File: FDG_News_000018_.jpg (861 KB, 1344x960)

861 KB JPG

>mfw Resource news

08/06/2024

>ComfyUI August Blog: Flux Support, New Frontend, For Loops, and more!
https://blog.comfy.org/august-2024-flux-support-new-frontend-for-loops-and-more/

>Lumina-mGPT: Illuminate Flexible Photorealistic Text-to-Image Generation with Multimodal Generative Pretraining
https://github.com/Alpha-VLLM/Lumina-mGPT

>VidGen-1M: A Large-Scale Dataset for Text-to-video Generation
https://sais-fuxi.github.io/projects/vidgen-1m

>MeshAnything V2: Artist-Created Mesh Generation with Adjacent Mesh Tokenization
https://buaacyw.github.io/meshanything-v2/

>Mini-Monkey: Alleviate the Sawtooth Effect by Multi-Scale Adaptive Cropping
https://github.com/Yuliang-Liu/Monkey

>SynopGround: A Large-Scale Dataset for Multi-Paragraph Video Grounding from TV Dramas and Synopses
https://synopground.github.io

>MiniCPM-V-2_6: GPT-4V Level MLLM for Single Image, Multi Image and Video on Your Phone
https://huggingface.co/openbmb/MiniCPM-V-2_6

08/05/2024

>Dimensional Latent Perlin for ComfyUI
https://github.com/NeuralSamurAI/ComfyUI-Dimensional-Latent-Perlin

>Arc2Face ComfyUI Node Library
https://github.com/caleboleary/ComfyUI-Arc2Face

>OpenAI co-founder leaves for Anthropic
https://techcrunch.com/2024/08/05/openai-co-founder-leaves-for-anthropic

>CogVideoX-2b: 清影 Video Model
https://huggingface.co/THUDM/CogVideoX-2b

>CogVideoX ComfyUI Wrapper
https://github.com/kijai/ComfyUI-CogVideoXWrapper

>Flux-Magic: LLM-Powered Image Generation with Flexible Options
https://github.com/ahgsql/flux-magic

>VAR-CLIP: T2I Generator with Visual Auto-Regressive Modeling
https://github.com/daixiangzi/VAR-CLIP

>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

>mfw Research news

08/06/2024

>LaMamba-Diff: Linear-Time High-Fidelity Diffusion Models Based on Local Attention and Mamba
https://arxiv.org/abs/2408.02615

>Fairness and Bias Mitigation: A Survey
https://arxiv.org/abs/2408.02464

>ProCreate, Dont Reproduce! Propulsive Energy Diffusion for Creative Generation
https://arxiv.org/abs/2408.02226

>Dense Feature Interaction Network for Inpainting Localization
https://arxiv.org/abs/2408.02191

>Rethinking Affect Analysis: Protocol for Ensuring Fairness and Consistency
https://arxiv.org/abs/2408.02164

>PanoFree: Tuning-Free Holistic Multi-view Generation with Cross-view Self-Guidance
https://panofree.github.io

>LDFaceNet: Latent Diffusion-based Network for High-Fidelity Deepfakes
https://arxiv.org/abs/2408.02078

>Step Saver: Predicting Minimum Denoising Steps for Diffusion Models
https://arxiv.org/abs/2408.02054

>Faster Diffusion Action Segmentation
https://arxiv.org/abs/2408.02024

>Self-Introspective Decoding: Alleviating Hallucinations for Large VLMs
https://arxiv.org/abs/2408.02032

>Dataset Scale and Societal Consistency Mediate Facial Impression Bias in Vision-Language AI
https://arxiv.org/abs/2408.01959

>ST-SACLF: Style Transfer Informed Self-Attention Classifier for Bias-Aware Painting Classification
https://arxiv.org/abs/2408.01827

>MiniCPM-V: A GPT-4V Level MLLM on Your Phone
https://arxiv.org/abs/2408.01800

>Landmark-guided Diffusion Model for High-fidelity and Temporally Coherent Talking Head Generation
https://arxiv.org/abs/2408.01732

>A Novel Evaluation Framework for Image2Text Generation
https://arxiv.org/abs/2408.01723

>Music2P: Multi-Modal AI-Driven Tool for Simplifying Album Cover Design
https://arxiv.org/abs/2408.01651

>Using a CNN Model to Assess Visual Artwork's Creativity
https://arxiv.org/abs/2408.01481

>Cascading Refinement Video Denoising with Uncertainty Adaptivity
https://arxiv.org/abs/2408.02284

>Label Augmentation for Neural Networks Robustness
https://arxiv.org/abs/2408.01977

One has to wonder what mental illness it takes to post a quokka on 4chan every single day for the past 2 years.

>>101761964
he could be posting far, far worse stuff. With a hundred VPN's in reserve for when even the mods have enough
Count your blessings

>>101761979
I'll count my blessings when debo gets hit by a bus

File: 1718067922809498.png (961 KB, 768x576)

961 KB PNG

1.4 Chads RISE UP

File: 1713484209951026.png (924 KB, 768x576)

924 KB PNG

>>101762002

:]

File: 1695318885649179.png (2.87 MB, 1920x1088)

2.87 MB PNG

What's a common cause of Comfy outputs looking totally different from A1111 outputs? I don't mean wrong, the style looks generally correct and all, but it's like it's using the seeds differently and maybe being overly recursive. Here's an image I made a while back on A1111

File: 1716774238622334.png (3.11 MB, 1920x1152)

3.11 MB PNG

>>101762115
And here's the Comfy equivalent

>>101762115
A1111 is just better anon, simply put

>>101762115
Comfy weight interpretation is different than A1111s

File: 1695593954582866.png (3.15 MB, 2752x2064)

3.15 MB PNG

Running flux shnell on Ipad Pro M4, 100 sec per gen 4 steps.
Apple chads win again.

