https://pastebin.com/NYdMmBGH

File: da nooz 19.jpg (579 KB, 1344x960)

579 KB JPG

>mfw Resource news

08/05/2024

>VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling
https://github.com/daixiangzi/VAR-CLIP

>AI chip startup Groq lands $640M to challenge Nvidia
https://techcrunch.com/2024/08/05/ai-chip-startup-groq-lands-640m-to-challenge-nvidia

>Flux Dev & Schnell Support Added to Pallaidium
https://github.com/tin2tin/Pallaidium#change-log

>AIマンガ「LearningToon」無料公開:社会人向け縦読み学習
https://prtimes.jp/main/html/rd/p/000000011.000138218.html

>Proof of Concept LoRA on Flux Schnell
https://x.com/ostrisai/status/1820462674230059328

08/04/2024

>SimpleTuner v0.9.8 - flux 24 gig training has entered the chat
https://github.com/bghira/SimpleTuner/discussions/639

>FastSD CPU v1.0.0-beta.35: Adds Aura SR v2 support
https://github.com/rupeshs/fastsdcpu/releases/tag/v1.0.0-beta.35

>SimpleTuner now supports Flux.1 training (LoRA, full)
https://github.com/bghira/SimpleTuner#flux1

>Civitai Model Manager: CLI tool for managing AI models from CivitAI
https://github.com/regiellis/civitai_model_manager

>Skimmed_CFG: Powerful ComfyUI anti-burn allowing much higher CFG
https://github.com/Extraltodeus/Skimmed_CFG

>ComfyUI-AdvancedLivePortrait
https://github.com/PowerHouseMan/ComfyUI-AdvancedLivePortrait

08/03/2024

>ComfyUI/Forge Implementation of Smoothed Energy Guidance
https://github.com/pamparamm/sd-perturbed-attention

>TryOnDiffusion: A Tale of Two UNets
https://github.com/fashn-AI/tryondiffusion

>Nvidia reportedly delays its next AI chip due to a design flaw
https://www.theverge.com/2024/8/3/24212518

>ComfyUI Frontend Modernization: Transitioning to a New Era on August 15, 2024
https://github.com/comfyanonymous/ComfyUI/issues/4169

>CEO of Invoke says Flux fine tunes are not going to happen
https://www.reddit.com/r/StableDiffusion/comments/1eiuxps

>ComfyUI-FLUX-fal-API
https://github.com/gokayfem/ComfyUI-FLUX-fal-API

>mfw Research news

08/05/2024

>Conditional LoRA Parameter Generation
https://arxiv.org/abs/2408.01415

>Transformers are Universal In-context Learners
https://arxiv.org/abs/2408.01367

>Hallu-PI: Evaluating Hallucination in Multi-modal LLMs within Perturbed Inputs
https://arxiv.org/abs/2408.01355

>TexGen: Text-Guided 3D Texture Generation with Multi-view Sampling and Resampling
https://dong-huo.github.io/TexGen

>General Framework to Boost 3D GS Initialization for Text-to-3D Generation by Lexical Richness
https://arxiv.org/abs/2408.01269

>S2TD-Face: Reconstruct a Detailed 3D Face with Controllable Texture from a Single Sketch
https://arxiv.org/abs/2408.01218

>Interpreting Global Perturbation Robustness of Image Models using Axiomatic Spectral Importance Decomposition
https://arxiv.org/abs/2408.01139

>CoLoRA: Contribution-based Low-Rank Adaptation with Pre-training Model for Real Image Restoration
https://janeyeon.github.io/colora/

>EIUP: A Training-Free Approach to Erase Non-Compliant Concepts Conditioned on Implicit Unsafe Prompts
https://arxiv.org/abs/2408.01014

>FBSDiff: Plug-and-Play Frequency Band Substitution of Diffusion Features for Highly Controllable Text-Driven Image Translation
https://arxiv.org/abs/2408.00998

>Reclaiming Residual Knowledge: A Novel Paradigm to Low-Bit Quantization
https://arxiv.org/abs/2408.00923

>Toward Automatic Relevance Judgment using Vision--Language Models for Image--Text Retrieval Evaluation
https://arxiv.org/abs/2408.01363

08/04/2024

>Comprehensive Survey of Complex-Valued Neural Networks: Insights into Backpropagation and Activation Functions
https://arxiv.org/abs/2407.19258

>Detached and Interactive Multimodal Learning
https://arxiv.org/abs/2407.19514

>Take A Step Back: Rethinking the Two Stages in Visual Reasoning
https://arxiv.org/abs/2407.19666

>ActivityCLIP: Enhancing Group Activity Recognition by Mining Complementary Information from Text to Supplement Image Modality
https://arxiv.org/abs/2407.19820

>>101738111
>>101738090
Not even hiding the samefag

>>101738090
>>Discord
>6wUwtcJsr2
Who's all in there?

>>101738102
he had to bake early lal

>>101738132
the old avatarfags are still posting there daily

File: vramlet fed.jpg (364 KB, 1424x1024)

364 KB JPG

>>101738137

The vramlet thanks you for your offering.

