File: FDG_News_00009_.jpg (971 KB, 1344x960)

971 KB JPG

>mfw Resource news

08/04/2024

>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

08/02/2024

>Optimizing Diffusion Models for Joint Trajectory Prediction and Controllable Generation
https://yixiaowang7.github.io/OptTrajDiff_Page

>UniTalker: Scaling up Audio-Driven 3D Facial Animation through A Unified Model
https://github.com/X-niper/UniTalker

>Smoothed Energy Guidance for SDXL
https://github.com/SusungHong/SEG-SDXL

>Mitigating Multilingual Hallucination in Large VLMs
https://github.com/ssmisya/MHR

>GalleryGPT: Analyzing Paintings with Large Multimodal Models
https://github.com/steven640pixel/GalleryGPT

>The Manga Whisperer: Automatically Generating Transcriptions for Comics
https://github.com/ragavsachdeva/magi

08/01/2024

>Stable Fast 3D: Rapid 3D Asset Generation From Single Images
https://stability.ai/news/introducing-stable-fast-3d

>mfw Research news

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

>More precise edge detections
https://arxiv.org/abs/2407.19992

>From Flat to Spatial: Comparison of 4 methods constructing 3D, 2 and 1/2D Models from 2D Plans with neural networks
https://arxiv.org/abs/2407.19970

>FedDEO: Description-Enhanced One-Shot Federated Learning with Diffusion Models
https://arxiv.org/abs/2407.19953

>FiCo-ITR: bridging fine-grained and coarse-grained image-text retrieval for comparative performance analysis
https://arxiv.org/abs/2407.20114

>OmniBal: Towards Fast Instruct-tuning for Vision-Language Models via Omniverse Computation Balance
https://arxiv.org/abs/2407.20761

>DMESA: Densely Matching Everything by Segmenting Anything
https://arxiv.org/abs/2408.00279

>Resilience and Security of Deep Neural Networks Against Intentional and Unintentional Perturbations: Survey and Research Challenges
https://arxiv.org/abs/2408.00193

08/03/2024

>Image Super-Resolution with Taylor Expansion Approximation and Large Field Reception
https://arxiv.org/abs/2408.00470

>Localized Gaussian Splatting Editing with Contextual Awareness
https://arxiv.org/abs/2408.00083

>Hierarchical Conditioning of Diffusion Models Using Tree-of-Life for Studying Species Evolution
https://arxiv.org/abs/2408.00160

>SynthVLM: High-Efficiency and High-Quality Synthetic Data for Vision Language Models
https://arxiv.org/abs/2407.20756

>Real Face Video Animation Platform
https://arxiv.org/abs/2407.18955

Fent OD arc when?

File: 02 (90).jpg (721 KB, 2000x2500)

721 KB JPG

>>101721011
does anyone have a working colab notebook for fine tuning sd 1.5 or sdxl?

Thanks!

Kohyas are deprecated or have package issues.

File: 1722697224915596.jpg (43 KB, 564x564)

43 KB JPG

Hi, newbie here. I'm starting again on SDXL using PonyXL model (using autism finetune really, but yeah). I have a 3060 with 12vram to use, right now I'm using these flags: "--xformers --cuda-stream --pin-shared-memory --cuda-malloc"

it works right, I was just wondering, if I should change something about if it works it just works?

i miss tranis depression posting
cant wait until he's fired (again) kek

File: delux_ggf_00015_.png (1.13 MB, 1152x896)

1.13 MB PNG

>>101721588
at this point, he's made so much money and so many connections that he prob wouldn't care less if sai went under

File: 000000_15843_.png (2.6 MB, 1024x1536)

2.6 MB PNG

>>101720862
YW Anon,

File: 00036-2842372522.jpg (646 KB, 1248x1864)

646 KB JPG

7 minutes b4 noon, goo morning.
thinking of buzzing off all my hair again
what samplers/schedulers are yall using? tech question

File: ComfyUI_01740_.png (1.08 MB, 1152x768)

1.08 MB PNG

>>101721437
>automatic
oh boy he doesn't know

>>101721645
thanks for reminding me to fill my invoice

>>101721925
>>101721167

>n-notice me
lolcow behavior

average gen time has inexplicably risen by 35%
not sure what I changed, if anything

https://github.com/intel/AI-Playground

intel won..

