File: fSDG_News_000026_.jpg (506 KB, 896x512)

506 KB JPG

>mfw Resource news

08/19/2024

>Safetensor UNet Loader / Merge Plugin for AUTOMATIC1111 Stable Diffusion Web UI
https://github.com/captainzero93/load-extracted-unet-automatic1111

>Generative Dataset Distillation Based on Diffusion Model
https://github.com/Guang000/BANKO

>Efficient Image-to-Label Diffusion Classifier for Adversarial Robustness
https://github.com/hfmei/IDC

08/18/2024

>AI + a16z: The Researcher to Founder Journey ft. Black Forest Labs founders
https://podcasts.apple.com/au/podcast/the-researcher-to-founder-journey-and-the/id1740178076?i=1000665678592

>FLUX.1-schnell-training-adapter: Train LoRAs directly on schnell
https://huggingface.co/ostris/FLUX.1-schnell-training-adapter

>XLabs-AI / flux-controlnet-canny-v3
https://huggingface.co/XLabs-AI/flux-controlnet-canny-v3

08/17/2024

>FastSD CPU v1.0.0-beta.36 release with FLUX.1-schnell OpenVINO support
https://github.com/rupeshs/fastsdcpu/releases/tag/v1.0.0-beta.36

>UNet Extractor and Remover for Stable Diffusion 1.5, SDXL, and FLUX
https://github.com/captainzero93/extract-unet-safetensor

08/16/2024

>SLCA++: Unleash the Power of Sequential Fine-tuning for Continual Learning with Pre-training
https://github.com/GengDavid/SLCA

>HAIR: Hypernetworks-based All-in-One Image Restoration
https://github.com/toummHus/HAIR

>bigdata-pw-Dinosaurs: 321 Dinosaurs
https://huggingface.co/datasets/bigdata-pw/Dinosaurs

>IVE: Towards Descriptive and Diverse Visual Commonsense Generation
https://github.com/Park-ing-lot/DIVE

>Civitai: Quickstart Guide to Flux.1
https://education.civitai.com/quickstart-guide-to-flux-1/

>Replicate: Fine-tune FLUX.1 with your own images
https://replicate.com/blog/fine-tune-flux

>AI-powered ‘undressing’ websites are getting sued
https://www.theverge.com/2024/8/16/24221651/ai-deepfake-nude-undressing-websites-lawsuit-sanfrancisco

>ComfyUI RyanOnTheInside Node Pack
https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside

>mfw Research news

08/19/2024

>xGen-MM (BLIP-3): A Family of Open Large Multimodal Models
https://arxiv.org/abs/2408.08872

>PFDiff: Training-free Acceleration of Diffusion Models through the Gradient Guidance of Past and Future
https://arxiv.org/abs/2408.08822

>Comparative Analysis of Generative Models: Enhancing Image Synthesis with VAEs, GANs, and Stable Diffusion
https://arxiv.org/abs/2408.08751

>Task-Aware Dynamic Transformer for Efficient Arbitrary-Scale Image Super-Resolution
https://arxiv.org/abs/2408.08736

>Adaptive Layer Selection for Efficient Vision Transformer Fine-Tuning
https://arxiv.org/abs/2408.08670

>SketchRef: A Benchmark Dataset and Evaluation Metrics for Automated Sketch Synthesis
https://arxiv.org/abs/2408.08623

>A New Chinese Landscape Paintings Generation Model based on Stable Diffusion using DreamBooth
https://arxiv.org/abs/2408.08561

>Achieving Complex Image Edits via Function Aggregation with Diffusion Models
https://arxiv.org/abs/2408.08495

>TEXTOC: Text-driven Object-Centric Style Transfer
https://arxiv.org/abs/2408.08461

>JPEG-LM: LLMs as Image Generators with Canonical Codec Representations
https://arxiv.org/abs/2408.08459

>PQV-Mobile: A Combined Pruning and Quantization Toolkit to Optimize Vision Transformers for Mobile Applications
https://arxiv.org/abs/2408.08437

>CT4D: Consistent Text-to-4D Generation with Animatable Meshes
https://arxiv.org/abs/2408.08342

>TurboEdit: Instant text-based image editing
https://betterze.github.io/TurboEdit/

>Segment Anything for Videos: A Systematic Survey
https://arxiv.org/abs/2408.08315

08/18/2024

>Survey: Transformer-based Models in Data Modality Conversion
https://arxiv.org/abs/2408.04723

>Weak-Annotation of HAR Datasets using Vision Foundation Models
https://arxiv.org/abs/2408.05169

>Scene123: One Prompt to 3D Scene Generation via Video-Assisted and Consistency-Enhanced MAE
https://yiyingyang12.github.io/Scene123.github.io/

>>101970411
Why is it still called Stable Diffusion General if nobody here uses SD anymore?

File: 1703704235041707.png (3.33 MB, 1920x1152)

3.33 MB PNG

>>101970828
I do

>>101970843
Why not Flux for something like this? Is there any point to SD anymore outside of porn?