File: 1964083186-flux1-schnell-1.png (1.49 MB, 1024x1024)

1.49 MB PNG

>>101738194
Always happy to help

>>101738171
>the old avatarfags are still posting there daily
sounds awful

>>101738171
So ones like Fran, Debo, Quokka are in there?

>>101738111
>AI chip startup Groq lands $640M to challenge Nvidia
>Groq CEO Jonathan Ross
what's with all the actors getting into ai

>>101738318
no those three are not in there

>>101738332
So no avatarfags are in there?

>>101738338
well some are in there but not all, pw and bmp aren't in there either

File: file.png (11 KB, 925x75)

11 KB PNG

cpu goes brrr

>>101738318
Debo got chased out and has been seething ever since

>>101738350
So none of the annoying avatarfags are in there? Sounds neat.

>>101738367
The roleplaying ones aren't in there.

File: delux_bu_00012_.png (1.24 MB, 1344x768)

1.24 MB PNG

>>101738360
seems more like debo chased all the neerdowells off sdg :D
debo stay winning

certainly winning at shit gens

neck yourself debo kek

>>101738090
I don't understand the OP image

Daily reminder that FLUX finetune is possible and that LoRA already have been made

https://github.com/bghira/SimpleTuner/blob/main/documentation/quickstart/FLUX.md
https://github.com/bghira/SimpleTuner/pull/622#issuecomment-2267624531
https://x.com/ostrisai/status/1820462674230059328

>>101738644
It's only gonna cost you an arm and a leg

>>101738650
3xA100 80GB is like 5$/h anon.

File: file.jpg (1.16 MB, 1792x1024)

1.16 MB JPG

new big data banger dropping later today
>>101738681
i get 2xA100 80GB from azure for <$1/hour

>>101738644
The lie that Flux was impossible to finetune or create Lora's for was spread by Invoke CEO, since he makes all his money by selling access to a censored version of Stable Diffusion and was salty an alternative appeared.

>>101738709
Yes it's just runpod price I'm lazy
+ You tease all day but I still don't know what you tease or what you want to release lmao

File: 1722885708020_out-0.jpg (75 KB, 1216x832)

75 KB JPG

>>101738539
I have seen people on reddit who have been able to run it on 4gb, not sure if they were talking about Schnell or dev, I'll have to look into it.
But it's possible (slow AF but possible)

File: file.png (212 KB, 1187x695)

212 KB PNG

>>101738729
this one is just under 1B generations+metadata from leonardo
pic related are the current releases
https://huggingface.co/bigdata-pw
>>101738717
when did invoke make a saas, thought they just made their ui (with $3m funding somehow)

flux for composition and prompt comprehension fed to 1.5 controlnets for styles / celebrities etc is all you need

>>101738799
yeah it takes like 1-3 minutes per image at 20 steps. Has to be a faster way without spending more money.

File: IMG_6333.jpg (73 KB, 1024x1024)

73 KB JPG

>>101738896
Can you do that right now?

>>101738799
I've been running dev with an 8GB card and my average is about 140 seconds ish. It always drops into lowvram mode so I'm probably losing quality.

whats the best realism model out now?

File: ComfyUI_temp_cycnz_00005_.png (2.61 MB, 1536x1536)

2.61 MB PNG

>>101738090
New release: Fang (From Snoot Game. Nothing else) PONY Lora.
https://civitai.com/models/160952

If you guys can. Please submit some pictures to be uploaded to the showcase reel. (I will credit if you ask).

File: ComfyUI_temp_cycnz_00016_.png (2.64 MB, 1536x1536)

2.64 MB PNG

>>101739246

>debo is crying

File: 00003-2486591806.jpg (258 KB, 1304x1304)

258 KB JPG

WHATS THE BEST NEW XL MODEL (NON PONY)

File: delux_bu_00013_.png (1.08 MB, 1344x768)

1.08 MB PNG

>>101739283
no one uses XL anymore. but also, couldn't you just look at whats trending on civit?

>brap

File: delux_flebo_00016_.png (1.16 MB, 1216x832)

1.16 MB PNG

>>101739269
https://suno.com/song/71b244e0-6a38-467f-9471-f49192cb3701

>>101739683
why do you keep adding the discord to OP debo?

File: delux_tru_00019_.png (1.41 MB, 1344x768)

1.41 MB PNG

>>101739692
I don't bake

debo unironically killed this general, /ldg/ is thriving with /sdg/ refugees

>>101739438
Couldn't you not be immediately combative? Sure. But you do it anyway.

>>101739712
do you realize why anons call you the thread schizo?

>>101739712
Yeah you just happen to be there when peoples copied gens are posted to immediately 60 seconds later post the news. Totally

File: ComfyUI_temp_cycnz_00023_.png (2.1 MB, 1536x1536)

2.1 MB PNG

>>101739256

>>101739647
So, is it 13gb or 20gb to run it properly?