File: delux_ggf_00018_.png (1.03 MB, 1152x896)

1.03 MB PNG

>>101722091
old news and went completely ignored

>>101721925
pedo

File: 2024-08-04_00587_.png (1.86 MB, 1280x1280)

1.86 MB PNG

>>101722091
meanwhile
>Intel to lay off 15,000
>https://techcrunch.com/2024/08/01/intel-to-lay-off-15000-employees/

I am not so sure if intel is winning these days.

File: GS77VUQW8AAs9d_.jpg (239 KB, 928x1232)

239 KB JPG

a.i. pics(by others) you think are especially neat?
by rogerhaus on xwitter

File: cyborg001.png (1.26 MB, 1024x1024)

1.26 MB PNG

sd

File: delux_chi_00008_.jpg (209 KB, 1024x1024)

209 KB JPG

>>101722091
>intel won..

how are you anons coping with the demise of your beloved SDG?

are you having fun in the doxcord? LDG is thriving btw..

File: schFLUX_00011-Minecraft V(...).png (856 KB, 1088x896)

856 KB PNG

Morning anons

File: angery.jpg (1015 KB, 2048x2048)

1015 KB JPG

>>101722300
Morning

File: delux_ggf_00022_.png (1.13 MB, 896x1152)

1.13 MB PNG

>>101722300
gm

File: cyborg002.jpg (430 KB, 1879x1513)

430 KB JPG

sd!!

>>101722274
people will be making bank on the cloudstrike dip too

>>101721250
Pick small subset of your data and train a low dim LoRA. Pick images from one photoshoot and tag.

So why does default SD suck ass at prompt adherence compared to something like Dall-e or Flux? Is it just tags?

>>101722479
CLIP

Free replicate api key to use flux :D
r8_7bbhIYeK4NEmCUPa7SufxaUqCFbQGZ10ow8SG

>>101722708
Nope

File: delux_ggf_00024_.png (1.16 MB, 896x1152)

1.16 MB PNG

>>101722708
get this to webapp anon. I think he ran out of keys

File: cyborg005.jpg (181 KB, 768x1248)

181 KB JPG

sd<3

File: delux_ggf_00025_.png (1.13 MB, 896x1152)

1.13 MB PNG

>>101723123
when your goth gf grows up and gets her goth law degree

>model recommends clip skip 2
>test it
>images are identical

File: freeplayturkey.jpg (165 KB, 640x1248)

165 KB JPG

sd is free

File: FLUX__00131_.png (1.07 MB, 1024x1024)

1.07 MB PNG

fucking hands man

File: 1715962427359627.png (1.38 MB, 1024x1024)

1.38 MB PNG

reminder if you set the weight to fp8 (23gb unet model) and fp8 clip model, it's much faster, only 4090s should be using fp16 imo

quality is still good, have a gundam.

>>101723448
define quality

>>101723448
I'm on 3060 12GB and 64GB RAM, t5xxl_fp8_e4m3fn model doesn't give me absolutely any increase in speed, it just takes a little less RAM than t5xxl_fp16.

File: 1722799231470_image.jpg (71 KB, 984x984)

71 KB JPG

>>101723364
Neat

>>101723501
there isn't a huge difference, i've tried fp16 and it's just really slow cause I have a 4080/32gb and fp16 wants a 4090/64gb RAM essentially.

*correction, only change the unet weight to fp8, im using fp16 for clip (the 9 gig model) and it works well. checkpoint weight should be fp16 for 24GB vram only or it's slow.

>>101723514
but same quality?

>>101723538
>>101723541
what metric is quality?

>>101723541
t5xxl_fp16 pictures look better to me.

>>101723565
whatever gets my dick hard

File: delux_ggf_00027_.png (1.19 MB, 896x1152)

1.19 MB PNG

>>101723545
does a pretty good job of avoiding sameface

File: 1700433958098678.png (1.07 MB, 1024x1024)

1.07 MB PNG

>>101723565
well with fp8 weight dtype and t5xxl_fp16 clip, the text is still super clear when you generate. ive tried fp16 weight with a 4080 it just takes a lot longer to generate as I dont have the VRAM recommended (24, have 16).

example:

File: 1698801442089391.png (1.15 MB, 1024x1024)

1.15 MB PNG

>>101723448
mobile suit gundam: the witch from mercury: Miku edition

>>101723591
Yep, somewhat weird that it also has consistently same faces when minimally altering tags.