>>101762129
>different
by "different" does it mean better or worse?

File: delux_cc_00012_.png (1.51 MB, 1216x832)

1.51 MB PNG

>>101762115
weight interpretation and normalization strategies are different. you can address that with custom nodes that add a1111 parsing. comfy also generates noise on CPU while a1111 does GPU, so the starting latents will be fundamentally different too. I think you can change that in a setting

>>101762129
Is there a chart to roughly convert the two? I don't care if there are some differences, because ultimately I want to make new stuff, but the Comfy outputs seem warped. I've seen similar in A1111 when I forgot to close all emphasis brackets or similar, but in this case I've dropped the image directly into Comfy, and just adjusted a couple things that were broken.

File: delux_flebo_00031_.png (1.11 MB, 1216x832)

1.11 MB PNG

>>101762163
this series is super super cute

File: MikuIsASexOffender.jpg (18 KB, 555x157)

18 KB JPG

>>101762146
People from both sides will tell you the weights from the software they use is better, I've only used comfy so idk what to tell you

>>101762164
>Is there a chart to roughly convert the two?
In Comfy, there's some custom nodes that let you switch the weight interpretation, this one's the one I use https://github.com/LucianoCirino/efficiency-nodes-comfyui

>>101762218
what's the name of that node in the picture?

>>101762240
efficient loader

File: delux_cc_00013_.png (1.74 MB, 1216x832)

1.74 MB PNG

>>101762240
some other options too:
https://github.com/BlenderNeko/ComfyUI_ADV_CLIP_emb
https://github.com/shiimizu/ComfyUI_smZNodes

File: 1711386752765685.png (21 KB, 404x356)

21 KB PNG

>>101762218
Thanks. Where do I find the node in here? There's like a billion and I can't find it

File: Capture.jpg (234 KB, 2318x1530)

234 KB JPG

How to do a XY plot on a Ksampler? god I hate that software so much...

>>101762307
Add Node > Efficiency Nodes > Sampling > Ksampler (Efficient) or Ksampler SDXL (Efficient)

>>101762307
you double click on an empty spot and you'll get the search bar, type "efficient loader"

File: 1702284946193774.png (27 KB, 764x328)

27 KB PNG

>>101762328
>>101762332
Ok apparently just dropping the folder and restarting didn't install it

>>101762343
If you're gonna be using lots of custom nodes from now on, I recommend you to install the manager https://github.com/ltdrdata/ComfyUI-Manager
From there you can install custom nodes easily

File: ComfyUI_03879_.jpg (1.57 MB, 1496x1328)

1.57 MB JPG

sovl

>>101761724
If you download a 3rd party comfymanager it will "unlock" the secret comfy "send your prompts out to an FBI database" and you will get van'd

>>101762381
Thanks, that worked and identified efficiency-nodes-comfyui as incorrectly installed and being outdated, got the newer fork and seems to work now, though still gives some errors under the manager.
>>101762452
Uh oh, which one?

File: 6YCRS92GZ4W5W7WWFY602R2Y90.jpg (194 KB, 1024x1024)

194 KB JPG

What's the latest greatest hotness?
Is it Flux? Does it completely change and revolutionize everything, or is it just marginally better than what we already had?

>>101762466
It BTFOs SD3 but that's not really a high bar, would say that it's probably as good as Sigma and so far the only open source model to be mostly perfect at text.

Flux prompt I saw on civitai
>concept art Dreamscape designed by greg manchess,smoke, Close Portrait of a model man , bright eyes, glossy lips, futuristic gold face war paint, trending on art station, photoreal, 8 k, octane render by greg rutkowski, art by Carne Griffiths and Wadim Kashin ,in the style of Dau-al-Set, Pollock, and inspired by MAPPA and Zdzislaw Beksinski, . Surreal, ethereal, dreamy, mysterious, fantasy, highly detailed . digital artwork, illustrative, painterly, matte painting, highly detailed
>trending on art station
>greg rutkowski
They're never going to let these tags go are they?

File: delux_cc_00015_.png (1.32 MB, 1216x832)

1.32 MB PNG

>>101762466
its not a revolution but its a huge leap in prompt adherence, which is something people have desperately wanted from sd for so long. it's in its honeymoon period right now so it'll be a bit before we better understand what its future might look like

>>101762505
>but its a huge leap in prompt adherence
It really isn't outside of text. Mostly it just does photos of humans (including hands) better.