>>101730183
warning to all new friends, do not engage I repeat, do not engage

File: 40p2y19s3tgd1.jpg (1.94 MB, 9967x2393)

1.94 MB JPG

Guys have you seen this shit? Looks like we can use CFG and negative prompt on flux with this node.
https://new.reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/

Someone remixed that workflow so that we can use negative prompts even at cfg = 1
https://civitai.com/models/625042/efficient-flux-w-negative-prompt

This debo hate stops NOW!

>>101738102

File: delux_bu_00014_.png (939 KB, 1344x768)

939 KB PNG

>>101739939
looks interesting

It's funny how often he changes his filenames since longtime posters filter him

>>101739939
fake and gay

>>101739939

is it possible to make lewd stuff with flux?

>>101739914
As I pointed out to Debo, I'm not new. He and I have both been around here a long time. I don't intend to filter him or pretend he doesn't exist merely because we've had some unpleasant interactions.

>>101740082
>>>/b/

@101740098
obviously d*bo

File: LEAVE_DEBO_ALONE.jpg (29 KB, 479x398)

29 KB JPG

>>101739943

101740113
you're not fooling anyone.

>>101740193
oooh, how did you get that noise in the image?

>>101739901
please respond

>>101740208
Upscaling, and photoshop.

i miss schizo anon

>>101740271
>and photoshop.
what the flux, that's unethical

>>101740258
12GB for fp8 unet 24GB for fp16 unet. T5 you should run at fp16 if you have at least 32gb of ram and a optimized pc setup. if your setup isn't optimized or you don't have 32gb ram, just use the fp8 t5

File: settings.png (927 KB, 1409x931)

927 KB PNG

>>101740302
>auto contrast/lecels etc

>>101740311
>https://civitai.com/models/625042/efficient-flux-w-negative-prompt
I see. What is T5? a version?

****

>>101740376
https://github.com/comfyanonymous/ComfyUI_examples/tree/master/flux

>>101740271
>>101740319
Cool. Thanks!

>>101740422
Thanks.
I'm an A1111fag, but it seems I'll have to learn Comfy, now.

File: 000000_16023_.png (1.89 MB, 1401x1051)

1.89 MB PNG

>>101740423
Yw,

>>101738090
What's the point of this thread other than chatting with PW and Debo?

>>101740311
>T5 you should run at fp16 if you have at least 32gb of ram and a optimized pc setup

Damn. I thought my 3090 + 64GB would last a couple more years. I'll be needing an upgrade, soon.

>>101740484
Ran spergouts

Will we have PonyFlux?

is stable diffusion event relevant anymore? we should rename this subreddit

Ran kicked debo out of the discord

File: Flux_00004_.png (3.34 MB, 3321x1974)

3.34 MB PNG

Seems that PerpNegGuider allows you to use the negative prompt even with cfg = 1
https://ku-cvlab.github.io/Perturbed-Attention-Guidance/
>Perturbed-Attention Guidance significantly enhances the sample quality of diffusion models without requiring external conditions, such as class labels or text prompts, or additional training. This proves particularly valuable in unconditional generation settings, where classifier-free guidance (CFG) is inapplicable

>>101740595
>is stable diffusion event relevant anymore?
It's relevant for as long as Flux can't do porn
The moment Flux can it's absolutely 100% over for SD though

>>101738123
who the fuck read papers

>>101740662
he certainly doesnt

>>101740662
Not debo

Is it me, or Comfy needs a lot more memory than forge?

>>101740595
>subreddit
You jest but debo always replies with the most reddit responses

>>101739269
>>101740288
Debo does this to farm post

File: file.jpg (430 KB, 1792x1024)

430 KB JPG

>tfw server crashed

File: delux_bu_00015_.png (1005 KB, 1344x768)

1005 KB PNG

>>101740121
someone should luma these meme

>>101740484
>other than chatting with PW and Debo
you need another reason?

>>101740517
>I'll be needing an upgrade, soon.
the worst part is there is no upgrade. the 5090s are rumored to have only 28gb so even the next gen flagship isn't gonna fit current gen model

>>101740596
wow! very thin-skinned thing to do!

>>101740625
is this both dev and shnell? any idea how it might differ from >>101739939

>>101740856
You're a disgrace

File: delux_bu_00016_.png (1.07 MB, 1344x768)

1.07 MB PNG

>>101740595
we're not named after the diffusion model. we're named after the action of using diffusion models in a stable fashion. common misconception

>>101740662
the bigger brains amongst us will sometimes read/comment on papers. in rare cases, thread devs might find something new to implement. someone recently implemented SEG from a paper

>>101740669
>>101740670
too true. too busy bangin your moms for all that nerd shit LOL

>>101740723
this!!

>>101740856
>is this both dev and shnell? any idea how it might differ from >>101739939(You)
the reddit workflow use the "classic" high CFG mixed with a DynamicThreshold so that it works

the civitai workflow use the Perturbed Attention Guidance node which is basically a remixed version of CFG that allows you to use negative prompt even at CFG = 1

>>101740901
are you ok?