File: 1709876330322600.png (3.03 MB, 1920x1088)

3.03 MB PNG

>>101970852
My 12GB video card can't run Flux efficiently enough to optimize the styles I want.

File: image_2024-08-19_094338989.png (1.25 MB, 944x1200)

1.25 MB PNG

>>101970828
i do mostly sdxl, but flux is diffusion too

>>101970863
Fair enough

File: 1692987324214609.png (3.03 MB, 1920x1088)

3.03 MB PNG

>>101970923
It's not fair at all, actually

File: 1700261113972839.png (3.14 MB, 1920x1088)

3.14 MB PNG

>>101970950

>>101971000
These are cool

File: image_2024-08-19_100134438.png (1.66 MB, 1152x896)

1.66 MB PNG

not sure if flux is even better with such stuff than sdxl

File: 1717762233539873.png (3.02 MB, 1920x1088)

3.02 MB PNG

>>101971066
Danke
>>101971074
Yeah Flux is best with realism and fine details; your pic is a great illustration of that, has a lot of the essential composition and more accurately, but it looks like a hybrid between a photo and some high-def CGI.

File: image_2024-08-19_101504403.png (1.42 MB, 1152x896)

1.42 MB PNG

flux can do wall-e's tho

File: image_2024-08-19_102012255.png (1.1 MB, 1152x896)

1.1 MB PNG

flux can follow the prompt pretty nicely, thats really cool, i'm excited for models to come.

An unfocused blurry humanoid Zombie in the foreground

>>101971284
drink your own bathtub water

File: 1723490068029313.png (2.86 MB, 1920x1088)

2.86 MB PNG

>>101971132

File: image_2024-08-19_103828515.png (1.33 MB, 1152x896)

1.33 MB PNG

>>101970863
did you try flux tho?
my 10gb 3080 does fine with the NF4 version of flux-s. its not perfect, but you get nice stuff out of it

>>101971330
Looks like it could be video game cut scene
>>101971332
Some Beksinski vibes in this one

>>101971332
love some stuff in the closer foreground

>>101971366
using forge webui , it does a good job with memory management and splits to cpu if needed, its fast too 15sec for 1024x1024

File: 00016-3007466800.png (1.01 MB, 896x1152)

1.01 MB PNG

>>101971294
I tried to do it. Prompt comprehension was good at least.

File: 1704796489768042.png (2.81 MB, 1920x1088)

2.81 MB PNG

>>101971366
There's so many versions floating around now, I tried fp8 and got some cool pixel art with it. Which version specifically is that and is anything else required?
>>101971383
Yeah it's a fusion between a Beksinski lora, a Range Murata lora, and a proprietary jank lora that adds a little extra weirdness
>>101971389
Agreed, the details are where it really shines

>>101970828
Its a containment general

>>101971447
this here
Flux.1-Schnell BNB NF4
https://civitai.com/models/638187?modelVersionId=714460

its the schnell version (4 Steps) and 4 bit makes it small

>>101971493
as far as i understand most is 4 bit, but the most important parts are still higher precision

>>101971366
>10gb 3080
I'm using dev-fp8 with one. Almost 3 minutes per gen though.

>>101971531
i think dev is too big for us low ram plebs sadly, need to stick with 4-step S, which is not too different i think

>>101971566
I will try switching to Schnell and see how that goes.

>>101971433
bad fake
irl her pussy looks like a gernade went off in it

File: image_2024-08-19_112504844.png (1.6 MB, 1280x1024)

1.6 MB PNG

ankha is ready for the apocalypse

>>101970411
Man OP is beautiful
Why can't modern video games look as good

>>101971813
https://www.youtube.com/watch?v=90oVkISQot8

>>101971813
>>101971857
i hate those obnoxious screenshakes tho, trying to make it look like a realistic handycam.

File: image_2024-08-19_120541837.png (1.51 MB, 1280x920)

1.51 MB PNG

hmmm somethings missing, not sure what

flickr 456m, probably do another update later
bigstockimage33m hopefully finished writing to webdataset by tonight

>>101972161
the invisible woman with prosthetic tits
nice

>>101972341
this is awesome

>>101972351
what if thats a man wearing fake tits?
with a big fucking schlong and hairy balls and male pattern baldness

>>101972399
thanks!
i'm working on a series of articles about working with these datasets that will be published on hf's community blog. it should help people learn how to filter and process them because i get the impression most people who do training are used to just having a folder of images with text files for captions
that's probably because the training scripts don't support things like webdataset or hf's datasets library, so i'll be looking at integrating those with the popular training scripts
also i'm going to review and categorize loras/finetunes to identify what people are training already, then i can filter the large datasets myself to produce smaller more specific sets, and determine what kind of data i should be acquiring, for example if i find that training on anime characters is popular (it probably is) then i'll work on big collections of characters by extracting them from the frames in the animes or finding them at some other source

question for experts
>>101970583

File: 00008-[flux1-dev-bnb-nf4-(...).jpg (406 KB, 2304x1792)

406 KB JPG

Is there a trick to get better skin tones from Flux? Also, they all look 50 yo.

i miss debo

How long will /ic/ will be buck broken over AI?