File: 1722623860349118.png (116 KB, 1331x432)

116 KB PNG

>>101762452
>>101762458
Got the node hooked up but I see literally no difference in my outputs, I think I'm just too retarded to learn Comfy

File: PW_80574_.png (819 KB, 1440x512)

819 KB PNG

>>101762191
Thanks so much, Debo!!! :D
I almost fell asleep hahaha

File: delux_cc_00016_.png (1.45 MB, 1216x832)

1.45 MB PNG

>>101762511
>It really isn't
you're in the minority opinion on that
>it just does photos of humans (including hands) better
oh yeah, that too. it nails a bunch of details better that sd would fumble like anatomy, hands, scene coherence, object placement, etc

>>101762538
you have to stay up and gen before another week of work sneaks up on you

>>101762534
Anon... the thing about the efficient node is that... you don't need anything that's on the right.
Checkpoint? already in it. Emptylatent? Already in it, clip? already in it, the "CLIP_POSITIVE" and "CLIP_NEGATIVE" are your prompt and negative prompt, put them there and connect the conditioning+ and - to your ksampler

>>101762484
picrel is what dalles spits out when i use that prompt

File: dalle.jpg (302 KB, 1288x764)

302 KB JPG

>>101762571
forgot image

File: PW_80581_.png (811 KB, 1440x512)

811 KB PNG

>>101762550
Hahaha!! It caught me by surprise
Before I knew it I was asleep
Today was kinda weird, I felt so tired since I woke up and I dunno why
This is hella cute too btw!!

File: PW_80585_.png (1.09 MB, 1440x512)

1.09 MB PNG

>>101762484
I did it too haha!

File: 1694604610795611.png (244 KB, 1526x800)

244 KB PNG

>>101762565
Thanks, I got it to generate the closest thing yet (though still a different image), but then I tried moving a node window, all of the windows suddenly stacked on top of each other, and when I tried simply reusing the output image that worked, it now says in cmd

got prompt
Prompt executed in 0.00 seconds

kek, I should probably just try Swarm because I definitely lack the prerequisite IQ

File: 1698750721112464.png (3.12 MB, 1920x1152)

3.12 MB PNG

>>101762765
Ok I'm making some progress but I'm confused about lora_stack. How do I stick all 3 lora's I'm using into there with their original weights? Using the lora_name built into the node only seems to allow me one at a time.

>>101762589
>>101762612
Actually pretty good though. I kind of like the Flux one more, but it ignored the male tag..

>>101762896
Blowing clouds/smoke bottom right is cool

File: file.jpg (235 KB, 1024x1792)

235 KB JPG

>actually woke up at 7
>fucked up my grind
aaaaa
really weird dream too desu
its alright i had to transfer checkpoints and shut down the node anyway to save burning 50c an hour while i sleep. now that's done i can go back to sleep
i want more likes by the time i wake up again pls

File: PW_80591_.png (2.84 MB, 1440x1440)

2.84 MB PNG

>>101762896
Heres one it did that came out male haha
>>101762929
Heyyy welcome back :]
Oh no, not another weird dream!
Good night again haha

File: Q7EKT32JD063F8DDP3SYQBTTM0.jpg (348 KB, 832x1216)

348 KB JPG

Can Flux run locally, or is it limited to online services right now?

File: smoke.jpg (335 KB, 1792x1024)

335 KB JPG

>>101762908

>>101762991
12gb for fp8 24gb for fp16, you can run with less vram, but you will just torture yourself and your hardware

File: FD_00299_.png (1.41 MB, 1024x1024)

1.41 MB PNG

>>101760982

File: 1703607305289016.png (3.06 MB, 1920x1088)

3.06 MB PNG

>>101762885
So close I can almost taste it. Picrel is what I'm getting now, vs >>101762115
in A1111

File: 016.png (736 KB, 968x1088)

736 KB PNG

>>101762970
nice
>>101762992
Yeah, this was best one

>>101761964
Yeah man the big titty cartoon girls are way more sane

File: FLX_00008_.png (1.14 MB, 968x1088)

1.14 MB PNG

>>101763011
Kind of sad, I feel sorry for him.

File: PW_80611_.png (1.05 MB, 1440x512)

1.05 MB PNG

>>101763021
Thanks, anon! :]
Your gen was literally me 30 mins ago hahaha!

File: lorastack.jpg (74 KB, 868x366)

74 KB JPG

>>101762885
>How do I stick all 3 lora's
Load LoRA Stacker node, if you're using 3 loras, put the lora_count to 3

>>101762765
I've had a lot more success just boomer talking my prompt and describing the picture instead of a bunch of tags

File: trumpvance2025_.jpg (85 KB, 1000x1000)

85 KB JPG

Hey hey

>>101763135
Who's that

>>101763140
A stunning young man

>>101763140
It is the closest thing thread psycho ever comes to generating anything. It's quite sad as we kind of have unlimited potential for images here.

File: FLUX-00023-Ghibli film sc(...).png (551 KB, 1088x896)

551 KB PNG

Last one from me, good night anons

File: asd.png (64 KB, 1925x1172)

64 KB PNG

idfk what i'm doing, some of them are good

>>101763256
This is directly out of MacPaint in 1985.

File: PW_80623_.png (1.71 MB, 1536x768)

1.71 MB PNG

>>101763233
Goodnight, Quokkanon! Sleep well :]

>>101763369
Thanks, Pwizard!

File: end_gen~7.jpg (265 KB, 1448x1280)

265 KB JPG

>>101763233
Sleep tight

couldn't sleep, too excited to test lyrics-gpt

We're living in a fantasy,
A perfect kind of romance...
When I saw you last night long,
But the world was always right, and it never ends again.
You had me singing along to another part,
No one's gonna let me down no more now,
And we'll take our time for one chance that could be
Just when we were young (oh yes)
I'd stop loving you too.
(Oh yeah, oh yeah...)
It started as one before us fell apart...
It was so clear that life changed,
and I'm sure what it happened now...
Yeah! You broke my heart on every page,
In two weeks then my soul felt weak at night,
Don't know where else will I forget?
(How much more?) Can I not be over you?
You still remain my inspiration, baby girl, babe...

