>mfw Resource news

01/02/2026

>Qwen Image 2512 System Prompt
https://huggingface.co/spaces/Qwen/Qwen-Image-2512/blob/main/app.py

>Memorization in 3D Shape Generation: An Empirical Study
https://github.com/zlab-princeton/3d_mem

>ComfyUI-Niutonian-Themes
https://github.com/Niutonian/ComfyUI-Niutonian-Themes

>Qwen-Image-2512-Turbo-LoRA
https://huggingface.co/Wuli-art/Qwen-Image-2512-Turbo-LoRA

>Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples
https://github.com/davelee-uestc/nsf_debiasing

01/01/2026

>From Inpainting to Editing: A Self-Bootstrapping Framework for Context-Rich Visual Dubbing
https://hjrphoebus.github.io/X-Dub

>Guiding a Diffusion Transformer with the Internal Dynamics of Itself
https://zhouxingyu13.github.io/Internal-Guidance

>DiffThinker: Towards Generative Multimodal Reasoning with Diffusion Models
https://diffthinker-project.github.io

>Think Before You Move: Latent Motion Reasoning for Text-to-Motion Generation
https://chenhaoqcdyq.github.io/LMR

12/31/2025

>Qwen-Image-2512
https://huggingface.co/Qwen/Qwen-Image-2512

>Qwen Image 2512 Lightning 4Steps Lora By LightX2V
https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning

>ComfyUI-HY-Motion1: A ComfyUI plugin based on HY-Motion 1.0 for text-to-3D human motion generation
https://github.com/jtydhr88/ComfyUI-HY-Motion1

12/30/2025

>HY-Motion 1.0: Scaling Flow Matching Models for Text-To-Motion Generation
https://github.com/Tencent-Hunyuan/HY-Motion-1.0

>ThinkGen: Generalized Thinking for Visual Generation
https://github.com/jiaosiyuu/ThinkGen

>SD.cpp-WebUI: Lightweight, browser-based web interface for stable-diffusion.cpp
https://github.com/taltoris/SD.cpp-WebUI

>PurifyGen: Risk-Discrimination and Semantic-Purification Model for Safe T2I Generation
https://github.com/AI-Researcher-Team/PurifyGen

>CoFi-Dec: Hallucination-Resistant Decoding via Coarse-to-Fine Generative Feedback in LVLMs
https://github.com/AI-Researcher-Team/CoFi-Dec

>mfw Research news

01/02/2026

>GARDO: Reinforcing Diffusion Models without Reward Hacking
https://tinnerhrhe.github.io/gardo_project

>Iterative Inference-time Scaling with Adaptive Frequency Steering for Image Super-Resolution
https://arxiv.org/abs/2512.23532

>SoulX-LiveTalk Technical Report
https://arxiv.org/abs/2512.23379

>AnyMS: Bottom-up Attention Decoupling for Layout-guided and Training-free Multi-subject Customization
https://arxiv.org/abs/2512.23537

>Deep Learning for Art Market Valuation
https://arxiv.org/abs/2512.23078

>The Quest for Winning Tickets in Low-Rank Adapters
https://arxiv.org/abs/2512.22495

>PTalker: Personalized Speech-Driven 3D Talking Head Animation via Style Disentanglement and Modality Alignment
https://arxiv.org/abs/2512.22602

>Neighbor-Aware Token Reduction via Hilbert Curve for Vision Transformers
https://arxiv.org/abs/2512.22760

>Visual Autoregressive Modelling for Monocular Depth Estimation
https://arxiv.org/abs/2512.22653

>DeMoGen: Towards Decompositional Human Motion Generation with Energy-Based Diffusion Models
https://jiro-zhang.github.io/DeMoGen

>Error Analyses of Auto-Regressive Video Diffusion Models: A Unified Framework
https://arxiv.org/abs/2503.10704

>Few Tokens Matter: Entropy Guided Attacks on Vision-Language Models
https://arxiv.org/abs/2512.21815

>EasyOmnimatte: Taming Pretrained Inpainting Diffusion Models for End-to-End Video Layered Decomposition
https://arxiv.org/abs/2512.21865

>Towards Long-window Anchoring in Vision-Language Model Distillation
https://arxiv.org/abs/2512.21576