File: 1722800373692_image.jpg (69 KB, 984x984)

69 KB JPG

>>101723703
kek!
Is that actually flux?

>>101723751
nope, it is deep fake. Don't believe everything you see

File: freeplayturkey2.jpg (445 KB, 1024x1024)

445 KB JPG

>>101723516
You too, zen round squirrel

>>101723751
yep.

>>101723593

this for example had this prompt: Bloomberg News breaking news broadcast with Donald Trump pointing at Miku Hatsune and gives her a handshake. Miku smiles. They are at an electronic music concert with bright neon lights and overhead lighting. The text on the bottom of the screen says "BREAKING NEWS", and below it "ANIME WINS". Below that text it says "Vocaloid makes America great again".

File: 1715508623702728.png (1.25 MB, 1024x1024)

1.25 MB PNG

>>101723629

How much VRAM do you need to run Flux? I've only got a 1080 with 8 GB of VRAM. Is Flux out for me?

>>101723884
You're looking at extremely long gen times anon. Multiple minutes at the least.

File: 1713231408253798.png (1.18 MB, 1024x1024)

1.18 MB PNG

>>101723873

File: freeplaysatoshi.jpg (422 KB, 1024x1024)

422 KB JPG

>Satoshi!...Wait, wait, wait. Are you kidding me?

File: file.jpg (284 KB, 1792x1024)

284 KB JPG

>>> IMAGES.estimated_document_count()
933816669

nearly finished, only a few days left
1.2B including the other sets

File: delux_ad_00003_.png (359 KB, 896x1152)

359 KB PNG

huh, this is interesting. I'm trying to make some fake amazon store pages and they're all coming out blurred

File: delux_ad_00004_.png (849 KB, 896x1152)

849 KB PNG

ok, it works sometimes

In terms of img2img, controlnet, ipadapter I know img2img+controlnet work fine together where controlnet more influences the form but img2img influences the colors

Does ipadapter just work all alongside that or does it slot in better in the place of img2img or controlnet in that equation?

File: delux_ad_00010_.png (1.39 MB, 1344x768)

1.39 MB PNG

totally weird prompt

File: FLUX_928134127923501_00001_.png (757 KB, 768x768)

757 KB PNG

FLUX is very impressive in some ways. This isn't cherry-picked, it's typical. Hands actually aren't perfect, and there's a hand where there shouldn't be one, but wow are they better than models we've seen. Every person in this photo has their own face, their own clothes. Belts have buckles, jackets have zippers, background furniture is somewhat coherent and can be made sense of.

The mistakes are an order of magnitude smaller and less frequent than with SD3. And in spite of that it shows signs of flexibility and breadth.

I'm going to put it through more challenges to see where it falls short; predictably, it can't do the style of a JAV dvd cover, whereas SD1.5 could. But that's not surprising.

File: 1716519037066661.png (1.4 MB, 1024x1024)

1.4 MB PNG

If the model can make Gundams it can make anything

>>101724545
as far as I know ipadapter acts like a style transfer, you get something similar but different and you have to adjust control weights to see how much it affects an image

if you want a new image exactly like the original in terms of lineart, use controlnet canny or depth, along with openpose for ensuring the character pose is the same. im still not 100% on how ipadapter works though.

File: ComfyUI_00070_.png (1.37 MB, 904x1024)

1.37 MB PNG

>>101725472
Mmmm industry.

File: ComfyUI_00037_.png (1.71 MB, 1024x1024)

1.71 MB PNG

>>101725501
pollute the world

>>101724871
Cheers, working on a streamline interface for SD so interested how people actually use the 3 together

File: ComfyUI_00061_.png (1.11 MB, 904x1024)

1.11 MB PNG

>>101725517
Got a sick ramp for trains here.