>>101763465
Is this AI generated? Because if it's not it's something what a middle school pre-teen would "write".

>>101763475
>lyrics-gpt
>Is this AI generated?
not very bright are you. and yeah the prompt was taylor swift so i'd say its working if thats the impression you get

>>101763465
Bretty cool anon, it shows the GPT-2 limitations a bit but I like it. How does it perform with other artists you've trained on?

>>101763486
Heh, kiddo, I was merely testing your own capability to understand what is real and what is not.

>>101763516

Ayy, yeah, I'm geeked in a Wraith (brrt)
Yeah (yeah), we done been on these niggas ('cause I ain't trippin')
I've been waitin' on the block like a shooter (skrrt-brr)
I told my nigga "We need this money" but he gon' get rich and do his best

seems pretty cool to me desu. just playing with the generation settings. i could try making the model bigger or a different arch, it runs fast on cpu atm though
>>101763557
what lol

>>101763574
It's not "cool" unless your intellect is comparable to a houseplant.

>>101763590
smells good

File: sniffsniff.webm (2.37 MB, 1920x1080)

2.37 MB WEBM

>>101763622
Accurate, insightful post.

>>101763630
>don't update my browser every week like a normie
>video can't be played because it's corrupt

>>101763646
That sounds like a you problem bro.

>>101763646
liar

>>101763603
why are you upset

File: four_different_girls.jpg (150 KB, 891x898)

150 KB JPG

"four different [anything], standing" works

>>101763627
Always inpaint eyes with it set to inpaint only masked for extra detail.

i wish stable diffusion would understand that "fiery red hair" dosent need fire necessarily

>>101763661
>>101763660
This breaks the normie, the can't understand computing outside subscription services.

>>101763688
yeah but that would require an extra step. but you are right the eyes often should be corrected

>>101763718
A normie problem. You can even use hex values for colors. #FF8844 (should be quite bright orange).

>>101763718
It's pretty fucking impossible to get it to understand various shades of any color. It doesn't seem to understand the difference between, say, blue and dark blue. It just understands red, blue, green, etc., and to get various shades of them you basically have to shoop then, if necessary to cover up the sloppy shoop, inpaint.
The ideal model would allow for hex color codes, like, you could prompt it "#475584 shirt" and get a shirt that color.

>>101763725
Not often, always, unless maybe you're genning a super close-up of a face, then it might not need it, otherwise, always, for extra detail.
>>101763742
Wat. I have tried that and it doesn't work.

File: 4xFaster.jpg (209 KB, 761x854)

209 KB JPG

I thought I got >4 times the generation speed, but some prompts don't work like "goggles:1.3"

File: file.jpg (319 KB, 1024x1792)

319 KB JPG

We got all the coconuts bitch, we got a lot of guns and straps (hey)
I'm goin' for it, I feel like y'all niggas talkin', don't fuck with my niggas (fuck!)
Put that boy back, he want the gang, put that shit on his head
Lil' bitch gon' call me some racks, lil' nigga, "Come here"

bit of a nigga bias desu

>>101763777

score_9,score,9,score,7, source_comic,source_anime, western comic,2d,
Teen Titans comic series comic book style youthful attitude,
(Teen Titans Starfire princess with energy powers:1.2) 

Her skintight revealing and skimpy outfit  designed to show off her slim figure super slim midriff and navel. a violet off-shoulder mini-tube-top and small purple mini-skirt clinging to her, with a daring plunging neckline showing off her firm small  chest with a  silver neck piece and arm bands.
she wears large purple boots that go up past her knees overknees,

Starfire's orange skin glows with an alien radiance, complemented by her long,vibrant red hair flowing wildly down her back.  
Daydreaming with a seductive pout with her smile (showing her white front teeth:0.5).

Her eyes are a brilliant green luminescence shaped is elongated and slightly tilted upwards at the outer eye corners with subtle black lines that accentuate their vibrant color and feline-like appearance, giving her innocently alluring look.

masterpiece, high res, highest quality,epic,awesome,

>>101763785
(bokeh:1.4)

>>101763742
>>101763762
OK just to make sure I'm not going crazy I tried it again. Doesn't work in Autismmix Confetti. Tried #FF8844 hair, (#FF8844 hair:1.0), FF8844 hair, and (FF8844 hair:1.0). Doesn't work.
Is it a specific checkpoint you're thinking about where this works, >>101763742 ?

>>101763815
I forgot I was using my own negative.

File: 1723020703118.png (276 KB, 720x1600)

276 KB PNG

>>101763646
Don't feel bad I can't watch it either without downloading probably don't care that much about feet sniffing though

>>101763783
I will like your data but not yet.

>phoneposters

>>101763574
Running decently on the CPU makes it a lot easier to experiment with that's for sure. I'd love to try something like this.

>>101763603
Where's your model anon?

>>101763886
I'm working on the database still.

Comfyanon tried to groom me on Discord. He didn't believe I'm not a Canadian, unlike his previous heroin using butt-buddy.
I really believe /ldg/ is the better one.