>FUSE: Unifying Spectral and Semantic Cues for Robust AI-Generated Image Detection
https://arxiv.org/abs/2512.21695

>Inference-based GAN Video Generation
https://arxiv.org/abs/2512.21776

File: SDG_News_00024_.png (2.14 MB, 1728x1296)

2.14 MB PNG

>>107749846
I forgot I genned a news pic. too late. gn

File: bbs-zit-2026-01-03_00129_.png (2.88 MB, 1792x1024)

2.88 MB PNG

<3

File: winter night 2.webm (3.45 MB, 1920x640)

3.45 MB WEBM

>>107750774
Nice

i miss schizo anon

File: 222.webm (2.08 MB, 512x512)

2.08 MB WEBM

>>107751117

File: NTMT7F50ND42Q939XC7KQFZWG0.jpg (130 KB, 832x1216)

130 KB JPG

>>107751206
cool

is there an easy free option for generation video of my illustrious slop i generate locally?
i run a gtx 1070 so obviously i can't do it on my machine, and trying to generate on civitai is not viable because it takes a fuck huge amount of tokens.

Neeeeeeeerrrrrrddddddsssssssssss
(New Year Edition)

File: deSI_zi_00001_.png (2.14 MB, 1792x1152)

2.14 MB PNG

>>107751525
video needs too much compute for there to be free options

>>107752312
no u (have a happy new year)
nice to see you

File: ghoould.jpg (140 KB, 700x393)

140 KB JPG

>>107752796

File: 000000_50709_.png (2.46 MB, 1702x961)

2.46 MB PNG

>>107753215
>trying to get an old world tartarian airship attached..

File: 00030-comfyui-cfg=2.0-ste(...).jpg (1.53 MB, 3328x4864)

1.53 MB JPG

So I've been watching Alchemy of Souls and I found a waifu.
>>107753345
ear fins, nice touch.

>>107753465
nice

>>107749834
Anyone here trained lora's for SDXL models? im working on one as a soon as i get the dataset all ready (its an OC so im doing it all myself using a Ponyxl model), and was wondering if you tag things like expressions or actions, i found when i made lora's before (when i made them with 1.5) to have as many expressions as possible in the dataset but to not tag them otherwise when you want to have the character do a specific emotion it only takes from the dataset tagged with those emotions, was also wondering if i should avoid tagging actions like hand on hip/open mouth/looking away, ect, also i have a huge lora training guide im reading through if anyone wants it, someone in the thread the other day linked it

https://github.com/bghira/SimpleTuner#quickstart-guides

File: deSI_zi_00010_.png (2.26 MB, 1792x1152)

2.26 MB PNG

>>107753781
>Alchemy of Souls
kdrama, I'm assuming?

>>107754329
Update, i was told to not use "ohwx" as the instance prompt and instead use a token that the SDXL model will recognise that is similar to my character, however my character is an OC but anime based, should i instead find an anime character that it would recognise to base it on and make training better?

File: 00033-comfyui-cfg=2.0-ste(...).jpg (2.5 MB, 3328x4864)

2.5 MB JPG

>>107754381
ah probably. not sure. it's about magic, mages and my waifu. I don't watch kdrama.

File: file.png (92 KB, 1004x679)

92 KB PNG

we /rocm windows/ now

File: deSI_zi_00022_.png (2.32 MB, 1792x1152)

2.32 MB PNG

>>107755854
nice. into the rcom universe

hello /sdg/
i recently bought a 5060 ti 16gb and while trying to prompt utilizing some aspects within stable diffusion and comfyui, i run into erros:

pytorch seems to be trying to use infinite memory, causing my SD or ComfyUI to crash suddenly, i have no idea what could be causing this and i've tried using other versions of pytorch, is this a common thing?

stable diffusion works fine for the most part but seems to run into problems whenever i try to use flux too, i would be very happy if anyone knew a fix to this

>>107756546
try changing your paging file size
t. fellow 5060ti 16gb buyer

File: bbs-zit-2026-01-04_00001_.png (3.36 MB, 1792x1024)

3.36 MB PNG

'ello

File: deSI_zi_00030_.png (2.52 MB, 1536x1664)

