/ldg/ - Local Diffusion General
Anonymous 01/20/25(Mon)12:12:28 | 315 comments | 138 images | 🔒 Locked
Come back in a year, brother
Then What Liquid Is It Edition

Discussion of Free and Open-Source Diffusion models.

Last Bread: >>103957379

Newfrien Guide: https://rentry.org/ldg-lazy-getting-started-guide

>UI
Metastable: https://metastable.studio
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI
Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge
reForge: https://github.com/Panchovix/stable-diffusion-webui-reForge
ComfyUI: https://github.com/comfyanonymous/ComfyUI
InvokeAI: https://github.com/invoke-ai/InvokeAI

>Models, LoRAs, & Upscalers
https://civitai.com
https://tensor.art/
https://openmodeldb.info

>Training
https://github.com/Nerogar/OneTrainer
https://github.com/kohya-ss/sd-scripts/tree/sd3
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts

>HunyuanVideo
Comfy: https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/
Windows: https://rentry.org/crhcqq54
Training: https://github.com/kohya-ss/musubi-tuner | https://github.com/tdrussell/diffusion-pipe

>Flux
Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050
ComfyUI: https://comfyanonymous.github.io/ComfyUI_examples/flux
DeDistilled: https://huggingface.co/TheYuriLover/flux-dev-de-distill-GGUF/tree/main
Pixelwave: https://civitai.com/models/141592/pixelwave

>Misc
Share Metadata: https://catbox.moe | https://litterbox.catbox.moe/
Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two
Archive: https://rentry.org/sdg-link
Samplers: https://stable-diffusion-art.com/samplers/
Digital Art Software: https://krita.org/en/
Txt2Img Plugin: https://kritaaidiffusion.com/
Collagebaker: https://www.befunky.com/create/collage/
Video Collagebaker: https://kdenlive.org/en/

>Related
>>>/aco/csdg
>>>/b/degen
>>>/c/kdg
>>>/d/ddg
>>>/e/edg
>>>/h/hdg
>>>/trash/slop
>>>/vt/vtai

>Local Sexting
>>>/g/lmg

>Maintain thread quality
https://rentry.org/debo
Anonymous 01/20/25(Mon)12:14:21 No.103969409
1713315254721390
>>103969385
>Come back in a year, brother.jpg
Anonymous 01/20/25(Mon)12:34:06 No.103969617
>>103959774
The main difference right now is down to model support but the thing that should be compared is at what point upping the rel_l1_tresh and residual_diff_threshold for each method do you get lossless output each caching method respectively with FLUX or some other model, no quantization, if that is even possible and then compare both speedup methods with that. Right now, people are just using settings willy nilly to compare based on feel when you can do an objective measurement. I feel like I am going to have to do this myself when I am done with a work project this week because people for some reason can't actually conceive of a good way to do image comparisons for this.
Anonymous 01/20/25(Mon)12:36:58 No.103969650
>>103969617
please do anon, I'm also havng a hard time o choosing between Teacache and First Block Cache, I feel they are really similar but there's surely a consistant winner out of this
Anonymous 01/20/25(Mon)12:39:33 No.103969676
00138-1329573222-aa506693-9671858123
A horizontal aspect ratio also tends to help if you don't want the subject to be walking straight towards the viewer.
Anonymous 01/20/25(Mon)12:45:25 No.103969734
00043-3526221880
Anonymous 01/20/25(Mon)12:47:31 No.103969754
>>103969385
requesting may-li cum image
Anonymous 01/20/25(Mon)12:48:44 No.103969763
bros, how do i escape the 1girl trap? I'd like to gen cool shit but i always end up just genning gigantic booba

send help
Anonymous 01/20/25(Mon)12:52:29 No.103969813
00046-715079202
>>103969763
gen monsters and landscapes
Anonymous 01/20/25(Mon)12:52:52 No.103969817
cm65c0wd900c0336zt3f8nbdd
regularposters are busy at work general

>>103969763
>how do i escape the 1girl trap?
unironically work on a project like a comic book or music video or worldbuilding with landscapes and architecture or something
i was able to reduce the 1girl market share in my mind (temporarily) by up to 20% with this method
just like breaking any addiction, its hard to start but gets easier the more you do it
Anonymous 01/20/25(Mon)13:23:52 No.103970149
>>103969763
seek inspiration, try to convey something, go beyon aesthetics and put some meaning behind your gens

1girling is also okay, in the end what matters is you're having fun, or are satisfied with your work
Anonymous 01/20/25(Mon)13:34:52 No.103970273
01732-2840231297
Anonymous 01/20/25(Mon)13:34:53 No.103970274
Blessed thread of frenship
Anonymous 01/20/25(Mon)14:24:35 No.103970767
>>103969763
Make loras, play with i2i
Anonymous 01/20/25(Mon)14:27:37 No.103970803
0
Anonymous 01/20/25(Mon)14:32:53 No.103970840
1714274769365523
>trying to gen a niche character with hardly any images
>look up if there's a lora
>there is
Goddamn this community is based, forgive the big booba slop, I'm just happy she has a lora at all.
Also I'm impressed with how well I've had illustrious models load pony loras
Anonymous 01/20/25(Mon)15:51:34 No.103971501
>Flex.1 alpha is a pre-trained base 8 billion parameter rectified flow transformer capable of generating images from text descriptions. It has a similar architecture to FLUX.1-dev, but with fewer double transformer blocks (8 vs 19). It began as a finetune of FLUX.1-schnell which allows the model to retain the Apache 2.0 license. A guidance embedder has been trained for it so that it no longer requires CFG to generate images.
>https://huggingface.co/ostris/Flex.1-alpha
Anonymous 01/20/25(Mon)15:51:37 No.103971502
1723961078554087
Anonymous 01/20/25(Mon)15:55:39 No.103971548
>>103970840
still bestiality, but very nice actually. be glad you got lucky with that lora! looks very good.
Anonymous 01/20/25(Mon)15:56:24 No.103971564
>>103970840
>illustrious models load pony loras
The what now? Never thought to try.
Anonymous 01/20/25(Mon)16:04:38 No.103971657
1710841644599624
Anonymous 01/20/25(Mon)16:05:22 No.103971660
>>103967130

I think this is devoid of any of my custom nodes so it should be open-able with the normal comfyui extensions.