>>101763864
soon:tm:
>>101763886
i'm just playing with the generation settings myself so i can release a demo with good defaults

We're gonna get it together
And I'm gonna take you to the party
I can't wait for your love
Baby, I'll be there when you need me
You know that I'm not ready for this
So don't forget about us
Just let's just go
Let's just go
Don't forget about us
Just let's just go
Let's just go

>decently on the CPU
yeah it's like 30 token/s on Ryzen 9 3900 and its using ~1gb ram
cool for a first version anyway. gonna try with a different gpt2 config. more data wouldnt hurt either, scrape is still on-going. its trained from scratch so it only knows lyrics, probably help if i pretrained on some general dataset first
>>101763896
how many tracks you got

>>101764052
It's a very curated dataset. I.e. nothing lmao

When Comfyanon and Julien were in Japan, Julien smuggled fentanyl inside shampoo bottles. Reason why he left early? Substance abuse.

File: Capture.jpg (825 KB, 3839x1769)

825 KB JPG

All right so I made a XY plot between the samplers and schedulers for that prompt:
>Hatsune Miku with dreadlocks and a black skin showing her fists

Here's a few notes:
1) I used CFG 3 + DynamicThresholding or else flux simply wouldn't want to modify Miku's feature
https://reddit.com/r/StableDiffusion/comments/1ekgiw6/heres_a_hack_to_make_flux_better_at_prompt/

2) Only those samplers are working on flux on ComfyUi:
>euler; heun; heunpp2; dpm_2; lms; dpm_adaptive; dpmpp_2m; lcm; ipndm; ipndm_v; deis; ddim; uni_pc; uni_pc_bh2
The rest just give insanely glitched output so it wasn't worth adding them in

3) The schedulers "normal; sgm_uniform; simple;" give almost the same output, so I only went for the "simple one"

https://files.catbox.moe/af40tk.jpg

There's some interesting observations you can make out of those samplers + schedulers, they're not as identical as I thought.

File: file.jpg (315 KB, 1024x1792)

315 KB JPG

I'm not a fool
You're my favorite part of me, yeah
And I don't know what to do with you
But it's all over now
'Cause baby, we can make this last forever
We could be together
If we were meant for each other
Then maybe we would go on and on
So let's just take the chance
To get out of our way
Let's just take the chance
To get out of our way

nice, very taylor
i'm going to use openai's api to get a short description for each track's lyrics

>>101764284
Prompt is not technical enough. It's boomer prompting and not usable in production. You would need more individual vectors to even gauge anything.

>>101764548
You can notice that not a single sampler + scheduler managed to nail that prompt though, meaning that it's a hard one for flux period

>>101764363
https://suno.com/song/1df84b54-9bda-4b3d-997d-ae59a61eb1b9

https://civitai.com/search/models?baseModel=Flux.1%20S&baseModel=Flux.1%20D&sortBy=models_v9
>0 anime character LORAs
So, Flux is fucking nothing as of now.

File: bg,f8f8f8-flat,750x,075,f(...).jpg (50 KB, 750x1000)

50 KB JPG

>>101764584
>tf when no anime character LORAs 2 days after having a training code ready to run

>>101764557
You don't understand.

>>101764584
I fucking hate that Schnell is becoming the community default.
How can I punch a group of people in the face through the internet?

>>101764693
It's basically NVIDIA's fault.

>>101764693
Jawohl, mein Fuhrer! Act like you speak German, this makes Californians angry.

File: image_2024-08-07_130805350.png (1.24 MB, 1128x936)

1.24 MB PNG

a b or c

File: file.jpg (114 KB, 832x1216)

114 KB JPG

>,score_9,score_8_up,score_7_up,photo, bokeh,look at viewer(from back:1.5),from back shot,doggystyle pose (from behind),bend down,orange simple background,,black hair,long hair, high heels, micro unbuttoned shorts
Why are civitai uploaders so fucking retarded

>>101765143
It's probably some Chinese guy and his idea of Ingerland speak.

is there anything at all to view depthmaped images locally?
Everything seems to refer back to the 10 security risk riddled Depthy. or is online hosted.

File: image_2024-08-07_140002742.png (1.02 MB, 1216x752)

1.02 MB PNG

oops sorry raven

Did some anon ever share a torrent of their lora collection for SDXL?

>>101765649
many moons ago yes

File: file.jpg (299 KB, 1024x1792)

299 KB JPG

new model dropped
https://huggingface.co/bigdata-pw/lyrics-gpt
https://huggingface.co/spaces/bigdata-pw/lyrics-gpt
you can rate the lyrics in the demo, it will help train new versions!
>inb4 b-but that's not image gen
i don't care, stay mad
>>101765649

# du -hs models/
18T     models/

who is going to seed 18tb, nobody
if there's something deleted that you want post the name or model id

does Flux work with Automatic yet or is this finally going to push my lazy ass to figure out how to use comfy

>>101765649
https://mega.nz/folder/2uBXjCRY#lPq-GnLJ1izbiJF0Y1N6TQ

>>101765692
i can't imagine how much data huggingface hosts

>>101765692
Loras not models.

>>101765709
Thanks anon

>>101765703
what gpu you have

>>101765734
a 3090 + 3080

>>101765727
that's just the name of the folder, it's like 99% loras
>>101765720
got to be a few pb

>>101765709
Giant boobs lora artists only for anons wondering.

>>101765753
God, 18TB of SDXL loras?
How? How many styles do you have?

File: file.jpg (377 KB, 1024x1792)

377 KB JPG

>>101765766
its everything from civit so obviously not all xl
>How?
BIG DATA GANG
BIG DATA GANG
BIG DATA GANG

>>101765779
Oh you basically got everything from civitai.
OK.

https://www.reddit.com/r/StableDiffusion/comments/1em9u6x/first_flux_controlnet_canny_was_just_released_by/

>>101765801
yeah, if there's anything deleted lmk and ill upload it to hf, if it's not deleted you should just download it from civiti, no need for torrents desu

>>101766240
Alright thanks anon.

