>>101792352
I got this ukyo-e lora somewhere, good stuff.
>>101792358
this is great.
>>
>>101792264
Nice
>>
File: 2024-08-08_00008_.png (1.39 MB, 1280x720)
1.39 MB
1.39 MB PNG
This is my take on "Grammar". a whip (discipline) and shepherd's crook (correction).
>>
>>101790835
>>
File: ComfyUI_temp_rrpzr_00033_.png (1.4 MB, 1024x1144)
1.4 MB
1.4 MB PNG
using my new nsfw detection node for evil (it's intended) purposes
>>
File: Capture.jpg (26 KB, 480x863)
26 KB
26 KB JPG
>>101790268
you can increase the negative guidance value to make the negative stronger than the positive
>>
File: ComfyUI_temp_rrpzr_00039_.png (1.41 MB, 1024x1144)
1.41 MB
1.41 MB PNG
>>
>>101792647
what do you use for detection?
>>
>>101792676
I should've looked at the text instead before posting
>>
Hey I got some pubes
Pubes are in the data set.
https://files.catbox.moe/r4bwnn.png
>>
File: ComfyUI_temp_rrpzr_00040_.png (1.42 MB, 1024x1144)
1.42 MB
1.42 MB PNG
>>101792684
I am retarded and don't know how to do it better so I bruteforced it, I took this thing:
>github.com/padmalcom/nsfwrecog
and wrote my own node, it didn't work for me right out of the gate with impact pack stuff
>>
File: out-3.jpg (167 KB, 1344x768)
167 KB
167 KB JPG
>>
File: a.jpg (1.22 MB, 3223x3051)
1.22 MB
1.22 MB JPG
Wtf I think I discovered something interesting by accident, even if you put nothing in the negative prompt, if you increase the Guidance Negative, the model will respect the style more and won't overcook Miku style to the model!
>>
>>101792737
Ah, I thought you used the recognition nodes.
Also lol at the git example, her face is nsfw?
>>
File: 1722226405194495.jpg (101 KB, 954x882)
101 KB
101 KB JPG
>>101792754
It's incredible how good flux is at small stuff.
>>
File: ComfyUI_temp_rrpzr_00048_.png (1.35 MB, 1024x1144)
1.35 MB
1.35 MB PNG
>>101792773
I dunno, the thing detects faces too. the classes are:
EXPOSED_ANUS
EXPOSED_ARMPITS
COVERED_BELLY
EXPOSED_BELLY
COVERED_BUTTOCKS
EXPOSED_BUTTOCKS
FACE_F
FACE_M
COVERED_FEET
EXPOSED_FEET
COVERED_BREAST_F
EXPOSED_BREAST_F
COVERED_GENITALIA_F
EXPOSED_GENITALIA_F
EXPOSED_BREAST_M
EXPOSED_GENITALIA_M
EXPOSED_BACK_F
EXPOSED_BACK_M
DILDO
>>
>>101792787
is it because of their VAE? Do we know if flux's vae is a 16ch one?
>>
>>101792807
I see, so not exactly nsfw, though most of it is.
>>
>>101792807
Is that image a pony to flux workflow? Or did you just inpaint the bubble speech?
>>
>>101792754
try it multiple times with different prompts.
>>
>>101792833
give me a prompt anon
>>
File: ComfyUI_temp_rrpzr_00051_.png (1.37 MB, 1024x1144)
1.37 MB
1.37 MB PNG
>>101792828
it indeed is. I'm that goblin anon that posted here today and yesterday
>>
>>101792807
maybe they meant to detect ahegao type faces? lol
>>
>>101792868
nah it detects any face as far as I can tell
>>
>>101792855
can you show the uncensored version pls?
>>
File: FD_00117_.png (902 KB, 1024x768)
902 KB
902 KB PNG
>>101792868
Flux can't into ahego
It doesn't know how to cross the eyes
:(
>>
File: ComfyUI_01751_.png (1.9 MB, 1024x1024)
1.9 MB
1.9 MB PNG
>>101792833
This is insane!
>A 1700s painting of a portrait of Hatsune Miku
https://imgsli.com/Mjg1Nzcz
>>
>>101792928
https://files.catbox.moe/26jghj.png
>>
File: file.png (2.34 MB, 1024x1024)
2.34 MB
2.34 MB PNG
>>
>>101792948
workflow?
>>
>>101792956
very nice titties
good work and thanks.
>>
>>101792968
https://files.catbox.moe/xlxd00.png
>>
>>101792998
thanks anon
>>
Come and get that loaf while it's hot...

>>101792978
>>101792978
>>101792978
>>
>>101793018
I'm wondering, how come there is always a new thread early?
>>
>>101793038
he expected 5 more images when he put in the application for a new thread
>>
Should we just ignore all non-collage bakes?
>>
>>101793038
Because debo is baking and he's a fucking idiot.
>>
>>101793241
This
>>
>>101793233
i think this is a good move
>>
>>101793233
>>101793292
If there is no collage and the first post is >mfw and posted before the bake is linked in the old thread let the thread autosage.
>>
>>101793334
why the random rules
a thread is a thread
>>
>>101792676
Do a hassidic woman.
>>
By all means anon bake a new one
>>
>>101793387
No. I don't want to shit up the boards and get our gen banned. We can use up the current one and sage the next debo bake.
>>
>>101793416
No one will care. You weren't there the last time there were more than two active image generation threads.
>>
I will bake if it's really that bad
Just tell me to do it
>>
>>101793449
no
>>
>>101793437
I was here when the threads were moving so fast we had a new bake every 20 minutes.
Don't cite ancient magic to me boy.
>>
>>101793475
So then you'd know another bake wouldn't get deleted by jannies
>>
>>101793486
Sorry, I am not a massive faggot like you, I don't like shitting up the board.
All you need to do is literally fuck off back to your own thread.
>>
bump
>>
>>101792868
>>101792935
try: Right eye looks to the left.
Left eye looks to the right.
>>
>>101792095
neat
>>
Thanks, will check out
>>
>>101796505
You're welcome, all me btw
>>
bump
>>
Right
>>
>>101796560
Left
>>
>>101796586
Right
>>
You keep forgetting to bump, anon.
>>
Well ok
>>
>>101796772
Left
>>
But why
>>
Its funny
>>
Sometimes it is
>>
It
>>
ok
>>
File: a.jpg (35 KB, 1087x118)
35 KB
35 KB JPG
>>101788051
laughs in fp16



[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.