>>101973378
Looks like the gen is burned. Try lowering your cfg.

File: file.png (4 KB, 279x106)

4 KB PNG

bigstockimage about 50%
tracks about 15% and the metadata acquisition is still in progress so there's more now
can you believe a certain undiosclosed music generation saas expected me to collaborate for free? i'm sure they already use copyright music but if they had what i have they wouldn't have bothered to reply. they'll just have to figure it all out themselves desu. it's not like i'm expecting a job or anywhere near what they'll be paying their existing engineers and researchers but it's taken a lot of research, time and resources to curate, it would cost me to transfer it to them, plus the legal risk considering what the dataset is which is why i'm not releasing publicly it in the first place. i just expect a mutually beneficial arrangement, quid pro quo, it's fair to say the potential improvement to their service with a better match of music to lyrics is worth millions of dollars in sales and i'd be able to do so much more with a bigger budget. they also flat out refused the possibility of confidentiality or non-disclosure agreements which would have afforded both parties much needed legal protection, so i don't mind sharing this here

>>101972592
i was thinking extracting frames from bluerays should be quite a nice source of images

>>101975006
yeah, it would be, i have a project in progress scraping biographies and photos of actors which i'll eventually use to build system for facial recognition to automate tagging of characters in extracted frames, it's up to ~4.2M actors atm, seems like a lot but it's covering multiple countries, decades of tv/film and all the different roles/characters

>>101975065
crazy but sounds fun

>>101975006
include the year in captions. would be cool to be able to type "movie from 1984" and actually get a specific quality from around that year

>>101970411
kino

>>101975159
i believe people already train with a random choice of multiple captions, i guess the choice happens once per epoch i'd have to check how the training scripts are working, but i think it would be useful to train on all the images with both specific and generic captions, with the generic caption being something like "movie from 1984", that way the both the characters/content are captured and the generic style

File: 1723997347621483.jpg (121 KB, 1020x1272)

121 KB JPG

>"stable diffusion general"
>it's all about flux

sdfg stable diffusion and flux general

>>101975315
thanks for the astute observation, autismo-kun. indeed not everything in life is literal. not to worry, this thread actually functions as a general for all kinds of image generation models, you can even post about GANs if you'd like

>>101975424
Wrong
Its the schizo containment thread, thats why everyone else left

>>101975451
yes, your posts are particularly annoying which explains why posters left to get away from you

File: image_2024-08-19_165810612.png (1.09 MB, 1024x1024)

1.09 MB PNG

flux pro from replica

>>101971493
You should use GGUF quants if you're going to use NF4. Q4_0 to be precise.

You new?

>>101975652
i tried q4_0 dev but got this error in forge

  File "D:\stable-diffusion-webui\backend\loader.py", line 59, in load_huggingface_component
    assert isinstance(state_dict, dict) and len(state_dict) > 16, 'You do not have CLIP state dict!'
AssertionError: You do not have CLIP state dict!
You do not have CLIP state dict!

>>101975687
the game is fucking called metal gear solid

>>101975451
Get a medication holy fucking shit or better yet, get mangled in a car accident. I beg you, please get killed.

>>101975720
Hi debo

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

>>101975822
Can you fuck off schizo anon, please?

>>101975869
>When in doubt, ask yourself; is this reply purposefully obtuse? is this reply off-topic? is this reply seething about this rentry?

>>101975869
>please
Delicious

File: image_2024-08-19_171641225.png (1.16 MB, 1344x768)

1.16 MB PNG

stop fighting or i send my girl over

>>101975932
Dont complain if you get her back in a schizophrenic state of mind

>almost 600 views in the rentry

how do you run flux gguf models on forge ? i always get

AssertionError: You do not have CLIP state dict!
You do not have CLIP state dict!

>brother you're shitting out some dumpster-grade cloud service gens. how tf am I supposed to know who you are

>>101976129
fucking kek

>please stop harassing me on an anonymous image board

>singular schizo cope

>nigbo

>>101972161
The annoying part

>>101976207
You are here every single fucking day you psycho

>>101976282
But enough about debo

>its one anon
Schizophrenic cope

File: delux_ci_00022_.png (2.28 MB, 1536x968)

2.28 MB PNG

>>101974977
>can you believe a certain undiosclosed music generation saas expected me to collaborate for free?
I guess they're dumping so much cash into their legal defenses against the record labels that they've got nothing left for contractors
>they also flat out refused the possibility of confidentiality or non-disclosure agreements
all their lawyers are too busy lol. but since they can't sign an NDA, you should just do it for resume fodder. one step closer to Chief Data Acquisitioner

>

File: delux_ci_00023_.png (1.72 MB, 1536x968)

1.72 MB PNG

>>101975315
we encourage all diffusion done in a stable fashion. meta-stable is ok too. no affiliation with the Stable Diffusion brand. common misconception

>>101975401
but then all the pixart/hunyaen people would whine, not to mention future models we don't even know about yet. idk why we need to pretend like the name matters.