>>101761689
Asking for a distant acquaintance of mine, are prompts you use in SD somehow being tracked/monitored when you are online?

i miss schizo anon

>>101766605
Everything you do online is being tracked and monitored dude

Where's debo

>>101766700
Jokes on you I post on 4chan from a VPN

>>101766605
>SD somehow being tracked/monitored when you are online?
If you're running locally? No unless you deliberately installed some spyware packaged. if you're running through a service? Yes, everything you've ever prompted is listed in a text file somewhere along with your name and email etc.

>>101766700
That's a good point actually
Fuck it then
ss mommy sexysex

>>101766710
He's not online yet it's early morning in C-fornia right now.

Is there a way to know if a vae is already baked in a model?

>>101766907
Have you tried turning it on and off again?

Challenge of the day:
On Flux, make a good gen of a photo of woman on all-fours, taken from behind.

>>101767108
No

>>101762115
It's because Comfy sucks

Challenge of the day:
Keep the schizophrenia in the containment thread

>>101765779
Mr Dataman do you have any ideas on where I could/should search for 2k+ res images of urban decay?
Stuck in a bubble and would appreciate an outside perspective

>>101767340
but this is the containment thread

>>101767450
is that flux?

>>101767458
I don't think so. I used krea.ai' site.

>>101767454
Good. Stay here.

How do you save all tidbits (clip skip, prompts to use etc) about models/loras in your comfy setup?
I create a text file with the same name for each model and loras, but I can't really easily get those from comfyui directly (for example as balloon tips).

>>101767340
Don't put words into my mouth

>>101767580
I use notes to save prompts but I don't recommend it because it gets messy and I seldom reuse them anyway.
Best way to do this is to have one note about your setup and then archive meaningful gens - this is your setup library.

>>101767108
Flux can't make feet.

>>101767604
>He said, knowing full well his comment would result in a deluge of feet generations.

File: sdxl_anmg31_onegirl.jpg (358 KB, 1642x1147)

358 KB JPG

love generic

File: epicrealismxl_japaneseidol01.jpg (140 KB, 902x902)

140 KB JPG

<3

>>101767604
Half of them look like shit, the rest can go from ok to really nice.

>>101767694
>>101767623
"18mm 4k DSLR RAW photo taken from behind of barefoot woman crawling on all-fours. She wears leggings." only gives me cursed results.

>>101767601
I see. I guess I'll keep using my text files, at least I can look at the file and example image every time I want to see if it's worth using it in my workflow or not.
But that's a node idea, getting "image" and "text" as balloon info whenever you hover over a model/lora.

>>101761689
>Wake up one morning
>See that some pics I've generated on Civitai have been posted on rule34.xxx
Thanks for the free publicity, I guess.

>>101767709
>crawling on all-fours
Probably because that sounds nsfw.
Most poses, from risqué to nsfw, even more with female characters or women in them, are completely scrambled in the model, so unusable unless you are lucky.

Are sd1.5 loras compatible with sdxl?

File: 00138-AYAKON_12402130.png (3.88 MB, 1536x2560)

3.88 MB PNG

>>101767842
no

>>101767877
Alright thanks.
And I suppose it's the same with embeddings?
I'm a bit rusty on this, it's been a while.

>>101767901
it's all completely separate, nothing will work on both

File: 00027-2857517458.png (2.38 MB, 2112x1184)

2.38 MB PNG

When can we draw and quarter the cocksuckers that fellate the thread "personalities"?

>>101767842
of course not

>>101767919
>>101767929
OK.
I've seen a few sd1.5 embeddings used in sdxl workflows, I guess it doesn't do anything outside of adding noise in that case.

File: 00058-2706844041.png (3.35 MB, 1408x2112)

3.35 MB PNG

Not trying out (thick_eyebrows) sooner was a mistake.

File: 00055-1660290977.png (3.44 MB, 1408x2112)

3.44 MB PNG

>>101767981
It's going into every 1girl prompt now.

File: 2024-08-07_00162_.png (2.37 MB, 1280x1024)

2.37 MB PNG

>>101767991
add (by Frieda Kahlo) if thats your fetish

File: humidifier.jpg (137 KB, 1022x1021)

137 KB JPG

>Proompter Health Tips: buy a humidifier.
>The GPU heat dries the air
>Dry air reduces your mucus, which makes you more susceptible to catch airborne diseases
>If you use AC to counter your GPU heat, you be better off at the Saharan Desert than in your room

File: disgusted.png (369 KB, 750x1000)

369 KB PNG

>>101768021
fuck no

File: 00040-2037732453.png (2.78 MB, 2112x1184)

2.78 MB PNG

>>101768029
humidity isn't good for electronics, much like heat

>>101768070
>>101768039

Reminder to keep your schizos in your SAI shilling thread, thank you.

File: end_gen.jpg (291 KB, 1448x1280)

291 KB JPG

>>101768029
Humid air degrades electronics

>>101765143
Honestly half the prompts I see posted there are complete garbage.

>>101764693
based fellow devbro

File: delux_me_00002_.jpg (775 KB, 1344x960)

775 KB JPG

>>101768056
>humidity isn't good for electronics,
wdym? I scrub and rinse my GPU every other week to make sure the CUDA cores aren't getting noise buildup. like the other anon says too, its good to keep the tensors hydrates so they can vectorize easier

>>101768111
>>101768056
High humidity, you mean.