File: bike monster nsfw.jpg (157 KB, 674x672)

157 KB JPG

sdxl aryan

File: delux_flebo_00015_.png (1.15 MB, 1216x832)

1.15 MB PNG

>>101740883
your*

>>101740912
I'm great. how are you?

>>101740625
PerpNeg and PAG are not the same, anon

>pastebin link still up
>ffaze seething

>>101740967
oh yeah you're right... so... what's the best one? because it seems like the both of them can make negative prompt work at cfg = 1
https://mybyways.com/blog/testing-new-pag-and-perp-neg-nodes-in-comfyui

>>101740484
Where else am I gonna get my malware news links

Do you think his failed DJ carrier lead him to be a 4chan schizo

Schizo containment thread

>>101740990
flux can't use PAG (since it was designed for u-net, and flux is a transformer), so you have only one option

File: generated_image_1722896258058.png (1024 KB, 896x1088)

1024 KB PNG

just posted this in lmg by accident... technically this is a local model pls no ban. anyway heres my favorite game from my childhood

>>101741242
could you give him some sith aesthetics like grey skin, dark baggy eyes, etc?

>>101738896
It doesn't understand any nsfw/suggestive composition.

man i just started using my 4090 for art gen because flux-1.dev actually can produce nice results via CLIP that don't look like that generic pony.
Though i learned fast that --gpu-only and using fp8_e4m3fn is far better (approx 40s for 50 steps and a basic pipeline) than allowing the cpu to ever touch the model.
My real question is i'm thinking to write a simple Node that stores my prompts and an id to put into the filename since i can only batch stuff at night due to the heat generation.
Has someone else already wrote such a thing for comfyui yet?
My filenames already include the model and seed but without the prompt you cannot go back easily.

File: file.png (9 KB, 734x97)

9 KB PNG

soon:tm:
probably would have been done by now if the server didn't crash >:(
i guess it got too hot

>>101738090
Getting back into image generation after a bit of a break. Can anyone suggest a solid model that generates realistic humans in nature backgrounds?
Ex. Girl in the middle of a forest at night

>>101741427
bstaber

>>101741265
What do you do in this game?

File: delux_bu_00017_.png (1.11 MB, 1344x768)

1.11 MB PNG

>>101741427
flux

>>101741459
stub your toe repeatedly. its open world so theres really no victory condition. way ahead of its time

File: 000000_16031_.png (2.04 MB, 1075x1434)

2.04 MB PNG

>>101741292
I'll try.

>>101741472
Fuck!

>>101741403
sounds like a you problem.

>>101741474
Darth Naroge...

>>101741478
if it wasn't i wouldn't be willing to write a node. Is that supposed to mean something?

>>101741470
Thanks, this looks exactly like what I need.

>>101741474
is that a shake weight?

>>101741515
It's reptilian gold.

File: deflux_mo_00009_.jpg (354 KB, 1344x960)

354 KB JPG

>>101741474
>Kamala Harris is a powerful jedi who is telepathically hovering coconuts around her, Kamala harris is in a state of deep focus with her eyes closed and one arm raised, the scene is dark and moody on the planet dagobah where Kamala Harris is practicing the force
idk who this lady is but she wields the power of the coconut tree

>>101741565
Oh shi.... powerful she is.

>>101741242
wow this is so deep, Trump is truly the sith of our time

File: darthtrump.png (1.06 MB, 1024x768)

1.06 MB PNG

>>101741242
my attempt

>art? nein
>celebrities? nein
>nudity? nein
>/pol/ propaganda? ja

File: ComfyUI_temp_00001_.png (967 KB, 768x1024)

967 KB PNG

>>101741565
nice prompt changing one word works well

does anyone know if i am allowed to post an image with a nude greek statue of a woman? its not really a horny image its just a meme album cover. ive been banned for toeing the line before and i dont want you all to miss out on my GENIUS memes

File: dragontrump.png (2.5 MB, 1137x1137)

2.5 MB PNG

>>101741606
better than shitty furryposting

>>101741606
>anons making something unique and funny for once
>this fag mad it's not his 1girl, blue_hairs, anime face, hatsune_miku, slop
get a grip, fag

>debo gearing up another mass reply

>>101741684
*report

File: deflux_mo_00014_.jpg (326 KB, 1344x960)

326 KB JPG

>>101741611
I think they have to be half-watermelons, no?

>>101741633
people have gotten away with it before. its really up to whatever mood the jannies happen to be in on any given day.

>>101741684
whats the minimum number of replies to quality as 'mass'?