>>101976446
i'll wait to see what the others say and play them off against each other desu

File: delux_ci_00024_.png (2.24 MB, 1536x968)

2.24 MB PNG

>>101975749
nta but good morning. what are you genning today?

>schizospamming

what was that point of attaching images to those posts?

>why are you posting images to an image board

igg image generation general
one family

>gm

>inb4 decoy purple schizo

File: delux_ci_00025_.png (2 MB, 1536x968)

2 MB PNG

>>101976633
just to highlight how stupid names are

/igg/ on-topic:
- local gen
- saas gen
- dalle
- twitter
- nai
/igg/ offtopic:
- training
- video gen
- dev work

no matter what name we use, things we want will be offtopic and things we don't want will be ontopic. the name just doesn't matter besides drawing arbitrary lines in the sand. a rose by any other name would smell just as sleep

I vote for staying here in /schizo debo general/

Cope

File: ComfyUI_00188_.png (1.41 MB, 1024x1024)

1.41 MB PNG

anyone using any of those flux merges out there? are they any good?
I'm wondering whether they get loaded and prompted on as flux or as sdxl

File: succ_0282.jpg (469 KB, 1664x2432)

469 KB JPG

I remember when I first started with SD when I would get body horror outputs it legitimately freaked me out. Now I run a tiled upscaler with denoising set to high and get something out of a horror movie and I barely flinch.

File: end_gen~15.jpg (269 KB, 1448x1280)

269 KB JPG

Nobody knows what pilaf is or what congress actually does

File: pilaf from dragon ball.png (3.16 MB, 2210x1674)

3.16 MB PNG

>>101977028
>Nobody knows what pilaf is
not wrong

Huh?

File: 2024-08-19_00228_.png (1.8 MB, 1024x1280)

1.8 MB PNG

>>101977028
pilaf is tasty, its just kinda fried rice but from india and surrounding countries, luckily I got many pajeet restaurants in my eu country so I can eat it sometimes

>>101977028
>congress
does nothing

>>101976872
no difference really

File: FLUX-00066-palette knife,(...).jpg (166 KB, 1170x800)

166 KB JPG

Morning anons

File: BMP_FLUX_07711_.png (738 KB, 1280x848)

738 KB PNG

>>101977028
Yeah I thought it was food

File: delux_ci_00026_.png (1.9 MB, 1536x968)

1.9 MB PNG

>>101977226
gm

Once again I have to complain, why are so many of these lora creator using AI pony images to train their lora!? WTF is wrong with them. The person even sounds like he's suprised that his lora looks like pony...

https://civitai.com/models/659930/standing-anal-hanging-suspended-congress?modelVersionId=738456

NSFW by the way

Debo killed my dog

>>101977343
buzz

>>101977343
>standing-anal-hanging-suspended-congress
What is wrong with you

>>101977365
I only clicked it because it looked like Pony and I wanted to judge.

>>101977343
there is no barrier to training on civit beyond whether you have the buzz or not. nothing is stopping any retard from clicking the train button so of course the end result is an ocean of slop

>>101977430
Sure

>>101977610
Come on trust another 4chan bro, I'm as innocent as you can get. For a 4chan anon that is.

>>101975424
>>101975466
>>101975720
>malding trannies

>this is nigbos only social contact

File: 00022-[fluxRealisticV1_v1(...).jpg (787 KB, 2688x3456)

787 KB JPG

Forge seems quite experimental. Is there a more stable branch?

>>101977864
how many times per week do you need to get a new IP, on average?

File: delux_ci_00028_.png (2.02 MB, 1536x968)

2.02 MB PNG

>>101977878
illy has pretty much always treated forge as experimental. I don't think he believes in stable releases

>>101977884
i dont get banned at all, the other schizofriends might sometimes

File: 00051-[epicrealismXL_v5Ul(...).jpg (806 KB, 2688x3456)

806 KB JPG

>>101977904
I'm not complaining for all the great features. I'm just getting confused with the VAE. There used to be an automatic option which worked great for me. Now it's gone and I'm not sure if it's me but the VAE don't seem to work.

>>101976565
Probably because otherwise someone would be like "you're a nogen. Your opinion doesn't matter"

>>101978209
>someone
I can think of only one poster who would care

File: 1724091134276_0.jpg (81 KB, 984x984)

81 KB JPG

>>101978169
Oh no Kitten glows...

>>101976989
Fk yea, handlebars!