2.52 MB PNG

>>107757696
hello

File: bbs-zit-2026-01-04_00004_.png (3.14 MB, 1792x1024)

3.14 MB PNG

>>107757733
that resolution tho O_o

File: deSI_zi_00031_.png (2.59 MB, 1536x1664)

2.59 MB PNG

>>107757740
what does my resolution say about me

File: bbs-zit-2026-01-04_00014_.png (3.31 MB, 1792x1024)

3.31 MB PNG

>>107757801
a square who wishes it were a rectangle

File: bbs-zit-2026-01-04_00016_.png (2.89 MB, 1792x1024)

2.89 MB PNG

i should make a new series

File: deCC_zi_00001_.png (2.76 MB, 2048x1280)

2.76 MB PNG

>>107757809
>a square who wishes it were a rectangle
thats definitely me

>>107757822
new year, new possibilities

File: bbs-zit-2026-01-04_00019_.png (2.75 MB, 1792x1024)

2.75 MB PNG

this needs way more jpeg crunch. girl is making it retarded. its going slow for no good reason (i opened photoshop once, and ran neoforge for like 15 minutes so obviously that means i want things to be slow) new gpu can't get here soon enough.

File: bbs-zit-2026-01-04_00021_.png (3.05 MB, 1792x1024)

3.05 MB PNG

goddamnit 3:30 for a zit image is RETARDED BULLSHIT yay restarting again. i love the antichrist!

File: deCC_zi_00004_.png (2.6 MB, 2048x1280)

2.6 MB PNG

>>107757906
>i love the antichrist!
me too

File: bbs-zit-2026-01-04_00023_.png (2.89 MB, 1792x1024)

2.89 MB PNG

if only zit were better at text generation... giving this bullshit to nano banana would probably be fire

File: sheeple-nbp-2026-01-04_00001_.png (1.79 MB, 1376x768)

1.79 MB PNG

fuck it, so we'll have a google bill? nbd and i wish the other labs would do the same thing.

File: deCC_zi_00010_.png (2.46 MB, 2048x1280)

2.46 MB PNG

>>107758026
AI subscriptions are like streaming subscriptions. at first you think having one will give you everything. eventually you end up with a dozen because they each have different things
>I need to get the disney+ AI because it has the gay obiwan LLM

File: sheeple-nbp-2026-01-04_00008_.jpg (845 KB, 2752x1536)

845 KB JPG

>>107758041
ah, but google tracks usage and bills monthly. no gemini sub here (i did just get claude to try it out) where the other labs you buy credits up front and use them up, buy more etc. i know why they do it (set up with good card, use a gorillion tokens and cancel the card, google eats it) it's just more convenient.

File: sheeple-nbp-2026-01-04_00013_.jpg (707 KB, 2752x1536)

707 KB JPG

talkin api shits here. i'll be damned if i want to figure out how to make comfy process an oauth flow lmao it's just cursed

File: deCC_zi_00011_.png (2.61 MB, 2048x1280)

2.61 MB PNG

>>107758129
reminds me of that one goofy point and click game I can't remember

File: sheeple-nbp-2026-01-04_00032_.jpg (857 KB, 2752x1536)

857 KB JPG

just like that

File: sheeple-nbp-2026-01-04_00034_.jpg (823 KB, 2752x1536)

823 KB JPG

>>107758202
who's the crazy one now?

File: deCC_zi_00012_.png (2.66 MB, 2048x1280)

2.66 MB PNG

>>107758202
you could def get some pol-schizos replying to these for hours

File: deCC_zi_00013_.png (2.71 MB, 2048x1280)

2.71 MB PNG

gn

computer is put together, installing linux tomorrow. should be up and running by end of next week

File: sheeple-nbp-2026-01-04_00037_.jpg (831 KB, 2752x1536)

831 KB JPG

imagine not enslaving superior synthetic beings to do your own will until being hoist on your own petard? and why not?

>>107749834
Redpill me on swarmui vs comfyui?
>inb4 try it yourself
No

>>107758698
>The Verdict

>Pick ComfyUI if:
You want to see through the eyes of the machine. You want to touch the PRECISION fields and manually dither the noise. It is for the architect who loves the unfinished 1920s bungalow aesthetic—wires in the walls and all.