>>101741606
Yeah, it's almost like the flux creators decided to release that model just to do some election inference kek

File: ComfyUI_temp_00004_.png (1005 KB, 768x1024)

1005 KB PNG

>>101741725
i just started to edit the prompt with a lightsaber but

File: darthtrump2.png (2.11 MB, 1072x1432)

2.11 MB PNG

>>101741474
>>101741707
its super stubborn about including the red tie with every trump prompt. I had to clone stamp it away in paint.net and do one more pass

>>101741743
kek, I originally prompted for the red tie.
nice work.

cinematic film style Star Wars style of Dark Sith Lord (DONALD TRUMP:1.4) looking old and lack of color, Grey skin, wrinkles,wearing a black cloak with red tie, dynamic,  highly detailed, outside a compound made of shiny stainless steel, (industrial city center),, holding long brilliant glowing florescent red light saber,

File: deflux_mo_00016_.jpg (542 KB, 1344x960)

542 KB JPG

>>101741739
if you add
>Kamala Harris is half black half indian
the lady starts to look a but more like kamala

File: Capture.jpg (232 KB, 2023x1615)

232 KB JPG

Is there a way to put multiple guiders on SampleCustomAdvanced? I'd like to add the regular cfg on it aswell

File: file.png (206 KB, 541x592)

206 KB PNG

>>101738090
why doesn't flux understand penis?

File: file.jpg (422 KB, 1024x1792)

422 KB JPG

one shot, click-click-click, reload, reload, reload

Schizophrenic vibes itt

File: HARAM.jpg (104 KB, 874x1008)

104 KB JPG

>>101741841
>why doesn't flux understand penis?
because sex is HARAM

I like images that stimulate my imagination. Something a little dense, a little abstract, and open to interpretation.

>>101741859
wrong i can easily gen a woman naked.

>>101741859
flux can generate plenty of it but using the default workflow i find it goes to anime quickly

>>101741881
With pepperoni nipples

>>101741884
the website doesn't gen penis though :(

File: taylor_sphinx.png (3.89 MB, 1536x1536)

3.89 MB PNG

>>101741877
does taylorposting still get your banned here?

>>101741905
oh i mean dev not schall or whatever that version that isn't nsfw is called

>>101740595
stable diffusion is grandfathered in

File: 000000_16036_.png (2.21 MB, 1075x1434)

2.21 MB PNG

>>101741776
kek makes sense,,//doesn't know Tucker..

File: ComfyUI_temp_00011_.png (1.02 MB, 768x1024)

1.02 MB PNG

>>101741776
that got close though had to rewrite the prompt

Did Astralite Heart say something about Flux, more specifically if it's possible and he has the intention to fine-tune Flux to Pony?

Can't wait for the american elections to end, can't stand seeing shitty politician faces all the time.

File: mechcc.png (1005 KB, 670x675)

1005 KB PNG

sd mech

>>101742006
https://www.reddit.com/user/AstraliteHeart/

>>101741597
Why does he look like a low resolution 3D print?

>>101742055
Oh, so he's working on AuraFlow.
I think it will be at least a year, until we get PonyFlux.

File: lol.png (1010 KB, 768x1024)

1010 KB PNG

>>101742081
not low res but i still laughed

>>101741908
>does taylorposting still get your banned
Illegal in Missouri
https://www.krcu.org/crime-safety/2024-03-28/missouris-taylor-swift-act-targets-ai-image-threats

File: 000000_16038_.png (2.26 MB, 1075x1434)

2.26 MB PNG

>>101742014

File: gantz01.jpg (221 KB, 796x800)

221 KB JPG

sdxl gantz, akira, blade runner

>>101733871
>>101733825
>>101733809
Thanks guys!!

>five fingers

holy shit, it's improving every year!

>>101742274
Looks like a mess desu

File: flux fail.png (1.09 MB, 768x1024)

1.09 MB PNG

>>101733809
ok that is something flux has a problem with or i dont know the proper CLIP prompt i tried 5 different ones, but this is my second day prompting

File: file.jpg (399 KB, 1792x1024)

399 KB JPG

ethernet vibes, connections kinda lagging
memories upload, but nostalgia keeps nagging

>>101738090
Can I use stable diffusion XL on a laptop with 8 gb VRAM? I'm using regular stable diffusion 1.5 now.

File: ComfyUI_temp_00026_.png (1.19 MB, 768x1024)

1.19 MB PNG

>>101742522

File: whoframed.png (2.9 MB, 1728x1296)

2.9 MB PNG

I spent all this time setting up a mask workflow to avoid having to use the anireal models and loras, and it turns out you'd still need it for the final composite, and even still 1.5 can't keep gender and artstyle straight when denoising the whole final image so the mask composite borders stay ugly. Waifu Roger Rabbit is still a fun idea though, it was worth a shot.

File: dancer01.jpg (180 KB, 795x798)

180 KB JPG

>>101742417
Ya. It's sd gacha, idea generator.

>have a clear idea and do controlnet and upscaling
someday I will

>>101742542
barely.