https://files.catbox.moe/h3xs28.png

But basically, the upscale is:
2x upscale, half with 4xFFHQDAT and half with 4x_NMKD_Siax_200k . Then img2img with 0.45 denoising using pixelwave.
Anonymous 01/20/25(Mon)16:07:36 No.103971678
>>103971660
Not that anon but thanks for sharing, your stuff is amazing
Anonymous 01/20/25(Mon)16:10:24 No.103971693
1713313000873828
>>103971564
Just tried it with Selvie lora from unicorn overlord and yep, works surprisingly well
Anonymous 01/20/25(Mon)16:13:03 No.103971723
>>103971657
This Gaze series is pretty cool.
Anonymous 01/20/25(Mon)16:16:52 No.103971767
>>103971660
*stares at workflow* so you are chaining DAT and ESRGAN upscalers, interesting. thank you very much, I am the anon who asked. I was messing around with similar content earlier this week and found that the basic (non R-) ESRGAN upscaler works fairly well too.
>>103971678
it is isnt it
Anonymous 01/20/25(Mon)16:20:48 No.103971797
I have a 4070 with 12GB and I'm trying to get back into genning after upgrading my PC. Woud appreciate some help beyond the newfriend guide. I got comfyUI and automatic to run but I'm a little at a loss about what is considered "the standard" for genning nowadays. It cant be SD1.5 right?
Anonymous 01/20/25(Mon)16:22:04 No.103971809
motherFUCKER windows update while I was out
how do I continue training from the last epoch
Anonymous 01/20/25(Mon)16:22:48 No.103971817
>>103971797
Try the base Illustrious v0.1 first, before diving into its mememerges and such, if you're a 2d enjoyer. If 3dpd then lustify. If 3d but not 1girl then I don't even know.
Anonymous 01/20/25(Mon)16:23:11 No.103971822
Day 3, still can't get Flux models to fun.....I don't want to stop at SDXL
Anonymous 01/20/25(Mon)16:23:32 No.103971827
cargo cult genning
Anonymous 01/20/25(Mon)16:23:41 No.103971830
>>103971797
Stable Diffusion WebUI Forge + Cat Tower checkpoint
Anonymous 01/20/25(Mon)16:24:05 No.103971840
1
Anonymous 01/20/25(Mon)16:25:50 No.103971857
b2dfdadf-60bf-4188-9a08-0351e17080bb_video
>>103969385
>cum girl is in the collage
>none of the rockets are
>/vp/NAPT/ still isn't mentioned anywhere

cool thread
Anonymous 01/20/25(Mon)16:30:07 No.103971903
>>103971797
uninstall a1111
install REforge, listen to this man >>103971817 go from there
>>103971857
op shady news at 11. fuck thats amazing, slowly warming up to the whole animation thing. 10 seconds?
Anonymous 01/20/25(Mon)16:32:23 No.103971931
>>103971678
>>103971767
np
another thing I sometimes do if I want to get more of a grainy look is in GIMP with GMIC installed to use one of its sharpeners at low amplitude, usually inverse diffusion. Actually there's a ton of great post-processing tools in GMIC.
Anonymous 01/20/25(Mon)16:47:01 No.103972078
ComfyUI_07302_
Anonymous 01/20/25(Mon)16:50:29 No.103972117
ComfyUI_temp_trvls_00099_
I am not a doll.
Anonymous 01/20/25(Mon)16:54:56 No.103972165
66fb5703-8f0a-4aaf-b3d5-0a57abbb7ed1_video
>>103971903
kling is free
but i dont recommend them for actual starting image generation
i use comfy\indrema for the anon asking last thread

i cant post from work kek
Anonymous 01/20/25(Mon)16:58:30 No.103972197
>>103971657
fuck that's so good
Anonymous 01/20/25(Mon)17:02:35 No.103972234
Anonymous 01/20/25(Mon)17:02:50 No.103972235
>>103970273
Artist tags? Nice.
Anonymous 01/20/25(Mon)17:05:39 No.103972267
Has anyone done a comparison showing the difference in "quality of output" between AUTO and Comfy? I know they use different token weighting along with GPU vs CPU noise but has anyone tried accounting for both and generating the same image?

I look at some of these AUTO gens and go "fuck, how did he make it look so sharp"...