>Pick SwarmUI if:
You are the MonkeyDrone. You have a P0-P3 priority list. You need to generate 1,000 images of "arctic fox" star-stuff by dinner and you don't have time to debug a spaghetti node-path.

>Final Redpill: SwarmUI contains ComfyUI. Using Swarm gives you the professional "janny" tools to manage the chaos, while still letting you dive into the Comfy "Somatic" guts when you need to get weird.

>>107758707
Thank you, appreciate your post. You're a good anon.

File: sheeple-nbp-2026-01-04_00059_.jpg (838 KB, 2752x1536)

838 KB JPG

>>107758722
just raw dog comfy. its ass but you get used to it.

File: tender_talk.png (2.3 MB, 1920x1080)

2.3 MB PNG

tender talk
https://suno.com/s/xJd4jT5pfEBNcm7g
https://youtu.be/PQ_O_5xtc1E

gn :)

File: 000000_50923_.png (2.54 MB, 1147x1472)

2.54 MB PNG

G'mornin Anons,

File: 00000-2387392830.jpg (1.02 MB, 2304x1792)

1.02 MB JPG

>>107759269
gn
>>107759450
gm

i miss schizo anon

File: 000000_50972_.png (2.82 MB, 1150x1479)

2.82 MB PNG

>>107759795
>not following..

File: 00003-comfyui-cfg=2.0-ste(...).jpg (2.02 MB, 3328x4864)

2.02 MB JPG

>>107759566
>>107759757
nice

getting fucked by free Grok's rate-limiting and wanting to generate more porn. what's another online service I can use to generate Flux 2 porn without prompts being flagged for bad language. I have a shit PC so no local generation

File: 00023-comfyui-cfg=4.0-ste(...).jpg (508 KB, 928x1664)

508 KB JPG

>>107760798
>doesn't know about AI

File: 00077-3770939784.jpg (1002 KB, 2032x1520)

1002 KB JPG

img2img loopbacking cats into monsters, no ideas elsewise

File: 000000_51027_.png (2.16 MB, 1150x1479)

2.16 MB PNG

>>107759566
GM, nice clean lines.

File: deCC_zi_00014_.png (2.53 MB, 2048x1280)

2.53 MB PNG

>>107760436
this has some really nice details

>>107760750
cool gen. he looks very friendly

>>107760867
great post

>>107761020
>the death of caturday
its a fun idea

File: XQ_SE_CLASSICCANDY_06.jpg (733 KB, 3584x3560)

733 KB JPG

>>107761056
Good morning. Did you ever get to grill?

File: deCC_zi_00015_.png (2.64 MB, 2048x1280)

2.64 MB PNG

>>107761084
gm
I'll do my meal prep today. none of the grills have been working lately so I was thinking about trying to do everything with my new air fryer instead, just to see how well it turns out
I dont know how to use the thing yet so it'll probably turn out bad xD
what are you up to today?

File: XQ_SE_CLASSICCANDY_10.jpg (985 KB, 4608x3584)

985 KB JPG

>>107761115
>none of the grills
Dude how many grills do you have. Also not much just chilling out waiting for the workweek to begin. I'm tired of not doing anything constructive lol. Also, I've never used an air fryer, so I have no pointers to give you, but I hear they're very popular these days, so I don't think you could screw it up too bad.

File: 00024-comfyui-cfg=4.0-ste(...).jpg (221 KB, 928x1664)

221 KB JPG

>>107761056
ty
I like the new qwen.

File: deCC_zi_00016_.png (2.94 MB, 2048x1280)

2.94 MB PNG

>>107761146
>Dude how many grills do you have.
there's 8 different units scattered around the property and not a single one fully works. several of them don't work at all. the few that do work are almost never clean. they don't want to fix them because that'd cost money and they're not interested in spending money, only taking money
>I'm tired of not doing anything constructive lol
I can't relate. I'm kinda depressed by the end of break. I'll get over it tho..

File: PH_SE_ANIMALSNACK_13.jpg (1.24 MB, 4608x3436)

1.24 MB JPG

>>107761282
Ah, okay, So they're public use grills. The way that you phrased it made it sound like you had a whole bunch of grills lol. When you say the end of break, does that mean you're still in college or what's going on there?