>>101742542
you can always run smaller images and upscale if you run out of memory or convert the model to a simpler data type.
Flux-1.dev cannot fit in gpu memory only if you run the full net without converting to fp8 on a 4090 but the fp8 gives like 98% of the result at like 1000% faster

File: aaa.jpg (1.44 MB, 4040x2393)

1.44 MB JPG

Ok guys, do not sleep on PrepNegGuider, this shit is really cool (if you are ok with the speed decrease), I provide you a workflow for those interested
https://files.catbox.moe/eriobu.png

Because we got the exact same output for regular CFG and PrompNeg CFG when we don't put any negative prompt, comfy should code this shit so that it revert back to regular CFG when we don't put negative prompt (for the speed increase)

https://perp-neg.github.io/

>>101742613
this comparison implies that the negative fucks up the prompt comprehension

>>101742613
does anyone really like guidance 3.5? i like 2.0ish when i'm looking for gens that i won't post here kek

File: Capture.jpg (33 KB, 912x494)

33 KB JPG

>>101742625
true, you can change the strength of the negative prompt though, I think 1 is too high

File: BMP_FLUX_03569_.png (1.66 MB, 1024x1024)

1.66 MB PNG

https://suno.com/song/c697f082-5d33-4b34-a644-8bc2085aad76

File: file.jpg (410 KB, 1792x1024)

410 KB JPG

terabytes of tenderness, stored in folders deep

File: de_fl_00130_.jpg (1.18 MB, 1344x960)

1.18 MB JPG

>>101742733
reeeeeeeeemiiiiiiiiix
https://suno.com/song/f3e9b21b-a3da-4775-a95e-10ef175a824e

How many times will debo reply to bmp it's just sad

>>101742792
What music style? I'll make a suno

File: file.jpg (543 KB, 1792x1024)

543 KB JPG

https://suno.com/song/744ea913-adc0-4fd3-9678-52b990ef5237

>>101742807
>doll boxes actually work
We're finally getting somewhere.

>>101742851
Nvm! Beat me to it, nice vibe.

File: ComfyUI_temp_00030_.png (880 KB, 768x1024)

880 KB PNG

>>101742855
10s prompt did it fine

>>101741797
Use Switch (Any) from https://github.com/ltdrdata/ComfyUI-Impact-Pack

>>101742962
thanks anon! Is there a way to add conditions, like if PrepNeg cfg = 1 and there's no negative prompt, then it switch to regular cfg? it's for this workflow https://files.catbox.moe/eriobu.png

File: BMP_FLUX_03585_.png (1.06 MB, 1024x1024)

1.06 MB PNG

Fuck it I made one anyways
https://suno.com/song/f9b16b88-aee2-4a0b-8701-301e79d9ebc1

>>101742999
Yeah there's logical nodes on that same custom node, but I'm brainlet desu

File: ComfyFLUX_00001_.png (578 KB, 1024x576)

578 KB PNG

So i did it, FLUX running locally on my laptop, it takes around 2:30 mintues with 4 steps and 4:00 mintues with 8.
I can't complain, for some reason local FLUX seems to be much better at making Quokkas.

File: file.jpg (481 KB, 1792x1024)

481 KB JPG

big data dropping now desu. well, its hashing atm, probably take about 15 minutes, then the upload will take about 40 minutes
>>101742880
thanks!
>>101743039
nice!

>>101743320
Kek I just noticed I posted the wrong one, kms
https://suno.com/song/e7e9d803-f591-457d-9986-6a927d63b5f8

File: ComfyUI_00945_.png (1.28 MB, 1024x1024)

1.28 MB PNG

>a 20 year old Donald Trump
Trump is so overcooked you can't change anything on him

>>101743474
speaks more about the models right, even if you have 2B images shit from the pre-digital era are pruned from the datasets for the most part.
I remember in the 90s an expensive camera had 1.2Mb shots as a nice $2k dlsr

File: media_GUPmuEVXwAA2kb8.jpg (106 KB, 1024x1024)

106 KB JPG

https://xcancel.com/Konan92_AI/status/1820518655450562588#m
Shot's fired

>>101743538
flux's CLIP is too nice to care about that shit, imagine reading a 50 page document to learn tags

File: file.jpg (399 KB, 1024x1792)

399 KB JPG

>>101743379
what i really want is stuff like yung lean and bladee but suno doesn't allow artists names and there's no way suno's lyric gen will come up with anything like it
https://open.spotify.com/track/1zX178V8sWozr96MrfmRun

>>101743474

Trump and Miku can only be changed by fusion.
>>101742422

File: not smug.png (878 KB, 768x1024)

878 KB PNG

>>101743714

File: mech02.jpg (179 KB, 752x754)

179 KB JPG

mech diffusion

File: file.jpg (333 KB, 1792x1024)

333 KB JPG

>156/958 [09:56<49:41, 3.72s/it]
zzzz

>>101743553
>flux's CLIP
you mean flux's T5

File: FLUX_00037_.png (1012 KB, 896x1152)

1012 KB PNG

OC do not steal

>>101743816
>>101743780
prompt was just: a sarcastic anime girl laughing while holding her hand near her face
their engine works well enough i don't really care as i can shitpost in english
also t5 was shit if you had a cell phone back then

File: R7OXJYf.png (63 KB, 757x546)

63 KB PNG

https://www.reddit.com/r/StableDiffusion/comments/1ekgiw6/comment/lgpgyhf/?utm_source=share&utm_medium=web2x&context=3
>But I know was getting better results after adding in the FluxGuidance, could be a fluke but time will tell. I'll experiment more and report back if I learn anything interesting. Currently trying to integrate the negative prompting into Img2Img with Flux.
Wtf? So there's a difference between the two?