The last time I tried AUTO again after being on Comfy for awhile, it was so alien I could just not make a decent image without struggling.
Anonymous 01/20/25(Mon)17:09:03 No.103972297
00137-3298318355
Anonymous 01/20/25(Mon)17:11:28 No.103972322
86263472-205098244
a1111, I remember a1111...
Anonymous 01/20/25(Mon)17:14:06 No.103972353
>>103972267
you can (and I tried and managed to a long time ago) create identical output but the devil is in the details. noise source, token weighting and a lot of other things. >>103972322 dad joke
>>103972165
ty. I tried making a kling account the other day but my browser refused to get the security check to work so I gave up lol. brz.
Anonymous 01/20/25(Mon)17:16:02 No.103972371
>>103971857
Why when you post things like >>103972347 ? On second thought, not worth further replies.
Anonymous 01/20/25(Mon)17:19:13 No.103972396
>>103972322
I member training 1.5 embeddings with the GUI. Took like 10 minutes to train had like two tunable parameters and you'd tab over and give it a try as soon as it was done. Now you got these loras with 8000 things to tweak and who knows
Anonymous 01/20/25(Mon)17:19:14 No.103972397
Anonymous 01/20/25(Mon)17:23:29 No.103972435
Carrot sloppa 324
>>103972427
Can you make this move?
Anonymous 01/20/25(Mon)17:24:23 No.103972442
>>103972427
why are you gay?
Anonymous 01/20/25(Mon)17:24:57 No.103972445
what's up with all the SaaS vidgen going on here? This is the LOCAL thread
Anonymous 01/20/25(Mon)17:25:03 No.103972447
>>103972442
hes not a very good envoy thats for sure
Anonymous 01/20/25(Mon)17:29:14 No.103972474
Anonymous 01/20/25(Mon)17:31:34 No.103972495
>>103972427
go back
Anonymous 01/20/25(Mon)17:32:28 No.103972506
ComfyUI_07394_
Anonymous 01/20/25(Mon)17:36:33 No.103972543
Anonymous 01/20/25(Mon)17:36:48 No.103972544
>>103972474
Thanks anon. I haven't tested it yet. Have you?
Anonymous 01/20/25(Mon)17:41:25 No.103972584
>>103972544
Nope. I'll step back into the Flux world once I upgrade or a decent finetune comes out. Whichever comes first.
Anonymous 01/20/25(Mon)17:47:44 No.103972654
>they deleted all the rocket posts
\ldg\ vs \sdg\ rivalry continues
*eats popcorn*
Anonymous 01/20/25(Mon)17:47:46 No.103972657
1711691453088308
Huh, so noob indeed works with pony loras to some extent. Native illustrious has no knowledge of this character.
Anonymous 01/20/25(Mon)17:48:41 No.103972668
classic cameltoe
Anonymous 01/20/25(Mon)17:54:07 No.103972714
>>103971767
>DAT and ESRGAN upscaler
whats the difference? is DAT supposed to be better?
Anonymous 01/20/25(Mon)17:56:35 No.103972733
>>103972654
He shouldn't have posted the nipples
Anonymous 01/20/25(Mon)17:57:56 No.103972749
So what happens if you make something outside of the recommended resolution?
Is the recommended resolution something you should always go with or does it not really matter?
Anonymous 01/20/25(Mon)17:58:52 No.103972755
>>103972749
SD stuff distorts, others don't seem to
Anonymous 01/20/25(Mon)17:59:00 No.103972759
>>103972749
>So what happens if you make something outside of the recommended resolution?
Generally just distorts and warps the image into tiles when you start base genning at high resolutions. That's why we upscale.
Anonymous 01/20/25(Mon)18:00:09 No.103972776
>>103972749
The most apparent consequence is things like borders being off or characters becoming distorted. You don't NEED to use the rec res but you should.
Anonymous 01/20/25(Mon)18:02:15 No.103972802
>15 minutes later
Anonymous 01/20/25(Mon)18:02:45 No.103972808
ComfyUI_temp_trvls_00135_
Anonymous 01/20/25(Mon)18:03:30 No.103972814
/lmg/ > /ldg/ > /lgbt/ > /de3/ > /sdg/
Anonymous 01/20/25(Mon)18:03:32 No.103972815
>>103972755
>>103972759
>>103972776
Ah I see. I've been using Pony since I started and doesn't seem to have that issue. I'm going to try SD though soon so that's good to know.
Anonymous 01/20/25(Mon)18:04:33 No.103972824
>>103972815
anon... pony is SD, no matter how much they say the opposite
Anonymous 01/20/25(Mon)18:04:44 No.103972827
>>103972815
Pony is SDXL. You just haven't gone high resolution enough yet if you haven't experienced distortion.
Anonymous 01/20/25(Mon)18:09:17 No.103972867
0
Anonymous 01/20/25(Mon)18:14:13 No.103972913
me rn
What's the best way to specify different characteristics for different characters in comfy UI? Example one wearing a hat and one with a bow tie.
Anonymous 01/20/25(Mon)18:16:25 No.103972938
tv-static
>>103970803
b e a u t i f u l.
Anonymous 01/20/25(Mon)18:16:36 No.103972940
HyVid_00027
so damn close, closer than any other local model for sure but just not there
Anonymous 01/20/25(Mon)18:18:54 No.103972959
1733424670816
we will never top this
Anonymous 01/20/25(Mon)18:19:00 No.103972962
>>103972940
If I wasn't watching that on loop and staring at it I would have been fooled.
Anonymous 01/20/25(Mon)18:19:17 No.103972964
>>103972938
REEEEEEEEEE
Anonymous 01/20/25(Mon)18:21:27 No.103972989
Anonymous 01/20/25(Mon)18:22:11 No.103972999
00012-4074270250
Anonymous 01/20/25(Mon)18:31:13 No.103973066
oi
>>103972714
bit of a long winded topic.. DAT upscalers were often trained with photography in mind and, from my experience, excel at working with faces. they need quite the processing power. ESRGAN is an older architecture. think of upscalers like models, trained on specific datasets.
Anonymous 01/20/25(Mon)18:34:07 No.103973099
00022-1192089892
Anonymous 01/20/25(Mon)18:34:40 No.103973105
Anonymous 01/20/25(Mon)18:39:34 No.103973161
We don't deserve a new model
Anonymous 01/20/25(Mon)18:41:14 No.103973176
diffusion-pipe now supports Flex.1-alpha finetuning by adding the option to bypass the guidance embedding.
I've only done a very quick preliminary lora test, but it seems to compete favorably with flux dev and flux dev de-destill despite being a smaller model.
I still don't have fp8 / quantization implemented for flux, but this being only 8b parameters means you can train loras on it with a single 24GB GPU, unlike full-size fluxes.
Anonymous 01/20/25(Mon)18:42:35 No.103973196
Okay I've been out of it for a while, but when the fuck did CFG lower down to 1.5-3 values and steps go all the way down to 13?? What changed?
Anonymous 01/20/25(Mon)18:42:50 No.103973200
how do I generate stuff with huanyuan in comfyui?
Anonymous 01/20/25(Mon)18:43:25 No.103973206
>>103973176
Example comparisons?
>but this being only 8b parameters means you can train loras on it with a single 24GB GPU, unlike full-size fluxes.
Neat.
Anonymous 01/20/25(Mon)18:43:41 No.103973208
Anonymous 01/20/25(Mon)18:51:30 No.103973277
>>103971660
>my custom nodes
What kinds did you write?
Anonymous 01/20/25(Mon)18:53:34 No.103973300
>>103972940
it's still impressive. non-geeks won't know the difference. btw, prompt :-)
Anonymous 01/20/25(Mon)18:54:16 No.103973310
https://huggingface.co/tencent/HunyuanVideo/blob/main/hunyuan-video-t2v-720p/transformers/mp_rank_00_model_states_fp8.pt

Can this model do anime?
Anonymous 01/20/25(Mon)18:58:23 No.103973344
00053-258931254
Anonymous 01/20/25(Mon)19:04:05 No.103973391
>>103973196
it shouldntve if your settings are exactly the same
Anonymous 01/20/25(Mon)19:09:01 No.103973428
ComfyUI_temp_trvls_00144_
Anonymous 01/20/25(Mon)19:12:07 No.103973443
>>103973344
Cool colors