File: file.jpg (290 KB, 1024x1792)

290 KB JPG

i have all the loras

>>101721011
Was told to repost this from ldg
Been running into all sorts of issues while trying to get sd working locally on my machine (AMD/Linux). The guide on the stable-diffusion-webui-amdgpu repo linked to in the rentry up above has gotten me the closest so far, and while the webui does now install/launch without errors, I can't seem to get it to actually start generating anything. The buttons in the webui seem to not work and there's no new output in the console after start up. Any ideas what might be causing this?
inb4 no model, this guide installed v1-5-pruned-emaonly.safetensors for me, but I also tried two other models from civitai without any luck.

File: de_fl_00122_.jpg (498 KB, 1344x960)

498 KB JPG

>>101725581

>>101725730
can you show the output from your cli

File: Flux_00039_.png (1.23 MB, 904x1104)

1.23 MB PNG

>>101725692

File: file.jpg (372 KB, 1024x1792)

372 KB JPG

>>101725733
167k of them, 18tb
if anything gets deleted i probably have it

So im assuming flux is the big dick on the block now, can we make our own loras and use them together?

File: 1714619355686760.png (1.21 MB, 1024x1024)

1.21 MB PNG

open source won, dall-e wont let you prompt this.

WE WON bros.

>>101725959
>dixar
kekkkk

File: delux_ggf_00029_.png (1.19 MB, 896x1152)

1.19 MB PNG

>>101725903
>flux is the big dick on the block now
yes, for now
>can we make our own loras and use them together?
information is conflicting. some say its possible, others say its not. if possible, it may be too expensive/demanding to be accessible to most

>>101724779
It can't do retro 90s video game styles at all

File: file.jpg (231 KB, 1024x1792)

231 KB JPG

i can see an open position at Meta that i'm pretty sure is for AITemplate. they still havent even sent me a rejection email for a different position i applied for
fuck those guys desu, i'm waiting on github to defork, might change the license too so they can't use it

>>101725959
wouldn't 'the border explorer' be more topical though anon?

Does anyone know where to get the new model of flux but through a torrent?

File: Disgusting criminal.jpg (182 KB, 1536x1536)

182 KB JPG

>>101725773
Thanks, officer. Thanks to your efforts, this disgusting criminal is off the streets.

So does this mean we will be seeing full comics with stories soon?

File: 00057.png (1.04 MB, 1024x1024)

1.04 MB PNG

>>101726048
>they still havent even sent me a rejection email
I usually don't even get one either. One place sent me one like 2 months later, then another rejection for the same position 2 months after that.

File: flux222.jpg (1.6 MB, 2048x2048)

1.6 MB JPG

flux is pretty fun

File: 1701041908339847.png (1.19 MB, 1024x1024)

1.19 MB PNG

>>101726080
you can prompt stuff like "4 panel comic" and get a comic. if you describe what happens in the first/second/etc panels, it can do that too.

>>101726110

Can it make up its own description? guessing no, but that would be next level shit. Make 200 page comic with a intriguing story about "xyz"

>>101726103
What kind of prompt did you use for Chun Li? I tried several times but it didn't seem to know the character.

>>101726103
Chun Li prompt?

>>101726126
no idea, im basically on day 1 with this stuff, but it can do lots of shit.

File: FLUX__00001_.png (1.2 MB, 1024x1024)

1.2 MB PNG

>>101726080
there's enough coherence to get actions down per panel, but I dunno about consistency, for now

File: 1692457085814186.png (1.12 MB, 1024x1024)

1.12 MB PNG

>>101726006
I take it back, switching from Schnell to fp8-dev seems to have improved that

>>101726188
cfg too high

>>101726233
I mean, to be exact, that's not really accurate to the graphics of the period. I think you were right the first time.

What's amazing about SD is how it can a little bit do EVERYTHING. When you start to get careful aesthetic dataset curation a la Dall-E you lose that. FLUX reminds me of Dall-E3 more than anything. And that's cool—Dall-E is great at some stuff. If you want to prompt Obamas, FLUX will mog SD3 hard.

>>101726048
would you even want to work for meta?

File: 1721481292354860.jpg (208 KB, 1024x1024)

208 KB JPG

>>101726326
My experience is the opposite, DALL-E 3 is the best at capturing aesthetic/style, but has the least compositional variety. Picrel is DALL-E 3's take on basically the same prompt for the image above, which is the gold standard I'm trying to reach.