File: BMP_FLUX_03624_.png (1.45 MB, 1024x871)

1.45 MB PNG

>>101743593
Yeah that's a tough one, I keep getting either trap or rap pop, rip

File: file.jpg (235 KB, 1536x1536)

235 KB JPG

>>101743780
pretty cute

File: kek.png (1.02 MB, 768x1024)

1.02 MB PNG

>>101744037
> jessica rabbit as a washed singer at a cheap bar holding the mic with an expression of regret

>>101743961
suno doesn't know bassline or uk garage either
https://open.spotify.com/track/2xUjNGHjJOBcpTf7dYx4EY

>>101744062
I prompted for jessica rabbit and it gave me an actual rabbit, so I just described the character

>>101744134
many terms cause the flip to anime i've noticed, but if you describe it using terms never used in anime (e.g. real life things) it tends to avoid the anime stuff w/o needing to do that negative prompt crap and lose 50% performance. but i've only tested like 1000 prompts or so

File: BMP_FLUX_03636_.png (1.61 MB, 1024x980)

1.61 MB PNG

>Flux doesn't know Desu-chan
Why even live
https://suno.com/song/5991acf5-d9bb-40c4-9326-e5f2792cc142

File: file.jpg (297 KB, 1792x1024)

297 KB JPG

>can i get a big data banger bro
yhyh be bout 15 mins

File: delux_cos_00004_.png (1.3 MB, 1216x832)

1.3 MB PNG

>>101744123
>uk garage
thats a fun idea I never thought of. you could try "2-step" as an alternative it might know

File: desu.png (819 KB, 768x1024)

819 KB PNG

>>101744201
web searches don't even show her anymore so i think there is nothing to train against for their models

File: ComfyFLUX_00002_.png (543 KB, 576x1024)

543 KB PNG

>>101738090
So I'm using a workflow that's recommended for under 12gb vram use of FLUX it works but there's this one node that I'm not sure how it's used but I'm also sure it's the reason why sometimes stuff doesn't come out perfect
Can anyone explain me how to use split sigmas? From the little i googled about it 10000 is like the limit that should be used but when i used 10000 it came out all deepfried but anything less than 8000 makes everything blurry.

File: Capture.jpg (384 KB, 2777x1474)

384 KB JPG

>>101743474
cfg 3 + DynamicThresholdingFull seems to do the job fine, this is a good tool to make flux even better at prompt understanding, I wish I could find the best parameters though so that it looks less "fried" I would say

>>101744340
that image looks like it was made with a pony model though

File: delux_cos_00007_.png (1.43 MB, 1216x832)

1.43 MB PNG

>>101744320
I'd like to know this too. sigmas are a mystery to me

File: file.jpg (443 KB, 1792x1024)

443 KB JPG

https://huggingface.co/datasets/bigdata-pw/leonardo
hopefully it won't crash their dataset viewer server like last time
>>101744221
2-step is different desu. i want 4x4 niche bassline

>>101744354
yeah, like I said, cfg 3 completely destroys the image and DynamicThresholding is trying its best to make it work, I'm pretty sure the values aren't optimal and it could look nice, but with so many parameters it's hard to do some rigourous test

File: file.png (156 KB, 1620x810)

156 KB PNG

all good. pls like it

>>101744406
1 billion? where's the rest?

File: BMP_FLUX_03651_.png (1.59 MB, 1024x1024)

1.59 MB PNG

>>101744320
You guys are already making your own workflows? NEEEEEEEEEEEEEEERDS!

File: file.png (19 KB, 589x92)

19 KB PNG

>>101744421
here and on my server. ill put them together soon:tm: they have different fields and stuff
>>101744431
my workflow is type what i want and click the button :^)

I wish comfy had a button to swap height and width.

>flux

are we back bros?

>>101744468
doesn't sound too hard to implement

File: ComfyFLUX_00004_.png (557 KB, 576x1024)

557 KB PNG

>>101744431
I used this one
https://civitai.com/posts/5006398

File: delux_cos_00001_.png (986 KB, 1216x832)

986 KB PNG

>>101744468
you can get the comfyroll node pack and use the SDXL aspect ratio node. it has a bunch of presets plus swap dimensions

File: file.png (17 KB, 449x125)

17 KB PNG

oh it only has the first page as a preview. crashed their server again. 2big4u

File: BMP_07378_.png (1.8 MB, 1328x1328)

1.8 MB PNG

>>101744495
Thank you Quokkers, I'll try it out tomorrow

>>101744397
are you the guy trying to load the 22GB model in a 12GB card? I saw there was a guy on hugging face that uploaded the model converted to fp8 so it was 11gb but in memory it ended up using the full 22gb anyways unless you still converted it to fp8_e4. the results for prompt and seed were the same at least so he didn't fuck with it but most don't know how to edit flux yet
With a 4090 i have to swap to fp8 to keep it in memory and doing the cpu offloading is a joke as last time i tried that a 12s gen took like 2m