>>103973428
Big girl
Anonymous 01/20/25(Mon)19:14:36 No.103973466
>>103969763
Seconding img2img, you get way more interesting stuff.
Anonymous 01/20/25(Mon)19:19:56 No.103973493
1721650253166681
>>103969763
Make assets for your dream game
Anonymous 01/20/25(Mon)19:45:02 No.103973634
00054-crystalClearXLPrime_v10-20240731_130319_852478-1306673122
>>103969763
you need to find the exit. stop genning with your cock. and if you have to, make it efficient. hone your 1girl craft to the point where you can generate supermega 1girl9000 at any time, it'll wear out. go outside, get laid. it's a long process. creativity can be found everywhere and your sex drive is a sneaky bastard. bla
Anonymous 01/20/25(Mon)19:55:03 No.103973721
>>103973634
>just be tall and handsome bro
Anonymous 01/20/25(Mon)19:55:08 No.103973723
>>103969817
>regularposters are busy at work general
cozy hours ultra edition remix
Anonymous 01/20/25(Mon)19:55:41 No.103973729
may li kelly balthazar
>>103969385
Who keeps making these kelly balthazar/mai li images? Whoever it is should upload a zip to catbox because that bitch still makes me diamonds after like 15 years.
Anonymous 01/20/25(Mon)19:58:45 No.103973755
>>103973729
there's a lora of her so you can even make some of your own
Anonymous 01/20/25(Mon)20:02:36 No.103973786
>>103973755
not the same anon. can someone share her hunyuan lora? this is for my neighbor
Anonymous 01/20/25(Mon)20:03:38 No.103973794
S3683742073; ntrMIXIllustriousXL_xiii; DPM++ 2M SDE; 20250118040343;
Anonymous 01/20/25(Mon)20:06:23 No.103973814
>>103970840
I'm not a furry, but...
Anonymous 01/20/25(Mon)20:22:36 No.103973978
Anonymous 01/20/25(Mon)20:41:30 No.103974140
*yawn*
Anonymous 01/20/25(Mon)20:42:14 No.103974147
file
the power of noobai
Anonymous 01/20/25(Mon)20:58:46 No.103974314
>>103974147
You're using the wrong sampler, try euler_a
Anonymous 01/20/25(Mon)21:01:38 No.103974337
>>103969385
Share May Li cum shower pic, please.
Anonymous 01/20/25(Mon)21:07:01 No.103974381
>>103974147
Did it give her a butthole, a pussy, or both there?
Anonymous 01/20/25(Mon)21:07:21 No.103974387
1
So what model is everyone using these days? I don't think I like flux because of how tiny the images are.
Anonymous 01/20/25(Mon)21:08:28 No.103974395
sovl
s-sovl
Anonymous 01/20/25(Mon)21:08:53 No.103974399
>>103972999
nice body
Anonymous 01/20/25(Mon)21:18:45 No.103974478
>>103974147
What's wrong with it?
Anonymous 01/20/25(Mon)21:25:27 No.103974530
sovl3
Anonymous 01/20/25(Mon)21:37:50 No.103974631
z
Anonymous 01/20/25(Mon)21:40:54 No.103974664
2025-01-20_00005_
>>103969813
>gen monsters and landscapes
Anonymous 01/20/25(Mon)21:42:05 No.103974672
>>103974314
Generally speaking, the answer is moar steps. But yeah, there can be issues with the steps resolved by ancestral.
Anonymous 01/20/25(Mon)21:45:46 No.103974704
2025-01-20_00006_
>>103974672
(But I will probably stick to fewer steps)
Anonymous 01/20/25(Mon)21:50:36 No.103974753
ComfyUI_03892_
Anonymous 01/20/25(Mon)21:52:22 No.103974768
00248-2511086760-6dddecd0-9671858123
Anonymous 01/20/25(Mon)21:57:08 No.103974826
Anonymous 01/20/25(Mon)22:01:03 No.103974867
011332
Anonymous 01/20/25(Mon)22:21:29 No.103975069
011340
Anonymous 01/20/25(Mon)22:24:28 No.103975091
Anonymous 01/20/25(Mon)22:29:38 No.103975134
>>103972396
I just had a flux finetune go bad after 4 days of training. This is a special kind of hell we have arrived in.
Anonymous 01/20/25(Mon)22:30:29 No.103975149
e
Anonymous 01/20/25(Mon)22:30:49 No.103975152
2025-01-20_00010_
>>103974867
>painting of a woman sitting on a bed with a large purple flower
Anonymous 01/20/25(Mon)22:30:49 No.103975153
S1251123904; ntrMIXIllustriousXL_xiii; DDIM; 20250114101857;
Hi guys, what is the best way to keep the traditional animation grain and have the final result not look like the pic? every time i add the “quality” prompts it removes the grain, even with the added weight.
model recommendations are welcome
Anonymous 01/20/25(Mon)22:31:52 No.103975169
Anonymous 01/20/25(Mon)22:41:35 No.103975254
Anonymous 01/20/25(Mon)22:43:51 No.103975277
d
Anonymous 01/20/25(Mon)22:47:13 No.103975310
f
Anonymous 01/20/25(Mon)22:49:14 No.103975325
Anonymous 01/20/25(Mon)22:50:15 No.103975333
>>103975277
give it an /ldg/ license plate, and give the helmet a reflection of a 1girl
Anonymous 01/20/25(Mon)22:52:21 No.103975355
>>103975310
>the game that could never be made
>slim waist, white face
Anonymous 01/20/25(Mon)22:52:41 No.103975358
1726155161386366
Anonymous 01/20/25(Mon)22:53:02 No.103975363
j
>>103975333
Can't be bothered with inpainting. I don't think I've inpainted a single picture in almost 2 years.
Anonymous 01/20/25(Mon)22:56:49 No.103975388
i
>>103975355
I'll make it.
Anonymous 01/20/25(Mon)22:57:53 No.103975396
n
Anonymous 01/20/25(Mon)22:59:05 No.103975410
Anonymous 01/20/25(Mon)23:00:22 No.103975418
>>103975396
Add a hobo touching her screen please
Anonymous 01/20/25(Mon)23:01:23 No.103975427
>>103975363
Give me the prompt and I'll have a go at prompting.
Anonymous 01/20/25(Mon)23:02:25 No.103975437
1737359024142636[1]
>>103973729
If I can make a request: use her more recent pictures especially her professional modelling pics rather than her trashy porn pics. They had bad lighting, and she was greasy/sweaty so she comes off overly shiny or greasy in the generated images.
Anonymous 01/20/25(Mon)23:04:09 No.103975457
1733452627024633
Anonymous 01/20/25(Mon)23:07:45 No.103975486
>>103975427
https://files.catbox.moe/0tcbv0.png

>>103975418
You can add it if you want:

https://files.catbox.moe/1vm7ue.png
Anonymous 01/20/25(Mon)23:08:47 No.103975493
Lol
Anonymous 01/20/25(Mon)23:09:06 No.103975494
Anonymous 01/20/25(Mon)23:10:08 No.103975508
Anonymous 01/20/25(Mon)23:10:12 No.103975509
011358
didnt mean to reply
Anonymous 01/20/25(Mon)23:13:21 No.103975542
5126126126
Anonymous 01/20/25(Mon)23:14:55 No.103975566
521512512
Anonymous 01/20/25(Mon)23:15:19 No.103975570
>>103975542
On point.
Anonymous 01/20/25(Mon)23:16:53 No.103975589
51251
>>103975570
Mayhaps.
Anonymous 01/20/25(Mon)23:18:10 No.103975600
you can spam images in sdg
Anonymous 01/20/25(Mon)23:21:29 No.103975642
52151261236
You can.
Anonymous 01/20/25(Mon)23:23:17 No.103975665
guys I'm going to spam my images here because otherwise they'll be ignored
of course I understand these are trite images have been posted for years now so I'll I'm showing is I own a GPU capable of diffusion while also having the IQ of a third-worlder, thinking that because no one is shitting on the street, it must be an opportunity to shit on the street yourself
Anonymous 01/20/25(Mon)23:24:01 No.103975673
>>103975589
https://apnews.com/live/trump-inauguration-updates
>The blossoming relationship between Donald Trump and Elon Musk was on full display throughout Monday’s inauguration ceremonies.