File: ComfyUI_00207_.png (1.5 MB, 896x1088)

1.5 MB PNG

>>101978169

>>101978501
Bretty cool

File: PW_82800_.png (853 KB, 1024x768)

853 KB PNG

Good morning, anons! I hope everyone is doing well :]

File: delux_ci_00029_.png (1.82 MB, 1536x968)

1.82 MB PNG

>>101978692
hello, gm. I know she's supposed to be in bed but she looks like shes in a coffin, lol

>>101976714

File: PW_82798_.png (808 KB, 1024x768)

808 KB PNG

>>101978727
Morning, Debo! Great to see you again :]
LOOOL it kinda does huh haha!
Cool wizard! Really nice detail

dropped a few articles today
https://huggingface.co/blog/hlky/web-scraping-101
this one is particularly fun

File: 1724093570248_1.jpg (142 KB, 1170x800)

142 KB JPG

>>101978692
Morning PW

File: PW_82808_.png (806 KB, 1024x768)

806 KB PNG

>>101979018
Morning, Hlky! This looks interesting! I'll give it a read when I get back home tonight :D
>>101979030
Good morning, Quokkanon! It's great to see you again :]
Nice jacket haha!

File: 1724094244592_3.jpg (135 KB, 1170x800)

135 KB JPG

https://r2.fluxpro.art/cm017tjw80cds3blvx4t0m26c/3.webp

>>101979177
Kek i pasted the link

File: file.jpg (370 KB, 1792x1024)

370 KB JPG

>>101979105
morning! i'll probably drop stage 2 tomorrow, i'm really tired, i couldnt sleep last night until like half 3, ended up doing some jewelry stuff and chatting to chatgpt which gave me the idea to write these articles, check earlier in the thread for some of the other plans ive come up with
flickr is up to 466m but i'm going to wait until 500m for the next update

>>101971284
>>101971433
These are surprising good, what did you use to make them? Could yuo catbox one of them?

File: PW_82888_.png (942 KB, 1024x768)

942 KB PNG

>>101979270
Nice! I'll try to get caught up with stage 1 tonight then :]
Hahaha you definitely need some sleep! Coffee is a great idea if you don't tho
I'm drinking cold brew right now, I love coffee! I gotta get out of here in like 2 hours, probably gonna get ready in an hour or so after I fully wake up

>>101978727
>I know she's supposed to be in bed but she looks like shes in a coffin
sometimes the most enjoyable thing I get from these threads is your dry humor. Hopefully that doesn't sound like a slam, because it's not.

>>101979447
Thread schizo

>nigbo

File: file.jpg (366 KB, 1792x1024)

366 KB JPG

>>101979423
i always have coffee, or energy drinks. i'll probably get food then watch tv or play something on the ps5 until its a better time to sleep. it's only 8.30, napping in the evening yesterday is what messed up my sleep
have fun going out!
let me know what you think about the article when you get chance, and as usual let me know if there's any datasets, articles or code that you'd like to see!

>hlky needing coffee is the best meme of the summer

WHAT IS THE DIFFERENCE BETWEEN DISTILLED FCG SCALE AND REGULAR CFG SCALE IN FLUX

>>101970411
Looks amazing as a realistic Samus, can you Catbox it?

File: PW_82923_.png (835 KB, 1024x768)

835 KB PNG

>>101979601
That sounds like a good plan! Going to sleep early really is counterproductive sometimes haha
Thanks! Unfortunately it's just work, but I like my job so it's ok haha
Last day of the week though, so i'm happy about that :D
Will do!! I've been really motivated to learn more stuff lately
Actually this is my v2 of my LoRA that i'm using :]
The first one was 35 images, this one is over 200 or something like that haha
I think it only took a few hours or so!

All my pics in flux are like unfocused what is this

>>101978727
foreshadowing
a powerful concept

File: 00075-[fluxRealisticV1_v1(...).jpg (107 KB, 896x1152)

107 KB JPG

>>101978767

File: BMP_FLUX_07905_.png (1.64 MB, 1280x848)

1.64 MB PNG

Aftanoon anons that just got on

File: delux_ci_00030_.png (2.13 MB, 1536x968)

2.13 MB PNG

>>101979447
I sincerely appreciate that!

>>101979678
truth

>>101979726
illy gave a very vague description of distilled but it didn't really answer anything. I think its 'distilled' because it operates on both t5 and clip but not in a 1:1 way. idrk though

>>101979868
>pics in flux are like unfocused
I tried following this write-up to address blur/focus/detail in flux. I couldn't really get it to work as expected but maybe you'd have more luck with it:
https://www.reddit.com/r/StableDiffusion/comments/1estj69/remove_the_blur_on_photos_with_tonemap_an/

I've seen some other people talking up the tonemap approach so it seems to work for some people

>>101980100
Looks like its working with the dynamic threshold fix!! Thanks!

File: PW_82957_.png (824 KB, 1024x768)

824 KB PNG

>>101980081
Good afternoon, Mouse anon!! :]
>>101980100
My LoRA makes pretty good debos!
It doesn't like making him a guy tho most of the time hahaha

>>101977343
The real question is why are these people training concepts pony can do ootb, I've seen this numerous times now.

Just learn the booru tags ffs

File: delux_ci_00032_.png (1.8 MB, 1536x968)

1.8 MB PNG

>>101980202
>My LoRA makes pretty good debos!
is this still the 3d one or is this a different one?
>It doesn't like making him a guy tho
yeah I've had the problem a bunch of anime models, lol. very 1girl biased

File: PW_82946_.png (718 KB, 1024x768)

718 KB PNG

>>101980241
It is, but I trained it with more images :]
This one is over 200 while the first was only 35 haha

>want to animate a series of images into a single video
whats the go to for interpolating with ai video?