File: file.png (42 KB, 761x502)

42 KB PNG

>>101726102
it's stupid desu. i'm top 10 by additions upstream just from the stuff ive bothered to make a PR for. i've literally worked with the original developers on their private fork and my own fork has tons of new stuff. plus modeling which nobody else really worked on at all, ive added examples upstream and i have probably 90+% of diffusers implemented
they would have seen that from my application but its just not good enough for a rejection email
meanwhile their actual new hires make brilliant contributions like pic related
its been that long i cant really remember what position i did apply for
>>101726329
i'd work anywhere that pays me desu. ait isn't really what i want to be working on though. there's a position at openai that is pretty much exactly what i want

File: delux_ggf_00030_.png (1.09 MB, 896x1152)

1.09 MB PNG

>>101726080
not without some kind of ipadapter + controlnet implementations. still need finer control and consistency for long form story telling

>>101726110
any idea how to get those style of eyes from the third panel reliably?

>>101726126
>Make 200 page comic with a intriguing story about "xyz"
you wouldn't want to do the story directly with imggen. you'd want to start with a long-context llm that can write out the story. then you'd need to do a second llm or multimodal model pass to break the story out into storyboards, chapters, pages, and finally image prompt. then you can plug it all into imggen and make your story

>>101726326
What sucks about flux is that every gen looks the same, the lack of a negative prompt makes every gen look very dalleish, also it lacks quality consistency, with the same settings you get very different quality outputs (one gen looks sharp, the next looks blurry, the next looks pixelated, etc), meanwhile with a nice sd workflow you get 99% the same quality in every gen

>>101726343
this could easily be a duke3d type game, neat

File: FD_00205_.png (986 KB, 1024x1024)

986 KB PNG

>>101726132
>>101726139
Closest I could get. No idea how the fuck he did it

File: 1722814223305_out-0.jpg (128 KB, 1292x738)

128 KB JPG

Where can I run Flux Pro online?
Is there a Collab?

File: 1705289803426886.png (1.67 MB, 1024x1024)

1.67 MB PNG

>>101726427
Yeah I wish I could figure out how to do it locally. Also contrast picrel

File: 1701023654244405.jpg (225 KB, 1024x1024)

225 KB JPG

>>101726452

>>101726452
in flux, using "pixel art" before the rest of the prompts seems to work really well, in making retro game types of images.

>>101726463
I'll try switching the order, the above was made with

>1993 1994 1995 2.5D doom fps video game scene set in hong kong, pixel art sprite style. POV holding crowbar, HUD displays health and ammunition. Tall buildings, vibrant atmosphere, sunset horizon. Street next to canal, polluted, yellow chartreuse color theme, pixel art, sprite graphics

>>101721011
Is there a good image to text model out there yet? It seems like an obvious missing link for erp because otherwise you would have to (poorly) describe the image in text to send back to the chatbot.

File: SGI_Wallpaper_Ice.png (3.46 MB, 1920x1080)

3.46 MB PNG

Can someone try on flux making vintage sgi/cgi art from the 90s, mid 90s, like bryce or truespace 3d plugged art.

>>101726490
florence is supposedly good. there's a lot of vlms but idk which are superior

>>101726490
I think gemma can do it..?
I dunno, something by google gives you a description rather than just tags, and it's open source

>>101724779
It can't do most nsfw, it can't recognize many characters or artists.

File: file.jpg (281 KB, 1024x1792)

281 KB JPG

i think i have enough lyrics for a first version of a lyrics llm
gonna work on that tomorrow desu

File: delux_ggf_00031_.png (1.18 MB, 896x1152)

1.18 MB PNG

>>101726503
idk what any of that means but I'll see what comes out

File: delux_ggf_00032_.png (1.17 MB, 896x1152)

1.17 MB PNG

>>101726568
>lyrics llm
that would be very cool. its really hard to get gpt to be creative with lyrics. it always sound like a 70 year old white guy wrote them

File: m8x6b3v7m9a81.jpg (355 KB, 2000x2000)

355 KB JPG

>>101726589

its basically really early cgi from the early to mid 90s whether consumer grade or professional.