>Musk sat near Trump during the ceremony in the Capitol Rotunda, and he pumped his fists when the president promoted the goal of putting American astronauts on Mars.

>The moment sharply illustrated the unusual partnership the billionaire tech titan, who has lucrative contracts with the federal government, has established with the incoming president.
Anonymous 01/20/25(Mon)23:24:38 No.103975680
imagine if we had a new model
wed be trolling about that or whatever
Anonymous 01/20/25(Mon)23:26:09 No.103975702
5901251
Positively seething.
Anonymous 01/20/25(Mon)23:27:28 No.103975720
2025-01-20_00022_
Loving this hand pose.

>>103975508
she cute

>>103975665
my iq could never compete with einstein-india
Anonymous 01/20/25(Mon)23:28:17 No.103975731
>>103975702
HOLY SHIT ANON I HAVE NEVER SEEN AN IMAGE LIKE THAT BEFORE? ARE YOU SAYING YOU CAN JUST PRESS A BUTTON AND MAKE THE SAME IMAGE OVER AND OVER? AND YOU CAN DO THIS WITH IMPUNITY 24/7 BECAUSE IT IS AN AUTOMATED PROCESS MADE BY SOMEONE NOT YOU, SHOWING THAT YOU ARE DUMBASS NIGGER WHO THINKS USING A CALCULATOR MAKES YOU SMART? LITERALLY ANYONE CAN DO WHAT YOU ARE DOING, WHICH MAKES WHAT YOU DO STUPID.
Keep it on your computer. Come back when you do something that requires more than 20 minutes of effort. That way your spam can be kept controlled.
Anonymous 01/20/25(Mon)23:29:02 No.103975739
>>103975665
>>103975731
You're going about this the wrong way, anon
Anonymous 01/20/25(Mon)23:29:39 No.103975743
>>103975739
No, he'll eventually start spamming because he's a sdg troll.
Anonymous 01/20/25(Mon)23:30:36 No.103975753
>>103975437
where wouldst one find said images? google is failing me
Anonymous 01/20/25(Mon)23:30:53 No.103975757
2
Everything on the internet only exists for you. The world revolves around you. It's all about you, little guy. I'm sorry your mommy never mentioned that, kid.
Anonymous 01/20/25(Mon)23:30:59 No.103975759
>>103975680
lmg has a new model, called Deepseek:
https://www.deepseek.com/

I am trying it out. It's supposedly fully open source, and it's literally going live on Trump-Vance Day 1/20. Wonderful, albeit crazy, day.
Anonymous 01/20/25(Mon)23:31:23 No.103975761
Honestly I'm shocked they don't just spam on Facebook to get Yous from grandmas.
Anonymous 01/20/25(Mon)23:31:55 No.103975763
>>103975759
and we get nothing :(
Anonymous 01/20/25(Mon)23:33:56 No.103975789
Engage the kino poasting
Anonymous 01/20/25(Mon)23:34:24 No.103975791
1448352152
Anonymous 01/20/25(Mon)23:34:42 No.103975796
ComfyUI_07446_
Anonymous 01/20/25(Mon)23:35:53 No.103975813
011367
Anonymous 01/20/25(Mon)23:40:42 No.103975855
Can someone explain to me \(\) ? I see it around series names like a video game or anime. Is it searching online or what?
Anonymous 01/20/25(Mon)23:44:00 No.103975884
Lmao
>>103975855
It provides context. If you type Elizabeth, it might not know who you mean. But if you type Elizabeth/(Bioshock/) it will generate Elizabeth from Bioshock, assuming it's even been trained on pictures of Elizabeth from Bioshock.
Anonymous 01/20/25(Mon)23:44:21 No.103975887
>>103975855
Parentheses around a word in the prompt normally mean to increase the importance of that token. I assume the \( \) syntax is escaping the parentheses so you get actual parentheses in the prompt string instead of having them interpreted that way
Anonymous 01/20/25(Mon)23:45:18 No.103975895
>>103975855
\ is can escape character typically when you want a control character written literally, ( ) are typically used for (((emphasis))) and are a control character. So if you want to literally write (1998) you must do \(1998\)

imagine listening to low IQ underaged sdg spammers
they don't even understand why things are the way they are
nothing is worse than an underage person trying to talk to non underage people
Anonymous 01/20/25(Mon)23:45:45 No.103975899
>>103975884
Does it require a lora or is it searching for trained images in the checkpoint?
Anonymous 01/20/25(Mon)23:46:46 No.103975911
>>103975895
Ahh okay that makes sense.
Anonymous 01/20/25(Mon)23:47:01 No.103975915
41251251
>>103975899
In the checkpoint.
Anonymous 01/20/25(Mon)23:50:10 No.103975942
2025-01-20_00025_
>>103975457
how'd you suppress booba?
Anonymous 01/20/25(Mon)23:51:38 No.103975957
1701261869375909[1]
>>103975753
i have a few saved. She has more professional profile pics and some modeling pics. She's healthier looking despite being older, and it's done under better lighting.
Anonymous 01/20/25(Mon)23:51:52 No.103975961
>>103975763
Hardly seems fair, huh?

>>103975855
It's literally just passing the parentheses to it, ignore the retard.

If you don't escape them, it just adjusts the weights.
Anonymous 01/20/25(Mon)23:52:43 No.103975968
1737274825257[1]
>>103975957
This is a video made based on these images, but I don't have the originals.
Anonymous 01/20/25(Mon)23:53:22 No.103975970
>>103975961
example

(frog). 1girl. woman. blue dress.

this makes frog stronger in weight than the others.

frog \(tv series\). 1girl. woman. blue dress.

would use the tag "frog (tv series)", if there is such a thing.
Anonymous 01/20/25(Mon)23:53:52 No.103975973
>>103975968
mayli...
Anonymous 01/20/25(Mon)23:55:20 No.103975982
Sneed
Whata masterpiece from such a simple prompt: score_9, score_8_up, score_7_up, sneed
Anonymous 01/20/25(Mon)23:57:58 No.103976005
1701261744953752[1]
>>103975753
https://duckduckgo.com/?t=ffab&q=kelly+jiang+wang+art+&ia=web

Go through her more recent photos and videos, and take some pics.
Anonymous 01/21/25(Tue)00:01:18 No.103976024
Anonymous 01/21/25(Tue)00:07:11 No.103976077
Eight more days.
Anonymous 01/21/25(Tue)00:07:46 No.103976084
TND
Anonymous 01/21/25(Tue)00:09:20 No.103976103
2025-01-20_00027_
>>103975982
Pretty similar to a blank prompt, probably.
Anonymous 01/21/25(Tue)00:10:21 No.103976112
>>103976084
he's cute.
Anonymous 01/21/25(Tue)00:11:28 No.103976119
2025-01-20_00028_
>>103976103
blank prompt version...
Anonymous 01/21/25(Tue)00:17:08 No.103976178
2025-01-20_00030_
>>103976119
>
empty prompts are the goodest.
Anonymous 01/21/25(Tue)00:21:44 No.103976222
2025-01-20_00031_
>>103976178
same seed, no lora. I just had a random lora.

loras are a great way to introduce variation, but without a total change like with a seed change. Sometimes that's how it works anyway.
Anonymous 01/21/25(Tue)00:46:29 No.103976462
2025-01-20_00034_
>>103975982
trying other seeds.