File: PW_82911_.png (833 KB, 1024x768)

833 KB PNG

Goodbye, anons! I hope to see you all later tonight :]

>>101980363
Take it easy dude!

File: delux_ci_00033_.png (2.03 MB, 1536x968)

2.03 MB PNG

>>101980363
have a good day

Look at this dude's heeeeeeeeeeeead
>>101980363
See ya later alligator

I hope ani is ok...

File: BMP_FLUX_07930_.png (1.61 MB, 1280x848)

1.61 MB PNG

Whoops forgot gen
>>101977322
>>101980485
These are cool

>>101970828
Comfy runs SD and Flux

File: file.jpg (344 KB, 1792x1024)

344 KB JPG

>>101979678
damn good coffee
>>101979796
nice work, looks great!

>>101980545
>feed me coffee or that bridge gets destroyed

File: 1724100343352_image.jpg (167 KB, 964x1446)

167 KB JPG

>>101970828
I do from time to time, it's good at art styles out of the box

File: BMP_FLUX_07938_.png (1.82 MB, 1280x848)

1.82 MB PNG

>>101980689
Wonder if the 'Zilla can do a sick kegstand

File: ComfyUI_00208_.png (1.64 MB, 1280x768)

1.64 MB PNG

>>101980713

File: image.png (604 KB, 1410x1040)

604 KB PNG

hello tech wizards, i downloaded stable diffusion, got the ui working

I dont have gfpgan
I dont have lora
I dont know how to import safetensor

are the results supposed to be this bad, and how can i fix it?
how can i import https://civitai.com/models/562557?modelVersionId=626703 and are there any dependancies/prior steps?

thank you very much for any help

File: BMP_FLUX_07977_.png (1.29 MB, 1280x848)

1.29 MB PNG

>>101981029
Nope, he knows how to party though
>>101981029
Purp'zilla

Thank you debo for the fix for blurry images

>>101981347
thanks thread schizo for continually making this general worse

>>101981417
It's annoying when debo posts in third person, isn't it?

"baker anon" listened and removed the discord?

>>101981951
you post this slop 20 hours a day on purpose, but for what purpose?

>>101981998
it would be difficult to post it accidentally

Align your steps GITS
whats special about those GITS 11 and 32 variants ?
https://github.com/lllyasviel/stable-diffusion-webui-forge/commit/9f5a27ca4ed28261734f10b0706e9221f35051a5

File: BMP_FLUX_08039_.png (1.3 MB, 1280x848)

1.3 MB PNG

I have a sudden urge to play DDR/Musica 1/2. I miss Round 1 so much

>>101982120
>whats special about those GITS 11 and 32 variants ?
The GITS (Get It Together Steps) 11 and 32 variants are special because they represent specific configurations of the sigma values used in the scheduling process for generating images in the context of diffusion models, particularly in the Stable Diffusion framework.


GITS 11 Steps: This variant uses a specific set of 11 sigma values that are designed to optimize the image generation process over 11 steps. The values are carefully chosen to balance the trade-off between image quality and computational efficiency. The interpolation method used ensures that the sigma values decrease in a controlled manner, which can lead to better convergence and more coherent images.


GITS 32 Steps: Similarly, the GITS 32 variant employs a more extensive set of 32 sigma values. This allows for a finer granularity in the diffusion process, potentially leading to higher quality images. The increased number of steps can capture more details and nuances in the image generation process, making it suitable for more complex tasks or higher fidelity outputs.


Both variants utilize log-linear interpolation to adjust the sigma values based on the desired number of steps, ensuring that the transition between values is smooth and effective for the model's performance. The choice between using 11 or 32 steps typically depends on the specific requirements of the image generation task, such as the desired quality and the computational resources available.

>>101982162
it can be used with Euler a or SDE it seems, do 11 SDE steps also match the 11 version ?

since SDE usually needs like half the step count i mean

>>101982146
how do I unsubscribe from Troon News?

>>101982316
>/ldg/

>>101982316
>debo still salty about being ignored by bmp and catjak

File: 000000_16575_.png (2.34 MB, 1075x1434)

2.34 MB PNG

>>101971447
>There's so many versions floating around now

ComfyUI, GGUF,
https://huggingface.co/city96/FLUX.1-dev-gguf/tree/main
These are supposedly faster

>>101971813
>Man OP is beautiful
TY, go back a thread or two I genned a few Samus',, All done with Flux Dev GGUF (Flux1_dev-Q8_0)

File: delux_ci_00038_.png (2.29 MB, 1536x968)

2.29 MB PNG

>>101982402
I'm salty about bmp ignoring me because I think he's preddy cool and we like a lot of the same things. I think he's too committed to the bit to give me a fair shake. on the other hand, I would be immensely ecstatic if ran ignored me instead of endlessly obsessing over me. so, wrong on both counts. I know you know the lore better than this. I'm disappointed in you

>>101982462
are you trying to gen a metroid? thats pretty cool but this insides need to be more nucleus/organ-like, I think. plus gotta get the gnarly fangs in there too. still sick tho

>>101982500
>>101982500
>are you trying to gen a metroid?
Yes, it's difficult with no Loras / contronets. Just got home, so I'll try a bit tonight.