>>101726609
>it always sound like a 70 year old white guy wrote them
You mean the guys who wrote the greatest and most creative poetry in history?

>>101726589
>idk what any of that means
It means Softimage, think the Abyss water creature.

>>101726640
Prompt for this one?

>>101726649

Sorry I should have said, that's not AI. It's from an old software called bryce

Are there any good guides for creating nsfw content off existing selfies and nudes? I have a bunch of a girl I'd like to make ai stuff of but idk where to start

File: 1700334417706827.png (1.33 MB, 1024x1024)

1.33 MB PNG

>>101726670
Oh I thought you found a prompt it in Flux. I'll give it a shot.

File: file.jpg (204 KB, 1024x1792)

204 KB JPG

>>101726609
yeah thats why i want it, chatgpt sucks for lyrics. suno's lyrics aren't that much better, they're probably using the api, maybe a llama finetune or something
it's going to work with artist names. like i don't have a problem with restrictions when its necessary to stop cunny or retarded /pol/ shit because there's no actual use case for any of that. there are use cases for lyrics in the style of an artist, its more specific than a style or genre and as long as its well trained its not going to generate the exact real lyrics

File: lqU1H2HGXwmd4_z0g2ERv.png (1.1 MB, 1024x768)

1.1 MB PNG

best i can do

File: delux_tru_00001_.png (1.31 MB, 1344x768)

1.31 MB PNG

>>101726640
>>101726644

Inspired by mid-90s SGI/CGI art evokes the iconic styles from Bryce and TrueSpace 3D. It features a surreal landscape bathed in a {neon|muted|dark} twilight with polygonal {mountains|buildings|scenery|jungle|abstract shapes} and reflective water surfaces that stretch infinitely. Futuristic geometric structures with {chrome|detailed|rusted|interesting} textures and glowing edges cast soft diffused light. The scene includes floating translucent objects and a gradient sky blending deep purples and vibrant oranges, creating a nostalgic yet futuristic aesthetic that captures the essence of the digital art revolution of the 90s.

this is what I'm getting from an LLM prompt. I can add 'Softimage' but I'm afraid that will give me software UI. any ideas for more tokens to try?

>>101726641
debatable but even if true, lyrics aren't poetry. lyrics are an avenue to make a voice an instrument. there's a deeper need for rhythm, groove and pop in a song lyric than a poem. part of why gpt sucks at lyrics is probably because it just write poetry

File: nIJFTSnzyv4A7mxdSJs1p.png (1.04 MB, 1024x768)

1.04 MB PNG

>>101726736

Yea thats impressive, now just gotta make it a little bit less sophisticated and you basically have it.

>>101726490
llava/cambrian/moondream are all fine

>>101726640
this looks so so close to the very first 3d render i ever did in like... 1999

File: delux_tru_00009_.png (1.46 MB, 1344x768)

1.46 MB PNG

>>101726728
>when its necessary to stop cunny or retarded /pol/ shit
based
> there are use cases for lyrics in the style of an artist
yeah its really annoying that suno rejects "in the style of X" because its really hard to directly describe an artists specific style or form

>>101726771
>now just gotta make it a little bit less sophisticated
easier said than done. was having this problem when I was trying to do morrowind obama gens. I couldn't get it to downgrade itself into 90s polygons

So what happens to stable diffusion now? Isn't comfyui like part of stability or some shit, seems kind of crazy that it works with it, what happens to a1111 ? its like all the advancements to stable diffusion don't mean anything anymore.

>>101726836
how do you have the iq to even use a computer with an opinion like this

>>101726843

How are they ever going to make money now lol

>>101726836
>Isn't comfyui like part of stability or some shit
Or some shit indeed.

File: 1699651044520549.png (1.38 MB, 1024x1024)

1.38 MB PNG

First attempt at Bryce 3D, should have specified something more outdoors, but I kinda like it

>>101726836
>Isn't comfyui like part of stability or some shit
He quit.
>what happens to a1111
he's slow to update and they don't like him so he doesn't get early access to the models
>advancements to stable diffusion
SD3 was a flop anon

>>101726871
>and they don't like him
Isn't it just because he's addicted to some videogame so he isn't really reliably working on his project?

>>101726736
so cool.