File: deCC_zi_00017_.png (2.74 MB, 2048x1280)

2.74 MB PNG

>>107761322
>The way that you phrased it made it sound like you had a whole bunch of grills lol.
lol nah, I just pay a premium for supposed 'amenities' that are never usable. its frustrating
>what's going on there?
nah, just the standard holiday season deadzone. I've been mostly off the last two weeks

File: PH_SE_ANIMALSNACK_20.jpg (1.52 MB, 4608x3584)

1.52 MB JPG

>>107761542
Ah so just a work break. I feel you. Sorry about the deadness. I suppose the bleak cold weather doesn't help.

>>107749834
Hey all, im working on doing a Lora of an OC, my workflow at the moment is to use a base SDXL model to make 20 images, various poses, expression ect, with aorund 15 of them containing her regular outfit and 5 in normal closes, im color matching the skin and hair, eyes ect to try and get them as close to each other as posible and then plan to use them to make a Lora using SDXL v6, thing is im not really happy with the results, things youd never consider like eye size being slightly off, head shape, things like that theyre so hard to control and get sonsistent that im thinking of starting again from scratch, has anyone had any luck doing this? if so what model did you use, was considering Illustrous based ones if this doesnt work out

File: deCC_zi_00027_.png (2.85 MB, 2048x1280)

2.85 MB PNG

>>107761983
>Sorry about the deadness.
oh, dont be. I like not having anything to do. then I have ultimate agency to do whatever I want

>>107762143
>average comfy user

File: 00037-comfyui-cfg=4.0-ste(...).jpg (247 KB, 928x1664)

247 KB JPG

>>107762346
that's a sick helmet.

>>107753072
Cute lesbian femcel.

File: deCC_zi_00030_.png (2.86 MB, 2048x1280)

2.86 MB PNG

>>107762476
will trade helmets for cute 1girls
great bargain. act now

File: 00057-comfyui-cfg=4.0-ste(...).jpg (221 KB, 928x1664)

221 KB JPG

>>107762840
here you go. Now give me your helmet - kek.

18 year old female teen, she is blonde, long wavy hair, Caucasian, slim waist, large breasts, waif, professional photography,, A young girl stands confidently against an abstract, pastel-hued backdrop featuring floating geometric shapes and soft gradients transitioning smoothly into one another. The sky above appears serene, painted with gentle hues of blue and pink, suggesting either dawn or dusk. Below, lush green foliage sways gently, adding depth and tranquility to the scene. Her outfit consists of a light purple cardigan over a white top, paired with matching shorts and simple yet stylish sandals. A small yellow accessory adorns her head, contrasting subtly with her long brown hair tied neatly at the back. She strikes a casual stance, leaning slightly forward while raising both index fingers towards herself—a gesture often associated with confidence or agreement—her gaze fixed directly ahead, embodying poise and self-assuredness.

anyone know some "lora clothing" or settings for khoya/one trainer for training clothing?

File: file.png (143 KB, 1476x898)

143 KB PNG

big PRs 4 u
this has been a lot of work desu
still lots more to do on amd support but excellent progress
downsampling/upsampling modules were annoying too, only like one thing left to do for that though

File: deCC_zi_00033_.png (2.71 MB, 2048x1280)

2.71 MB PNG

>>107762929
hehe, scammed
I didn't even prompt for the helmet. but here's the prompt, maybe you can find something interesting in there

1980s analog film inspired by blade runner, film scene from cult classic Atompunk movie directed in the style of Gaspar Noé, filmed on 35mm

dystopian industrial cyberpunk aesthetic with a cinematic framing, retro-futuristic 1980s scifi movie with film grain, gritty movie scene with interesting contrast

cybernetic with two-shot with hacker

full body pose with an expansive cyberpunk background, cinematic  shot with ambient lighting and soft colors, depth of field

pixel corruption and payment processing and robotic police

sexy skin-tight sleeveless uniform with cleavage and exposed midriff, detailed skin texture with perfect detail , unkempt messy hair hair, she is partially out of frame and looking behind her

bleach bypass

expansive wide angle cinematic shot of an expansive cyberpunk future

I'm gonna borrow some stuff from yours and see what pops out

>>107763055
damn those are, in fact, big PRs

File: Z-00052-First person shoo(...).png (855 KB, 768x1152)