>same cope
>for over a year straight

>posts hand with an extra finger

>>101982590
you don't want to know what she does with the extra finger

i have 6 (six) (1, 2, 3, 4, 5, 6) followers on hf

File: BMP_FLUX_08147_.png (1.75 MB, 1280x762)

1.75 MB PNG

>>101982722
On what now? I hit 400 on pixiv last month but I haven't posted in a while. Also unpopular opinion pizza tastes better when square sliced, don't ask me how it just does.

>>101975315
Embrace the future, old man

File: huggingface.png (58 KB, 1024x941)

58 KB PNG

>>101982814
hugging face
strong agree on square pizza btw

is there a 1000 foot overview of the image gen 'scene'? i played around with SD on automatic111 way back when (2 years ago?). there's so much stuff that has happened. It would be cool if someone was writing a small book or something -- not necessarily about how to use the tools to generate art, but the evolution and history of what is going on. ie, people using loras (what is that?), control nets (what is that?), which new models came and went (SDXL, SD3), private models (dall-e).

just like, collected information in a somewhat cohesive way, and not a training manual. maybe the scene moves too fast for this to be a thing.

File: BMP_FLUX_08151_.png (1.53 MB, 648x1200)

1.53 MB PNG

>>101982842
I just released after I posed, I am big baka

File: Untitled.png (1.78 MB, 1024x1024)

1.78 MB PNG

>>101982995

>>101982859
too much for a quick post

>>101983268
yea, was hoping someone was compiling something larger like a wiki or even a book, but i guess not

>>101978564
Radium kittens

File: 000000_16584_.png (1.96 MB, 1434x1165)

1.96 MB PNG

>need denticles

>>101981055
safetensor is just a filetype used for both models and loras. For models, like the one you linked, just download it and put in your models/stablediffusion folder.
>are the results supposed to be this bad
You're using the base model and no negative prompt, so you definitely could improve it. But SD will probably struggle with "fish with dog head" because it doesn't handle multiple concepts well and I'm sure has no knowledge of that concept. Flux might do better.

File: 1719850304991200.jpg (531 KB, 2496x3648)

531 KB JPG

any of you anons have more of pic related? is it even SD?
t. lurking retard

File: 1710517775812544.jpg (89 KB, 984x984)

89 KB JPG

>>101970411
Time for a stupid question: what are the absolute minimum VRAM requirements for running Flux?

>>101984560
It looks like Pony. I can tell, because it makes those lines on the face when you use the tag cyborg.
>>101984638
I think 8. There are some smaller versions
https://huggingface.co/RichardErkhov/chanwit_-_flux-7b-v0.2-gguf/tree/main

>>101984638
you can maybe fit the smallest quant into 4gb

>>101984699
>>101984638
I think Q_4 is good for lower vram

>>101984720
>>101984722
Nta. Are there any ways you can force forge webui to split the model ways in between the GPU and CPU? I have a 16 GB GPU but I don't think that's enough to run Flux. Won't it crash if I try to run it?

>>101984760
Forge does have the "swap method" toggle button, but I've never tried to do that
>Then you want to move the model to GPU. But the model is 12GB, bigger than your GPU memory 8GB. Then how? The anwser is to split the model to two parts. One part to GPU, one other to a "swap" location. If you select CPU as swap location, your model will load to CPU memory and GPU
https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/981
It won't crash, you'll just get an error

>>101983132
which artist did you train on to get this

File: 00010-82470249-Australian(...).png (1.1 MB, 816x1192)

1.1 MB PNG

After many failed attempts, i finally did it.
I was able to run FLUX on Forge!

File: delux_po_00040_.png (1.99 MB, 1024x1344)

1.99 MB PNG

>>101985176
are you using a low quant gguf model? if not, that frees up a lot of vram. also, you can try swapping out your t5 encoder for even more vram:
https://huggingface.co/city96/t5-v1_1-xxl-encoder-gguf/tree/main

but if you've already got it where you like it, congrats! how long does each gen take for you?

File: tmpjbzcgwdu.png (1.27 MB, 1152x896)

1.27 MB PNG

Headphones is one thing. Hands on headphones is another. Any ideas?

