>mfw Resource news

03/01/2026

>Accelerating Masked Image Generation by Learning Latent Controlled Dynamics
https://github.com/Kaiwen-Zhu/MIGM-Shortcut

>Open-sourced a one-click ComfyUI setup for RTX 50-series on Windows
https://github.com/hiroki-abe-58/ComfyUI-Win-Blackwell

>stable-diffusion-webui-codex v0.2.0-alpha
https://github.com/sangoi-exe/stable-diffusion-webui-codex

>ComfyUI SeedVR2 Tiler
https://github.com/BacoHubo/ComfyUI_SeedVR2_Tiler

02/28/2026

>Anima 2B Style Explorer
https://github.com/ThetaCursed/Anima-Style-Explorer

>LoRWeB: Spanning the Visual Analogy Space with a Weight Basis of LoRAs
https://research.nvidia.com/labs/par/lorweb

>Nexa - Your On-the-Go ComfyUI Companion
https://github.com/Arif-salah/Nexa_comfyui

>Z-Image-Turbo Controlnet Union 2.1 version 2602
https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union-2.1

>FL PixelGen: Pixel-space diffusion text-to-image generation and LoRA training nodes
https://github.com/filliptm/ComfyUI-FL-PixelGen

>ComfyUI NKD Sigmas Curve
https://github.com/Nekodificador/ComfyUI-NKD-Sigmas-Curve

>Capable, Open, and Safe: Combating AI Misuse
https://bfl.ai/blog/capable-open-and-safe-combating-ai-misuse

02/27/2026

>WISER: Wider Search, Deeper Thinking, and Adaptive Fusion for Training-Free Zero-Shot Composed Image Retrieval
https://github.com/Physicsmile/WISER

>Aligning Few-Step Diffusion Models with Dense Reward Difference Learning
https://github.com/ZiyiZhang27/sdpo

>Huihui-Qwen3.5-27B-abliterated
https://huggingface.co/huihui-ai/Huihui-Qwen3.5-27B-abliterated

>ComfyUI Yedp Action Director
https://github.com/yedp123/ComfyUI-Yedp-Action-Director

>Nano Banana 2: Combining Pro capabilities with lightning-fast speed
https://blog.google/innovation-and-ai/technology/ai/nano-banana-2

>ComfyUI适配 FireRed-Image-Edit-1.0: general-purpose image editing
https://huggingface.co/FireRedTeam/FireRed-Image-Edit-1.0-ComfyUI

>mfw Research news

03/01/2026

>Mode Seeking meets Mean Seeking for Fast Long Video Generation
https://primecai.github.io/mmm

>SwitchCraft: Training-Free Multi-Event Video Generation with Attention Controls
https://arxiv.org/abs/2602.23956

>Diffusion Probe: Generated Image Result Prediction Using CNN Probes
https://arxiv.org/abs/2602.23783

>SenCache: Accelerating Diffusion Model Inference via Sensitivity-Aware Caching
https://arxiv.org/abs/2602.24208

>MSVBench: Towards Human-Level Evaluation of Multi-Shot Video Generation
https://arxiv.org/abs/2602.23969

>Enhancing Spatial Understanding in Image Generation via Reward Modeling
https://dagroup-pku.github.io/SpatialT2I

>DesignSense: A Human Preference Dataset and Reward Modeling Framework for Graphic Layout Generation
https://arxiv.org/abs/2602.23438

>Interpretable Debiasing of Vision-Language Models for Social Fairness
https://arxiv.org/abs/2602.24014

>Venus: Benchmarking and Empowering Multimodal Large Language Models for Aesthetic Guidance and Cropping
https://arxiv.org/abs/2602.23980

>A Difference-in-Difference Approach to Detecting AI-Generated Images
https://arxiv.org/abs/2602.23732

>Fixed Anchors Are Not Enough: Dynamic Retrieval and Persistent Homology for Dataset Distillation
https://arxiv.org/abs/2602.24144

>U-Mind: A Unified Framework for Real-Time Multimodal Interaction with Audiovisual Generation
https://arxiv.org/abs/2602.23739

>Joint Geometric and Trajectory Consistency Learning for One-Step Real-World Super-Resolution
https://arxiv.org/abs/2602.24240

>NAU-QMUL: Utilizing BERT and CLIP for Multi-modal AI-Generated Image Detection
https://arxiv.org/abs/2602.23863

>RAViT: Resolution-Adaptive Vision Transformer
https://arxiv.org/abs/2602.24159

>PixelRush: Ultra-Fast, Training-Free High-Resolution Image Generation via One-step Diffusion
https://arxiv.org/abs/2602.12769

>Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified T2I Generation
https://arxiv.org/abs/2602.13585

File: salt_in_your_mouth_alt.png (665 KB, 576x1024)

665 KB PNG

>>108281043
https://suno.com/s/CVdLn8Ay2Acckcy2

i miss schizo anon

File: 00000004-619152000979513-(...).jpg (1.44 MB, 3072x2048)

1.44 MB JPG

>>108284156
better but what does that have to do with /sdg/ lel
also some of those lyrics are a bit...forced

>>108284195
Forced soul vs artificial soul.