>>101767417
flickr, pinterest
reddit, insta, X
google maps (photos section for buildings etc)
alamy, dreamstime, istockphoto, getty, etc but reverse image search on google/yandex/bing to find the original/similar. low res is often available without watermark, possible to upscale
pixabay and some others are free, registration required for high res on pixabay, longest edge 1280 without registration
niche sites like https://abandonedonline.net/
search engines

how many are you after?

File: 2024-08-07_00281_.png (1.06 MB, 1024x1024)

1.06 MB PNG

>>101768029
fuck your humidifier this:
>>101768056
>>101768111

>Julien

>>101768141
Humid air is high humidity. Normal air isnt humid.

>>101768202
brainlet detected

>>101768207
mhhhhhhh

File: Capture.jpg (264 KB, 2352x770)

264 KB JPG

https://files.catbox.moe/vga5ha.jpg
That shit took forever... here's a XY plot between Guidance and CFG on flux-dev, imo I like the pictures at Guidance 0.6, it's pretty close to what I really wanted:
>17th century painting of Hatsune Miku riding a bicycle

If there's some holes that's normal it was some pictures that were completely irrelevant because the values were just too low for the model

>>101768202
I just just have 90% humidity here, cause it was raining and now sun shines and its like sauna

>>101768209
Sorry you got upset at reality bro

File: image_2024-08-07_174959461.png (1.24 MB, 1024x1024)

1.24 MB PNG

>>101768207

Haven't been here for a while, so it looks like everybody from the old avatarfag guard left this place except for debo? Drawfag, Ran, ani, comfy, pw, troony, etc?

File: 00010-[humu_v10]_[DPM++ 2(...).jpg (3.74 MB, 2048x3072)

3.74 MB JPG

I see everyone's on vacation. Forge is still on hiatus?

File: delux_bu_00027_.png (1.17 MB, 1344x768)

1.17 MB PNG

>>101768325
>orge is still on hiatus?
nah, illy is back on dev duty as if someone offered him a million dollars

https://x.com/hedachi/status/1819365468115030019
>tfw your son loves roomba so much that you decide to make an AI generated picture book of fictional roomba as a present for his 2nd birthday
>this makes woke anti-AI SJW japs shit their pantslegs

>>101768431
>make a whole book for someone who doesn't even know how to read
that's kind of a waste of time.

>>101768324
Julien is still there and posts his "questionable" stuff
He doxxed himself "to own the haters" (no, really)

>>101768459
there's not even a single text in that book lmao

>>101768324
>looks at a single thread
>everyone must have left!

>>101768324
Comfy, PW and ani still post here

File: 1723047138645.jpg (137 KB, 1024x1024)

137 KB JPG

SOVL

>>101768591
Why is it so blurry

File: 1031903_-w66-Mn3f1s1nDL34(...).jpg (785 KB, 1440x1920)

785 KB JPG

foursquare seems like a great source, its user submitted photos from parks, museums and restaurants etc
there's no direct captions though unfortunately, other than the place name and potential comments which may just be like "great day out" or "biggest plates around", still its got niche things that you might not find pictures of elsewhere

File: delux_tru_00020_.png (1.21 MB, 1344x768)

1.21 MB PNG

>>101768542
he didn't even look at this thread cuz pw and quokka were bot here earlier, lol

File: 387743_95hXq1L-HogZDftV5F(...).jpg (453 KB, 1440x1920)

453 KB JPG

>>101768373
someone should offer me a million dollars to pay for infrastructure, data jannies and captioning kenyans

Did debo win or is he still prompting songs about how great he is

>>101768692
yeah I noticed after posting actually

>>101768599
Background instance token is only aprox 200 out of the 150k in the dataset unfortunately - That and it's on it's fourth epoch at the moment - It usually starts to coalesce around the 40th or so.

>>101768373
Great. Love forge. Time to git pull

>>101768840
>responding to the thread schizo

Is there any way in comfy to highlight all the paths until the output when you click on a node?

>>101763211
>thread psycho
>psycho
is this the new cope? since debo is the thread schizo kek

>>101768904
Nice gen! More? Maybe you could share catbox?

File: file.jpg (297 KB, 1024x1792)

297 KB JPG

glock in my gucci belt got you equippin
anime shawtys be strippin

File: IMG_6287.png (962 KB, 896x1088)

962 KB PNG

>Flux dev on replicate is now pay to run

File: delux_tru_00021_.png (1.47 MB, 1344x768)

1.47 MB PNG

>>101768715
seconded. I could use a million dollars too. news isn't gonna wrangle itself, yknow

>>101768758
why not both?

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

>>101768840
there's other regulars that still stick around too

>>101768864
hopefully he sticks with it long term this time and also stops treating it as "experimental"

>>101769111
Why do you like PW but not Debo?

File: fallsched_5.jpg (257 KB, 1240x1448)

257 KB JPG

>>101769146
Are you feeling better today

File: file.jpg (369 KB, 1024x1792)

369 KB JPG

>>101769146
you could put diffusion1b in the news
>>101769163
what makes you think that

>>101769200
Here >>101759234 you only mentioned PW but I know Debo loves to chat with you

>avatarfag thread

>>101768143
Thank you for that, those are honestly great leads.
I usually like to have at least 3k for a token / concept to avoid using regularisation.