File: schFLUX_00012-Minecraft V(...).png (732 KB, 1088x896)

732 KB PNG

Quokkraft!

File: delux_tru_00010_.png (1.25 MB, 1344x768)

1.25 MB PNG

>>101726836
>Isn't comfyui like part of stability
no
>what happens to a1111
voldy adds other models to it but hes a perma-drunk video game addict so he gets to it when he gets to it
>its like all the advancements to stable diffusion don't mean anything anymore.
BFL are mainly ex-sd so they were able to apply a lot of what they learned there to make flux. SAI also has more models coming out so we can't count them out yet

>>101726851
they seem to be lining up for a similar model to BFL. limited-use mid-level models as open source offerings, then premium models powering their paid saas platform

>>101726858
nice

File: 1691767403419115.png (1.49 MB, 1024x1024)

1.49 MB PNG

>>101726920

File: delux_tru_00012_.png (1.13 MB, 1344x768)

1.13 MB PNG

>>101726936
very biased on rivers and hallways

>>101726964
Yeah I've noticed that with urban gens as well, you'll often get down-street/down-canal pics; I'm guessing that many models take the vanishing point into consideration

File: 1716641335501127.png (730 KB, 1024x1024)

730 KB PNG

pixel art. Resident Evil 1 in game screenshot. The main character is Hatsune Miku. She is running down a hallway. She is holding a silver pistol. Miku is wearing a black beret. The setting is inside a mansion.

does flux accept (weights:2.0)?

File: file.jpg (265 KB, 1792x1024)

265 KB JPG

video game screenshot dataset would be cool desu
>>101726818
yeah exactly, genres are too varied, even different tracks by the same artist can be varied

>>101727024
In my experience they don't do anything at all.

File: delux_tru_00013_.png (1.48 MB, 1344x768)

1.48 MB PNG

imagine if I could but "high polygon count, setting sun" in a negative prompt....

bored yet?

>>101727024
They seem to work but is prompt dependent.

File: delux_tru_00014_.png (1.39 MB, 1344x768)

1.39 MB PNG

>>101727070
been on this grind since 2022 and I'm still here 27 hours a day

I wish I didn't have a job otherwise id immerse myself into this shit full time.

>>101727101
sounds miserable. you need structure in your daily routine or everything falls apart

File: 37.png (873 KB, 1024x1024)

873 KB PNG

>>101726879

>>101727117

Its the ultimate blue pill, I know it wouldn't be good for me long term, but it would be pretty based for say a few weeks. At least we get off labor day coming up and a 4 day work week.

File: delux_tru_00015_.png (1.3 MB, 1344x768)

1.3 MB PNG

>>101727101
what would you do if you had infinite time? would you gen 1girls? would you experiment? would you write code? how would you spend your time?

>>101723514
Prob that's your max speed then. I mean vram+ram speed combined. The only way to speed up is if we can isolate it to run on vram only.

>>101727163

Grab a camera, take photos of things and make my own loras, try new methods and ideas that are being shared. I have ideas but it would be very time consuming and maybe it wouldnt even work but it would be cool to try and ive had luck in the past with some gens that were almost all custom.

File: delux_tru_00017_.png (1.2 MB, 1344x768)

1.2 MB PNG

>>101727216
sounds based. hope you carve out time for your ambitions in the future. who knows, maybe you'll luck out and get fired

>>101727117
My structure these days is pretty much:
>wake up
>eat
>exercise
>shower
>watch anime
>work at home
>eat
>gen
>fap
>sleep

>>101727276
>no lunch
>one fap sesh
>doesn't call his mom
deeply flawed

>>101727303
There's snacking involved here and there.
My whole family is dead.
Too old to fap more than once per day. When you get older you lose a lot of sensitivity down there after the first fap of the day and barely anything comes out the second time of the day.

>>101727320
are you 90?

>>101727378
kek no

File: delux_tru_00018_.png (1.47 MB, 1344x768)

1.47 MB PNG

I can't escape river prison

>>101727320
>When you get older you lose a lot of sensitivity down there after the first fap of the day and barely anything comes out the second time of the day.
Exercise more, eat less crap.
It's legit giving libido a boost.