File: 00000005-393156861681559-(...).jpg (1.13 MB, 3072x2048)

1.13 MB JPG

>>108284224
how about natural soul

File: deCA_zi_00041_.png (3.41 MB, 2560x1152)

3.41 MB PNG

gm

>vibe killed

File: Z-00141-clean, crisp vect(...).png (1023 KB, 768x1152)

1023 KB PNG

Morning anons

File: 00000017-1095254098680946(...).jpg (1.97 MB, 3072x2048)

1.97 MB JPG

>>108285562
gm

File: deCA_zi_00042_.png (3.1 MB, 2560x1152)

3.1 MB PNG

>>108285562
gm
thumbnail made me think quokka had a zoomer broccoli cut fr fr no cap

File: paper-zit-2026-03-01_00187_.png (1.94 MB, 1792x1024)

1.94 MB PNG

>>108284195
i was just fuckin around and kinda liked it, only spent like 10 minutes on it.

Quite the ded shitter

File: 1772564771791_a535b62a-14(...).png (831 KB, 768x1152)

831 KB PNG

Officer Quokka
>>108285610
Kek, how is that cut even called?

File: 21e.jpg (486 KB, 873x1332)

486 KB JPG

>>108286450
I've only ever heard it called a broccoli cut

File: deCA_zi_00043_.jpg (1.18 MB, 2560x1152)

1.18 MB JPG

>>108286534
we can use AI to derive the next popular vegetable based haircut

File: 00011-3744966258.png (1.72 MB, 1120x1472)

1.72 MB PNG

>>108286593

File: deCA_zi_00045_.png (3.51 MB, 2560x1152)

3.51 MB PNG

ew gross, it replied to me

File: 00034-comfyui-cfg=1.0-ste(...).jpg (498 KB, 1200x1800)

498 KB JPG

good pm anon
nice gen up there.

File: deCA_zi_00047_.png (3.15 MB, 2560x1152)

3.15 MB PNG

>>108287764
cool alien dude

this site has seedance 2 btw, paid only btw but its real full seedance 2 https://yapper.so/i/CSJG2AYC96

File: 1741108368507388.png (1.22 MB, 1024x1024)

1.22 MB PNG

UNIT

File: deCA_zi_00049_.png (2.7 MB, 2560x1152)

2.7 MB PNG

is 2048 the max dimension limit for zimg? I've also been noticing many loras are trained at 1024 for some reason

File: paper-zit-2026-02-05_00098_.png (3.7 MB, 2304x1296)

3.7 MB PNG

>>108288368
When i was using 2304x1296, it did weird things like repeating stuff on the edges, images being off-center, etc. how much it matters really depends on the scene. even in these you've been doing if you look closely at the right side it's always just kind of... off from the rest of the image. pic rel is a good example, the right side it started putting another book for no reason. you can usually spot some artifacting in those sections too. i don't know what the max actually is, probably based on total megapixels desu

File: deCA_zi_00050_.png (3.93 MB, 2560x1152)

3.93 MB PNG

>>108288429
>the right side it's always just kind of... off
yeah, thats where the question came from. I was doing some other gens too that were doing similar things cuz they were too big. I'd gen smaller and upscale but zit isn't great in my upscaling workflow either cuz it artifacts a lot there too

>>108288449
nice
hero or villain?

File: models-zit-2026-03-03_00126_.png (2.45 MB, 1792x1024)

2.45 MB PNG

>>108288485
the weirdest part is that the rest of the image comes out great, it's like it starts another image tiled on the extra width

>>108289268
you're heckin' cute and valid
let no one tell you otherwise

File: deCD_zi_00003_.png (3.37 MB, 2688x1072)

3.37 MB PNG

>>108289384
can I also be cute an valid

File: deCD_zi_00005_.png (3.29 MB, 2688x1072)

3.29 MB PNG

>>108290133
nice watercolor effect

here I was hoping nigbo was finally hit by a bus

sadly I was wrong

File: deJS_zi_00006_.png (2.98 MB, 1328x1496)

2.98 MB PNG

I finally got this prompt doing decent stuff. still tons of errors tho

>>108290550
>still tons of errors tho
that's a signature in every one of your crap gens

File: deJS_zi_00007_.jpg (968 KB, 1792x1920)

968 KB JPG

>>108290550
where I was before

>>108290556
hahaha you got me xD

File: deCD_zi_00012_.png (3.54 MB, 2688x1072)

3.54 MB PNG

gn

File: Ffcfsq4XkAEx3rq.jpg (539 KB, 2048x1920)

539 KB JPG

I would like to do this.

i miss schizo anon

nano banana 2 is so good, i wish there was a way to jailbreak it

File: 000000_60800_.png (3.28 MB, 1098x1464)

3.28 MB PNG

G'mornin Anons, have a great day!