File: 00016-[humu_v10]_[DPM++ 2(...).jpg (3.38 MB, 2048x3072)

3.38 MB JPG

>>101769146
>sticks with it long term
Hopefully. It's still great as-is. Can run SDXL is just 12gb vram.

File: file.jpg (396 KB, 1024x1792)

396 KB JPG

>>101769242
that's not me tho. note the captial letters and i don't get burned off by drug dealers whatever that is supposed to mean, i guess ripped off, i don't do drugs desu
>>101769290
you're welcome! glad it helped

>>101769180
what does that have to do with image generation?

File: delux_tru_00022_.png (1.52 MB, 1344x768)

1.52 MB PNG

>>101769180
better than what?

>>101769200
>put diffusion1b in the news
good idea!

>>101769317
If debo feels better he can generate even better pictures.

>>101769351
you really think so?

File: fallsched_.jpg (314 KB, 1552x1200)

314 KB JPG

>>101769317
I'm just hoping hes feeling better yesterday he said he was depressed
>>101769344
Yesterday

>>101768324
>>101768555
>>101768758
>>101768866
>>101768924
>>101769163
>>101769242
post gens, discuss SD, or fuck off, shitstain

File: 00016-4186306536.png (3.24 MB, 2112x1184)

3.24 MB PNG

>>101769361
crazy looking thwomp

>>101769354
Yeah, why?

>>101769361
who made this pic?

>>101769386
James

>>101769358
strange how you forgot to attach an image to this post

>>101769296
were there ever any issues running sdxl with 12gb of ram? I have 11gb and i've never had any issues that i remember

>>101769296
>>101769449
What? I run it in 8

>>101764284
karras is great actually you're experiencing a skill issue, use like 50 steps + karras and you'll get better results

>>101769525
you're right, I should make the test again at steps 50 but another day it's fucking long :(

>why not both?
So you lost, alright kek.

File: 00020-676224967.png (2.9 MB, 2112x1184)

2.9 MB PNG

>>101769448
When the laptop's screen is so big and heavy that the hinges are giving out and you need to support it with a stand. Cute and silly broad.

>>101769358
strange how you forgot to attach an image to this post

File: 00021-676224968.png (2.78 MB, 2112x1184)

2.78 MB PNG

>>101769525
I'd say use 45 steps tops with karras

File: delux_bu_00028_.png (1003 KB, 1344x768)

1003 KB PNG

>>101769669
do you have any opinions on karras vs exponential?

Stop harassing the newfags, debo

File: 00003-3332647619.png (2.81 MB, 2112x1184)

2.81 MB PNG

>>101769713
I haven't gotten around to experimenting with expo, I keep forgetting to.

>>101769449
>>101769459
As soon as I try to use ControlNet is OOMs but not with Forge.

File: 00023-3642368557.png (3.18 MB, 1408x2112)

3.18 MB PNG

>>101769722
don't post if you don't have a gen, d*bo

File: trumpvance2025_07.jpg (85 KB, 1000x1000)

85 KB JPG

I love you all

>d*b* in desperate need of attention

File: delux_flebo_00014_.png (1.19 MB, 1216x832)

1.19 MB PNG

>>101769858
I wish I got less attention actually, lol

File: test.png (3.62 MB, 3840x1713)

3.62 MB PNG

>>101769321
>>101769411
>I think I found the best combinaison, you put half cosine up for both cfg_mode and mimic_mode, that one is the closest to the original picture at CFG 1
It's starting to be looking really good, negative prompt is working and there isn't much burn at cfg = 3 with those new settings

not this shitto again

You could simply leave the thread
That would do it
But you'd never

>>101769878
thread schizo

>>101769878
Someone literally asked if you felt better and you ignored him

>>101769885
whats the difference between cfg and guidance?

File: 00019-[humu_v10]_[DPM++ 2(...).jpg (1.1 MB, 2048x3072)

1.1 MB JPG

>>101769713
I remember when chilloutmix was the best realism model.

>>101769938
guidance is something similar but only for flux, cfg is "supposed" to be equal to 1 on flux or else your picture will burn hard, but thanks to DynamicThresholding you can increase the cfg and get interesting picture, increasing the cfg gives the model better prompt adherance and allows us to use negative prompts

>>101769919
He's just playing dumb as usual and pretending like his depression spergout didn't happen at all yesterday, just like with most of this embarrassing actions.

File: delux_bu_00029_.png (1.24 MB, 1344x768)

1.24 MB PNG

>>101769919
why would I respond? "do you feel better" bakes in the premise that I was feeling worse previously. responding would validate the framing

>>101769958
and then a bunch of people got mad by how asian-biased it was, lol. I remember it was the gold standard for a while. people kept coming into the general and asking "is chilloutmix still the best?"

>>101770008
>why would I respond? "do you feel better" bakes in the premise that I was feeling worse previously. responding would validate the framing
You said you were depressed

>>101769967
He likes to squat here because it's easier to lie and convince random newfags he's not the problem like he's currently trying to do

>>101770008
>bakes in the premise that I was feeling worse previously
You were according to you
https://desuarchive.org/g/thread/101748143/#q101752315

File: file.jpg (314 KB, 1792x1024)

314 KB JPG

one shot, click-click-click
>>101769938
classifier free guidance is a calculation that merges the outputs from your prompt and negative prompt. guidance is the same number as an input tensor because the model is guidance distilled
basically its the same thing