Most have been uninteresting, this one is alright.
Anonymous 01/21/25(Tue)00:57:23 No.103976537
Anonymous 01/21/25(Tue)01:05:19 No.103976604
00024-832884187-8ae40a13-9671858123
Anonymous 01/21/25(Tue)01:07:49 No.103976630
2025-01-20_00045_
Drumph, orange hitler, fascist triumphist, lord of the flat Earth, and the sphere space, and the 8 parsecs.
Anonymous 01/21/25(Tue)01:09:44 No.103976647
>>103976537
Some other guy demonstrated that to me when I first got Flux working.
Anonymous 01/21/25(Tue)01:10:47 No.103976652
Anonymous 01/21/25(Tue)01:14:01 No.103976679
Anonymous 01/21/25(Tue)01:49:23 No.103976959
>>103973493
What model/prompt?
Anonymous 01/21/25(Tue)01:50:23 No.103976970
Anonymous 01/21/25(Tue)01:53:46 No.103976994
1721507219409298
>>103976959
Flux

https://files.catbox.moe/qf54i6.png
Anonymous 01/21/25(Tue)02:07:43 No.103977106
Anonymous 01/21/25(Tue)02:08:43 No.103977113
>>103976994
That's pretty elaborate.
Anonymous 01/21/25(Tue)02:14:04 No.103977146
1727457084515960
>>103977113
I've just been taking other peoples workflows and gradually smashing nodes together. The pixel art lora and the pixelate node are the most essential parts outside of the prompt, and within the prompt, emphasizing a plain white background and the specific placement of items seems to do most of the heavy lifting. I've had a lot more success with fantasy stuff probably owing to the sheer volume of fantasy art loras, but more contemporary items can also be convincing.
Anonymous 01/21/25(Tue)02:26:08 No.103977245
Anonymous 01/21/25(Tue)02:57:31 No.103977476
Anonymous 01/21/25(Tue)03:04:55 No.103977539
>>103977476
Love this, catbox?
Anonymous 01/21/25(Tue)03:28:49 No.103977745
huh, hunyuan has native flashing
I dunno why I assume that if there's a lora then the base model can't do it
Anonymous 01/21/25(Tue)03:31:58 No.103977780
ComfyUI_257504_
https://huggingface.co/comfyanonymous/cosmos_7B_fennec_girl_test_lora
Anonymous 01/21/25(Tue)03:43:22 No.103977872
>>103977780
why did you not post an example video?
Anonymous 01/21/25(Tue)03:46:34 No.103977892
ComfyUI_257520_
>>103977872
I didn't train it on videos.
Anonymous 01/21/25(Tue)03:47:36 No.103977899
>>103977892
does cosmos work in old comfy, or do you need the updates (I'm afraid to update)
Anonymous 01/21/25(Tue)03:48:12 No.103977901
>>103977892
so it just makes images now? i don't understand
Anonymous 01/21/25(Tue)03:53:32 No.103977942
ComfyUI_257533_
>>103977901
I trained a lora with a few images of fennec girl as a test and it worked very well. There's no reason you can't use a video model to gen single images.

>>103977899
yes you need the updates.
Anonymous 01/21/25(Tue)03:58:28 No.103977971
>>103977942
are there gguf quants of cosmos 7b? i'd like to try it
Anonymous 01/21/25(Tue)03:58:45 No.103977973
>>103977942
your loras always sucks and you have a habit of using models nobody cares about
Anonymous 01/21/25(Tue)04:08:52 No.103978056
ComfyUI_257545_
>>103977971
You most likely don't need quants if you are genning single images with it.
Anonymous 01/21/25(Tue)04:24:16 No.103978153
Comfyfag spamming these threads with his shit, low quality gens is the single biggest thing that stopped me from using ComfyUI. I'm sorry, buddy, but if that shit you post is the best your UI is capable off, then I'm not fucking using it, lol. Hire a marketing department and buy an ad or something. Also, A1111 won.
Anonymous 01/21/25(Tue)04:59:00 No.103978336
Anonymous 01/21/25(Tue)05:10:21 No.103978405
>>103976652
>>103976630
>>103976970
>>103977106
Do you have one where his followers are left holding the bag when Drump lets his meme coin scam collapse and runs off with the loot?
Anonymous 01/21/25(Tue)05:20:15 No.103978474
>>103975968
Why do I find her so damn attractive?
Anonymous 01/21/25(Tue)05:26:12 No.103978522
>>103978405
Trump just pardoned 1500 Trump supporters.

January 20th is Trump-Vance day. Yeah, I celebrate it.
Anonymous 01/21/25(Tue)05:27:14 No.103978534
>>103978474
Older you get, the more you're into milfs. Pretty wild huh?
Anonymous 01/21/25(Tue)05:28:44 No.103978543
1734317534699[1]
>>103978474
Pretty and she cleans up well.
Anonymous 01/21/25(Tue)05:31:24 No.103978554
>>103969763
the pressure to make lewds using the infinite custom porn generator is too great
Anonymous 01/21/25(Tue)06:25:10 No.103978857
>>103978554
>the pressure to make lewds using the infinite custom porn generator is too great
It will be so over when we actually get a good local video generator
Anonymous 01/21/25(Tue)06:51:44 No.103979001
00010-3767020125
>>103974387
CatTower
Anonymous 01/21/25(Tue)06:58:46 No.103979037
00012-0
hm, I wonder.
what if I put the negative prompts in the positive field, and the positive prompts in the negative field?
oh...that looks better than expected.
Anonymous 01/21/25(Tue)07:07:16 No.103979087
Are we going to have local video generation when the 5090 releases at the level that kling and hailuoai offer?
Anonymous 01/21/25(Tue)07:17:00 No.103979119
Why can I never make a human and an elf together? If I so much as type elf, everyone becomes an elf.
Anonymous 01/21/25(Tue)07:40:43 No.103979273
1708343278200062
Anonymous 01/21/25(Tue)07:51:21 No.103979346
>>103977780
no one cares for your garbage, go back to your containment thread
Anonymous 01/21/25(Tue)08:03:11 No.103979459
>>103975149
Random letters like this reminds me of pokémon games
Anonymous 01/21/25(Tue)08:31:37 No.103979672
file
>>103979119
Seems to just work for me on Flux models with decent rates of success. I'm guessing it might also work for Hyvid?