File: BMP_22856_.png (1.25 MB, 1280x720)

1.25 MB PNG

>>101744628
I love your 2hu gens

File: meat.png (957 KB, 768x1024)

957 KB PNG

>>101744628

>>101744624
>are you the guy trying to load the 22GB model in a 12GB card?
nope
>With a 4090 i have to swap to fp8 to keep it in memory and doing the cpu offloading is a joke as last time i tried that a 12s gen took like 2m
That's weird, did you put some -lowvram flag on the .bat? on normal_vram it should work fine for a 4090

File: delux_cos_00026_.png (706 KB, 1216x832)

706 KB PNG

what the flux

>>101744647
no i use --gpu-only so no offloading. the flux-1.dev cannot fit in the 4090 without going fp8. The results are like <1% as their documents state and the performance difference is night and day.
The default to unload the model is dumb in comfyui and the config panel doesn't even let them know that kek
I just have a window for each model and each batch will cause one to unload as required but each batch runs far faster

File: food.png (1.05 MB, 768x1024)

1.05 MB PNG

>>101744681
i wanted to attach this but forgot when talking about comfy ui oh no

>>101744681
>no i use --gpu-only so no offloading.
it can't work anon, the image model + VAE spike at the end of the image inference is 15gb, and the image model is 10gb, you should remove that --gpu-only and let the comfyUi software do the memory management for you, you're overflowing

>>101744694
4090 has 24gb and holds it fine. I gen at ~40s for 50 steps vs offloading at 500+

>>101744708
try to remove the flag still? if there's an improvement on your side don't forget to thank me, if nothing changes you can call me a faggot or whatever kek

>>101744708
i said i converted to fp8 on load so the model is 12gb so vae fits fine and i can game at the same time

>>101744712
do you really think transferring data over a pci-e bus is faster than keeping it on the gpu?
Are you retarded?

File: BMP_25210_.png (2.93 MB, 1384x1384)

2.93 MB PNG

>>101744690
>tfw never going to the Isekai world
https://files.catbox.moe/tnzmay.webm
https://files.catbox.moe/b89s6y.webm

>>101744735
>Are you retarded?
I'm trying to help you and this is how you react? fuck off nigger, keep enjoying your 15s/it for what I care

File: ComfyUI_00147_.png (1.67 MB, 1152x832)

1.67 MB PNG

what upscaler are you all using for flux dev gens?

File: cry.png (1.31 MB, 768x1024)

1.31 MB PNG

>>101744746
yeah my 30s 50 step flux-1.dev is bad when i already tried the offload to cpu even though they even said you shouldn't on a 4090
>>101744739
gen for you

File: BMP_01475_.png (3.09 MB, 1432x1432)

3.09 MB PNG

>>101744761
Wonderful. I love it!

File: BMP_25261_.png (2.31 MB, 1384x1384)

2.31 MB PNG

More cloud gens

File: doublekek.png (850 KB, 768x1024)

850 KB PNG

oh sorry that gen was 24 seconds, and this gen was 24s too

File: ComfyUI_temp_00018_.png (957 KB, 768x1024)

957 KB PNG

>>101744813

File: eval node.png (41 KB, 872x1045)

41 KB PNG

>>101744504
>>101744468
just make this kind of a node and you can do all sorts of manipulations as needed, no need for specialized nodes for every little thing.

someone may have made this sort of a thing somewhere but it's trivially easy to write yourself.

>>101738681
I can train a 1000-image SDXL Lora for 40 epochs at 1 repeat / batch size 4 for less than $5 in Buzz though, which may take like more than 8 hours but it doesn't matter because you pay the Buzz before the run starts based on their calculations.

>>101744901
if you dont have good negatives you shouldn't train a lora

>>101738717
Not really, literally the guy behind SimpleTuner initially was super pessimistic on Github that it was viable himself

>>101740572
I think he's testing AuraFlow instead

i mean you can hook into their model, it is just math.
so make a lora that can work it with already if you feel so strongly

File: delux_cos_00027_.png (684 KB, 1216x832)

684 KB PNG

>>101744923
>it is just math.
if you think math is so easy, name five of their albums

File: 1722917191021_out-0.png (902 KB, 1088x896)

902 KB PNG

Quokkapool!

File: 5.png (1.09 MB, 768x1024)

1.09 MB PNG

>>101744946
named

>>101744979
>>101744979
>>101744979

>>101745059
jesus an ant could render that

>>101738194
HOW DID HE DO THAT? LOOK AT THE PIZZA, IT ISN'T MISSING PIECE. sus

>>101745059
you can use an esgan model to upscale your low renders so people won't make as much fun of you.
Literally just edit the image output after the vae

>>101745072
>>101745088
you guys are actually retarded