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

1.41 MB PNG

>>101727730
he said he exercises daily tho. very admirable for a man of his years

File: 00017-3418930421.png (1.21 MB, 864x1304)

1.21 MB PNG

>haven't diffused in 5 months
fug bros i feel empty

File: delux_tru_00025_.png (1.27 MB, 1344x768)

1.27 MB PNG

>>101727769
>>haven't diffused in 5 months
you got 5 months of diffusing to get caught up on. get to it

>>101727749
This is why I said more.
Not just cardio, actual hiit stuff.

>>101727803
Weight training only results in a 15 minute to 1 hour testosterone boost.

File: delux_tru_00026_.png (1.44 MB, 1344x768)

1.44 MB PNG

>>101727842
regular weight training leads to sustained elevations in baseline testosterone levels over time. but also, test levels aren't a foundational measure to long term good health

with that said, I think I'm giving up on this truespace prompt. its made cool stuff but I can't get anything close to the 90s vibe no matter what I do. I wonder why flux hates the 90s so much

>>101726449

https://fluxpro.art/

>>101727961
>make an account
no

File: 1718538280485865.png (763 KB, 1016x760)

763 KB PNG

>>101727961
Ah yes, my next generation AI working wonders

>>101728108
footfags get the rope

>>101728108
Uhhhhh, fluxbros? Our answer?!

File: 1711690665370931.png (23 KB, 377x35)

23 KB PNG

>>101728168
At least it can generate carpet fringe with exceptional accuracy

File: 1718791991424727.png (1.08 MB, 1024x1024)

1.08 MB PNG

>>101728168
works for me

File: sd_bikes.jpg (921 KB, 1598x1598)

921 KB JPG

sd's imagination is beyond human.
understood the proper use of generative AI and controlnet.

>>101728285
fuck off emad nobody likes your shitty model

File: delux_tru_00028_.png (1.34 MB, 1344x768)

1.34 MB PNG

>>101728396
emad is off shilling some kind of crypto thing now. he doesn't care about sd anymore

>>101728453
It's kind of sad.

File: 2024-08.png (233 KB, 588x575)

233 KB PNG

>>101728453
are we sure bro is doing okay?

File: delux_tru_00030_.png (996 KB, 1344x768)

996 KB PNG

>>101728479
its only sad if you had believed in emad in the first place

>>101728483
his twitter has always been the worst. it was so brutal following him in the hopes of getting scraps of AI/SD news only to get injected with an endless stream of garbage

File: schFLUX_00013-1995, squig(...).png (1.29 MB, 768x1344)

1.29 MB PNG

I want a PC that can run FLUX locally so bad...

>>101728396
sd users hate emad, so stfu you newfag

File: fakezaku.jpg (232 KB, 800x800)

232 KB JPG

>>101728396
It's just a tool.
I don't care who made it or for what purpose.

>>101728524
Ooof,
To be honest I only really check his feed for my dose of schadenfreude. Mentally I've written off any public statements he makes considering how liberal he can be with the truth which is a shame.

>>101728585
Target next Christmas, by then we should be able to finally get some nsfw out of it.

File: delux_tru_00031_.png (1.06 MB, 1344x768)

1.06 MB PNG

>>101728585
I thought you were using your gainful employment to save up for a local machine?

File: F_00080_.png (1.15 MB, 968x1088)

1.15 MB PNG

>>101728585
How does it do with quokkas?

File: delux_tru_00035_.png (1.32 MB, 1344x768)

1.32 MB PNG

<egg>

File: 1722827531317_out-0.png (739 KB, 1536x640)

739 KB PNG

>>101728599
I hope by next Christmas someone figures a way to run heavy models like FLUX on lower end systems.
>>101728600
I still am, i think right now i have at least enough for a laptop with a 3050 or a 4070 but considering FLUX needs 24gb i don't think a laptop will be enough.
I need to go desktop.
>>101728605
Very strangely, just the word Quokka gives you animals that are closer to rats or hamsters, i need to specify that i want a Quokka with all the features that cascade and sdxl gave.
So far the best way to get closer to cascade Quokka is by writing

Australian brown fur fat cheeks smiling small

Before the word Quokka and even then the ears sometimes come out way too big

can we get a bake

>>101728822
>>101728822
>>101728822