>>103977780
at least it can make a fennec girl now, but cosmos still sucks
Anonymous 01/21/25(Tue)08:41:36 No.103979775
>>103979273
god i wish i were that apple
Anonymous 01/21/25(Tue)08:44:45 No.103979800
>>103979087
>>103978857
Hunyuan is good enough we just need stronger GPUs at home so it feels as fast as the cloud models
Anonymous 01/21/25(Tue)08:49:02 No.103979843
>>103979800
>Hunyuan is good enough
based on the stuff i've seen, i'm not impressed with the boob physics for example
Anonymous 01/21/25(Tue)08:50:41 No.103979852
Anonymous 01/21/25(Tue)08:52:32 No.103979869
1716017715104053
>>103975942
Didn't actively try to suppress em. It just didn't gen em.
Anonymous 01/21/25(Tue)08:53:15 No.103979876
file
>>103979119
another. perhaps try this checkpoint https://civitai.com/models/941929
Anonymous 01/21/25(Tue)09:01:54 No.103979959
>>103979843
1. The Asian boob guy is probably using his minimax prompts and not trying to iterate or improve them for hyvid, because
2. we don't have h100s at home so it's excruciating to test new concepts and prompt templates, but
3. Unlike cloud models you can always train a Lora to improve the models understanding of a concept
Anonymous 01/21/25(Tue)09:38:32 No.103980228
00021-0
>>103979273
inspirational
i wish i could do the painted look as good as you
Anonymous 01/21/25(Tue)09:41:23 No.103980251
00023-0
ill keep trying
Anonymous 01/21/25(Tue)09:55:54 No.103980378
00058-0
Anonymous 01/21/25(Tue)10:00:51 No.103980422
cm66lj9zp002q336zm3s00syh
>couldn't generate for 6 hours yesterday because "Server Busy" due to paypigs
Two more weeks and I am liberated
But on the other hand the cyberpunk future is already here so who even gives a shit about cyberpunk themed art anymore
Anonymous 01/21/25(Tue)10:14:08 No.103980534
Am I imagining things or do loras for v-pred require more training than the usual ones? I have so many that go past 100 epoch
Anonymous 01/21/25(Tue)10:59:41 No.103981038
come on image corpos do something
Anonymous 01/21/25(Tue)11:10:57 No.103981139
>>103980422
she looks like she's sitting on the toilet pushing a turd out
Anonymous 01/21/25(Tue)11:13:28 No.103981166
392198985
Anonymous 01/21/25(Tue)11:58:05 No.103981567
Anonymous 01/21/25(Tue)12:04:33 No.103981655
1733850610487984
>>103979119
Just generate two humans and edit the ears. Even on illustrious, prompting two original characters with consistent details is an ass. I seriously consider making loras for my OCs
Anonymous 01/21/25(Tue)12:06:43 No.103981673
1722002662713204
Anonymous 01/21/25(Tue)12:14:00 No.103981767
S3136007305; ntrMIXIllustriousXL_xiii; DPM++ 2M SDE; 20250121134933;
Anonymous 01/21/25(Tue)12:37:47 No.103982013
How's your outputs folder? Mine since 2022 fall:
101,0 GiB (108 454 630 232)
117 240 files, 805 sub-folders
Anonymous 01/21/25(Tue)12:39:21 No.103982026
1731418275924673
Anonymous 01/21/25(Tue)12:39:56 No.103982034
Anyone know a consistent way to expand a limited dataset? I only have a few full body images and she’s smiling in all of the images… can I use KLING or something to cycle through expressions or get a video of her rotating to capture her behind?
Anonymous 01/21/25(Tue)12:42:18 No.103982061
>>103981139
>she looks like she's sitting on the toilet pushing a turd out
the prompt is for "office cubicle" but mochi can't do cubicles?
i was going to post a video of her with a disgusted expression in response to (You) but
>Server Busy
and I can't even paypig because I'd only get 8 "fast" videos
honestly after recent events in america I will be doing everything in my power to obtain a 5090 by the end of february because money might not matter soon (my retirement money definitely wont by the time i retire)
Anonymous 01/21/25(Tue)12:48:55 No.103982134
>>103982026
style tags?
Anonymous 01/21/25(Tue)12:54:57 No.103982203
>>103982061
>after recent events in america
Why, you losing your job to a pajeet or what?
Anonymous 01/21/25(Tue)13:02:36 No.103982256
cm66s7mzf005s336zeh9g4j8t
>>103982203
there's like 5 different reasons the 5090 is probably going to be expensive or difficult to get (tariffs/trade wars but also china fuckery but also america fuckery)

i was already 80% set on getting the 5090 (this mochi fuckery pushed me to 90%) but now that cyberpunk clown world dystopia is full steam ahead I am 100% down to clown

oh also I think we can all not be idiots here and see the writing on the wall that GPUs are going to be regulated for individuals soonish, which might be part of china's plan with releasing all this strong open source stuff for free
Anonymous 01/21/25(Tue)13:03:42 No.103982267
>>103982256
>GPUs are going to be regulated for individuals soonish
kek, you're far past delusional.
Anonymous 01/21/25(Tue)13:06:31 No.103982295
>>103982267
>kek, you're far past delusional.
im not talking about america. you can sell cameltoe videos of your 11 year old daughter on telegram and make 100k a month and report that to the IRS with no issues in america.