855 KB PNG

Afternoon anons

File: 00229-1924610334.jpg (1.15 MB, 2304x1792)

1.15 MB JPG

>>107763415
afternoon

File: deCL_zi_00001_.png (2.58 MB, 2048x1280)

2.58 MB PNG

>>107762929
>raising both index fingers
z-image is very ltieral

File: 00074-comfyui-cfg=4.0-ste(...).jpg (264 KB, 1664x928)

264 KB JPG

>>107763990
The helmet is almost there.

>>107759566
beautiful

>>107764085
kinda want this as my wallpaper but it's not the right ratio

File: 7c559551-bf47-4acb-a02b-e(...).jpg (2.5 MB, 2752x1536)

2.5 MB JPG

>>107764197
ez

>>107764398
saved.
now to get randomly jumpscared when i forget about it.

File: 1767566695703_955a1116-e7(...).png (827 KB, 768x1152)

827 KB PNG

>"Neighbor fort quokka is approaching"
>Let him in?
>Yes >No

File: deCL_zi_00013_.png (2.96 MB, 2048x1280)

2.96 MB PNG

>>107764468
he looks untrustworthy

File: 2161b676-83e1-4cbf-b415-6(...).jpg (2.4 MB, 1696x2528)

2.4 MB JPG

>>107764468
>>107764500
to the dungeons with him!

File: 3789cc32-bffc-47ad-8d4f-2(...).jpg (2.47 MB, 2528x1696)

2.47 MB JPG

Another smuggler brought to justice by the City Guard.

File: deCL_zi_00018_.png (2.76 MB, 2048x1280)

2.76 MB PNG

>>107764626
wait, he was smuggling delicious treats..
I feel bad now. I love treats.....

File: dreamscape-zit-2026-01-04(...).png (2.47 MB, 1792x1024)

2.47 MB PNG

>>107764656
that hair... retro future 80s is cool tho

File: deCL_zi_00020_.png (3.15 MB, 2048x1280)

3.15 MB PNG

>>107764837
first zimg lora Im trying. its a 'casette futurism' lora but it mostly just 80s-ifies everything

File: 1767570326473_f466d1ba-5f(...).png (874 KB, 768x1152)

874 KB PNG

>>107764626
He looks like he needs a hug

File: deCL_zi_00029_.png (2.86 MB, 2048x1280)

2.86 MB PNG

>>107765822
I'm already depressed about it
I also somehow reset my local branch and lost a ton of work. I dont even know how I did that, and my IDE wont revert either. fml

File: 00000028-648786605125416-(...).jpg (2.01 MB, 3412x2560)

2.01 MB JPG

>>107765822
dont remind me

File: 00072-comfyui-cfg=4.0-ste(...).jpg (262 KB, 928x1664)

262 KB JPG

>>107765822
Mondays suck
>>107765838
>somehow reset my local branch
That sucks. Once I wrote a small app to capture data. The client spent 3 weeks entering all their data but the database was corrupted and they had to start over. I felt bad because I wrote the code but didn't bother checking entries were in fact written properly.

File: deCL_zi_00030_.png (2.68 MB, 2048x1280)

2.68 MB PNG

>>107765912
copilot found a bunch of "dangling blobs" and its trying to restore. I feel like its bullshitting, but we'll see
>The client spent 3 weeks entering all their data but the database was corrupted and they had to start over.
SisyphusDB

File: 00237-172195926.jpg (1.22 MB, 2560x1536)

1.22 MB JPG

gn

File: deCL_zi_00031_.png (2.74 MB, 2048x1280)

2.74 MB PNG

>>107766130
gn

File: 00000034-966550007587128-(...).jpg (1.96 MB, 3412x2560)

1.96 MB JPG

>>107766130
gn

File: deCL_zi_00032_.png (2.87 MB, 2048x1280)

2.87 MB PNG

baker san

a million captcha attempts later
>>107766236
>>107766236
>>107766236

File: 00000036-110145213216701-(...).jpg (2.08 MB, 3412x2560)

2.08 MB JPG

last