File: 00011-82470249-portrait_p(...).png (785 KB, 984x984)

785 KB PNG

>>101985223
>Gguf model?
Yes someone recommend me to use FLUX fusion (Dev and Schnell merge) at 8 steps
>Swapping out your t5 encoder:
Which one of those should I get if I'm using Q4 0?
>>101985300
Try hands on ears

File: 00007-1875872791-An_oil_p(...).png (1.45 MB, 816x1192)

1.45 MB PNG

>>101985223
>How long does each gen take for you.
A 984x984 image like >>101985472 took around 4 mintues and 15 secs, with 30s/it

File: delux_po_00039_.png (1.83 MB, 1024x1344)

1.83 MB PNG

>>101985300
>hands on head
>hands holding head
>hands pressing headphones
just keep trying diff variations and weights til something hits

>>101985472
>Which one of those should I get if I'm using Q4 0?
idk :(

>>101985486
oh shit he did it

File: tmph0u65qdb.png (847 KB, 768x768)

847 KB PNG

>>101985472
Doesn't seem to be working out that well

In other news, I tried to get Hatsune Miku with a musical note and got a foot pic.

https://litter.catbox.moe/gtaqp9.png

Captcha: MADP

>>101985472
>Which one of those should I get
>It's therefore recommended to use **Q5_K_M or larger** for the best results
That's all the readme file says

>>101985548
Ty, Gn Anons, kek //need a 4090!

File: delux_po_00067_.png (2.2 MB, 1024x1344)

2.2 MB PNG

>>101985604
gn

File: 00002-873418332-quokka_wa(...).png (1.23 MB, 984x984)

1.23 MB PNG

>>101985604
Gn anon

>>101985597
Well, also
>smaller models may also still provide decent results in resource constrained scenaruos.
It sounds like the less resources you have the smaller model you should try. He misspelled scenarios, not me.

File: 00032-2481129750.png (1.15 MB, 668x1184)

1.15 MB PNG

>people trying to make me "buy" their lora with buzz

File: delux_po_00066_.png (1.77 MB, 1024x1344)

1.77 MB PNG

>>101985741
thats a thing now? civit is such a meme

>>101985763
Yeah, people have been doing it with Flux loras
https://civitai.com/models/664172/n64-game-style-f1d?modelVersionId=743275
425 buzz to buy, or they have a donation goal of 64,420 and will make it public

File: BMP_FLUX_08323_.png (871 KB, 1024x1200)

871 KB PNG

>>101985014
Buy me a coffee prompt in Flux

File: file.png (150 KB, 1912x514)

150 KB PNG

>>101985597
i downloaded Q5_K_M and put it on the text encoder folder but it's not showing up on the UI, do i need to put it somewhere else?

File: delux_po_00063_.png (1.89 MB, 1024x1344)

1.89 MB PNG

>>101986036
maybe forge can't recognize gguf files as encoders yet (because its never been done before)

File: Screenshot 2024-08-19 230047.png (9 KB, 172x167)

9 KB PNG

>>101986036
Yeah, it might be that Forge just doesn't recognize the file yet. I think the text encoders are new. Hopefully he will update. Weirdly it doesn't see my Flux vae, just the Pony one.

>>101986036
Oh, but also make sure you restarted after you added it in case you didn't.

>>101986089
from last i checked the command line option to select a vae folder is broken, so your vae must be in the forge /models/vae folder and not somewhere else
i dont think that's been fixed yet
try putting the text encoder there too

File: ComfyUI_00299_.png (1.76 MB, 1024x1208)

1.76 MB PNG

I'm getting some blurriness that I just can't seem to get rid of.

Is there a way to sharpen up my gens?

File: 00004-3086111536-An_oil_p(...).png (1.02 MB, 816x1192)

1.02 MB PNG

>>101986173
Which model, anon?

>>101986178
Sorry Using Autism Mix and the ponyXL.

>>101986118
>/models/vae
That is where I put it. I copied it into the text encoder folder just to see also. I can run dev and schell without it but end up getting an error with some of the other Flux variant models because they need the vae set.

>>101986173
Try a different sampler. Are you upscaling/using hires fix? If so, don't use latent upscaler it causes blurriness with a lot of models.

>>101986216
im using comfy UI

File: 1704211681052801.png (3.35 MB, 1920x1088)

3.35 MB PNG

>>101971493
>>101975652
Surprised the thread is still up, thanks, will give it a shot

https://civitai.com/models/656083/copycat-flux-testfp8fp16?modelVersionId=744540
>Trigger Words 1girl
base

File: BMP_29061_.png (1.75 MB, 1064x1128)

1.75 MB PNG

>>101986601
>Yuck, pseudo anime style!
What does that even mean?

>>101986976
I guess he thinks it doesn't look like real anime. It does kind of look like a 1.5 model

>>101987064
Hmmm weird. I thought the catgril in the mountains looked pretty decent. I just noticed after I posted I'd call that "pseudo anime" more than anything.

>>101987064
Definitely higher quality and detail. Kind of interesting that it could carry over that 1.5 look. I was originally just joking that 1girl prompting was back, but I might download it.

>>101987208
meant to quote this one >>101987084

>>101987285
>>101987285
>>101987285

>>101984760
>I have a 16 GB GPU but I don't think that's enough to run Flux. Won't it crash if I try to run it?
Kek, what a retard. I'm running Flux on 12GB right now, no problems.