I however live in a much less free country
Anonymous 01/21/25(Tue)13:07:24 No.103982303
>>103982256
I agree with this, but I decided not to wait until the 5090 drops (it will be impossible for me to get it here anyway), so I just went and bought a PC with a 4090 the other day before they're all sold out like the 3090 was sold out before I could buy one from a store. I frankly doubt that the 5090 will be all that much better than the 4090 (ignoring the slightly bigger amount of VRAM that doesn't matter much to me anyway). My honest advice? Just buy a 4090, while you still can get one. The 5090 will be ridiculously more expensive for a marginally better increase in power.
Anonymous 01/21/25(Tue)13:17:10 No.103982396
cm66sjqeb00aa336zlm0u5sfk
>>103982303
>4090
maybe its just my inner jew, but something really rubs me the wrong way about paying MORE than MSRP for OLD tech

in my lizard brain i can justify the markup on the 5090 because its "new and shiny", and if the speed increase is 30% that's very nice. and I can always buy a second one if i want to generate 1440p hunyuan thots i guess
Anonymous 01/21/25(Tue)13:18:21 No.103982409
GPUs are silicon gold now. Just stockpile as much as you can, they will never devalue again especially if you keep them brand new.
If you don't need them you can just sell them for a profit.
Anonymous 01/21/25(Tue)13:32:35 No.103982566
cm66t9t6p002q336ust51kqre
>>103982409
>GPUs are silicon gold now.
GPUs are mostly glorified space heaters. It's mostly CUDA that makes them gold
Anonymous 01/21/25(Tue)13:36:50 No.103982627
1711681939310216
You guys like white girl or Japanese gyaru?
Anonymous 01/21/25(Tue)13:37:38 No.103982642
ComfyUI_SDXL_Upscaled_0002
Anonymous 01/21/25(Tue)13:39:58 No.103982668
>>103982642
It's over for me
installing stable diffusion now (again after a 2 year hiatus)
Anonymous 01/21/25(Tue)13:52:33 No.103982808
>>103982396
No one will ever be able to buy a card at MSRP unless it's 5 years old and out of warranty or something. I tried looking for a 3090 at MSRP (or below preferably, since it's as you say "old tech"), and I the only one I could find in my area was some dusty space heater for $700 that I'm sure would explode in a few months on me, if I bought it.
Anonymous 01/21/25(Tue)13:57:32 No.103982850
>>103982627
i never understood the appeal of gyarus for non-4chan incels that didn't hate white women

>>103982808
exactly, so I can at least lie to myself that the 5090 has a "new and shiny" tax, which I can't for the 4090
but if it's as dire as you say then even if I pay that tax for the 5090, it might not end up being a tax at all
>ANKH
Anonymous 01/21/25(Tue)14:06:42 No.103982939
>>103982642
What model is that?
Anonymous 01/21/25(Tue)14:16:46 No.103983054
Anonymous 01/21/25(Tue)14:24:52 No.103983120
>>103982939
https://civitai.com/models/997769/naixlmmmmix?modelVersionId=1286203
Anonymous 01/21/25(Tue)14:26:29 No.103983134
>>103982939
>>103983120

prompt from here:
https://civitai.com/images/51644312
Anonymous 01/21/25(Tue)14:28:14 No.103983155
>>103982939
horny anon
Anonymous 01/21/25(Tue)14:31:57 No.103983197
Anonymous 01/21/25(Tue)14:34:21 No.103983218
00063-2059903492
Anonymous 01/21/25(Tue)14:34:42 No.103983223
images
Is possible to train a pix2pix model using google collab?
Anonymous 01/21/25(Tue)14:36:48 No.103983252
>>103982013
98gb + 21gb across two hard drives
Anonymous 01/21/25(Tue)14:37:35 No.103983260
>>103982013
I don't compare penis size because it would make anon upset.
Anonymous 01/21/25(Tue)14:43:40 No.103983325
Anonymous 01/21/25(Tue)14:45:29 No.103983338
>>103983325
I'm willing to make a dataset of 200 pictures of anime girls and draw a 2-3 minutes sketch of each one.

But no clue over how to train the model on those dataset.
Anonymous 01/21/25(Tue)14:45:49 No.103983343
>>103980534
>I have so many
By all means please share some with the class
Anonymous 01/21/25(Tue)14:48:49 No.103983378
>>103982939
CatTower.
Anonymous 01/21/25(Tue)14:56:53 No.103983461
i still have yet to see anything from the cosmosverse that even comes close to the kino posted by anon when hyvid first dropped
Anonymous 01/21/25(Tue)15:01:07 No.103983503
>>103982013
i just converted some of the older pics to jxl and now it's down to 48.6GiB

also rather alarmingly file count went down by two files so perhaps my jxl conversion script isn't perfect and i lost two pictures. oh well.
Anonymous 01/21/25(Tue)15:12:46 No.103983651
hqdefault
Is there a way to cook a lora for illustrous using google collab?
Anonymous 01/21/25(Tue)15:22:49 No.103983756
Anonymous 01/21/25(Tue)15:43:20 No.103983976
1729165981255468
Anonymous 01/21/25(Tue)15:48:55 No.103984039
00063-827746328
Anonymous 01/21/25(Tue)16:38:41 No.103984620
>>103969385
can someone explain what samplers do in ComfyUI? Im too retarded to get where to put the output
Anonymous 01/21/25(Tue)16:40:44 No.103984647
ComfyUI_00002_
Anonymous 01/21/25(Tue)16:45:45 No.103984716
>https://www.reuters.com/technology/artificial-intelligence/trump-announce-private-sector-ai-infrastructure-investment-cbs-reports-2025-01-21/
>Trump to announce up to $500 billion in AI infrastructure investment, sources say
Good news I guess. Make 1girl great again
Anonymous 01/21/25(Tue)16:46:48 No.103984738
>>103984620
drag one of the images into your comfyui

https://comfyanonymous.github.io/ComfyUI_examples/sdxl/

there are a lot more examples if you go back 1 page
Anonymous 01/21/25(Tue)16:48:17 No.103984758
>>103984738
thanks alot! discovered you can drag input/outputs into empty area and it will suggest options with which to connect, works like a charm!
Anonymous 01/21/25(Tue)16:53:57 No.103984820
ComfyUI_00003_(1)
Anonymous 01/21/25(Tue)17:02:48 No.103984914
>>103984716
I don't trust any politician desu
Anonymous 01/21/25(Tue)17:05:26 No.103984953
>>103981655
There a guide on making loras anywhere? I need to do this for my OCs pronto.
Anonymous 01/21/25(Tue)17:12:42 No.103985057
ComfyUI_SDXL_Upscaled_0015
Anonymous 01/21/25(Tue)17:12:55 No.103985059
tips on removing body horror? ie 3 legs etc
Anonymous 01/21/25(Tue)17:15:11 No.103985090
>>103985059
lock the seed and play with positives/negatives, add shit like "malformed" to negatives and see if it helps without altering composition too much
otherwise, open pain, manually remove the third leg and inpaint the area
Anonymous 01/21/25(Tue)17:16:35 No.103985105
>>103985090
I'd suppose that by instigating
>body horror
>3 legs etc
in negative prompt.
Anonymous 01/21/25(Tue)17:22:55 No.103985206
Fresh on a slow Tuesday

>>103985188
>>103985188
>>103985188
Anonymous 01/21/25(Tue)19:27:45 No.103986699
Anonymous 01/21/25(Tue)20:54:10 No.103987655
>>103983338
why not use a controlnet for that instead?
why pix2pix?