8giglet here, should i even bother trying to gen with wan on this machine or just pay up and use something like runpod
have no interest in actually buying a bigger gpu cause i'll probably get bored in a week

>>108292495
>>>/g/ldg

>>108293210
what exactly is the point of this general these days?

File: Z-00142- watercolor scene(...).png (1.45 MB, 768x1152)

1.45 MB PNG

Morning anons
>>108292495
I have 6gb, sadly there's really not many option for us as even the most optimized models would take a while to gen, it's not worth it, if you have the money, just pay for it.

File: models-zit-2026-03-03_00213_.png (2.64 MB, 1792x1024)

2.64 MB PNG

>>108292495
z image turbo works great on 8gb

>>108292495
ofc i didn't read the post, wan your fucked

>>108294024
Your gens look like shit
Do you really never try out new stuff? You post the same slop all the time and that's super boring

File: file.png (759 KB, 1024x1024)

759 KB PNG

>Ask it to make the background transparent
>does this

File: deCD_zi_00015_.png (3.16 MB, 2688x1072)

3.16 MB PNG

the official racism dataset

>Large-Scale Dataset and Benchmark for Skin Tone Classification in the Wild
https://arxiv.org/abs/2603.02475

File: deCD_zi_00016_.png (3.53 MB, 2688x1072)

3.53 MB PNG

>>108294476
https://github.com/lllyasviel/LayerDiffuse

File: 1772651370792_473a1f41-20(...).png (959 KB, 1152x768)

959 KB PNG

>>108294494
Holy based

File: 1772654662347_110ce0eb-3b(...).png (1.37 MB, 1024x1024)

1.37 MB PNG

>>108294476
>Pet box
>No breathing holes
Oh no

File: deCD_zi_00024_.png (3.24 MB, 2688x1072)

3.24 MB PNG

>>108295955
new lora status?

>>108295653
>>108296113
Hey debo please fuck off from /ldg/ you're not welcome in the real thread

File: deCD_zi_00026_.png (3.52 MB, 2688x1072)

3.52 MB PNG

>>108296277
sorry, sdg rules is that you must post a gen for your opinion to be considered

File: models-zit-2026-03-03_00106_.png (2.46 MB, 1792x1024)

2.46 MB PNG

that just sounds like an invitation

File: 00000007-883746406175674-(...).jpg (1.41 MB, 2048x3072)

1.41 MB JPG

>>108296113
idk why the z-turbo lora failed but the z-base is mostly ok, although it turns her asian more often than not
i'm too tired lately to try again

File: deCD_zi_00027_.png (3.35 MB, 2688x1072)

3.35 MB PNG

>>108296447
not sure if this is relevant but I saw this comfy node yesterday

Been using Z-Image Turbo and my LoRAs were working but something always felt off. Dug into it and turns out the issue is architectural, Z-Image Turbo uses fused QKV attention instead of separate to_q/to_k/to_v like most other models. So when you load a LoRA trained with the standard diffusers format, the default loader just can't find matching keys and quietly skips them. Same deal with the output projection (to_out.0 vs just out).

Basically your attention weights get thrown away and you're left with partial patches, which explains why things feel off but not completely broken.

So I made a node that handles the conversion automatically. It detects if the LoRA has separate Q/K/V, fuses them into the format Z-Image actually expects, and builds the correct key map using ComfyUI's own z_image_to_diffusers utility. Drop-in replacement, just swap the node.

Repo: https://github.com/capitan01R/Comfyui-ZiT-Lora-loader

File: 00000010-950786313398987-(...).jpg (1.89 MB, 2048x3072)

1.89 MB JPG

>>108296561
yah in my case it's more the lora itself than comfy (since i havent updated in a while), i just have to re-do it again more carefully, i just rushed it last time and all samples and images were just ugly blurry mess. i'll probably try again over the weekend. but the base lora works on z-turbo well enough for now
i did try that node and it made no difference to the z-turbo lora i made

File: 00000007-1046366095584942(...).jpg (2.54 MB, 3072x2048)

2.54 MB JPG

>>108297051
i noticed zturbo has a harder time with feet at certain angles compared to klein

File: 00040-comfyui-cfg=1.0-ste(...).jpg (230 KB, 1200x1800)

230 KB JPG

>>108297059
Hmm. Perhaps. I didn't try Klein enough to make a feet comparison. The feet other my z-turbo gens seems ok except for the occasional 6th toe.

File: 00000010-158767495711844-(...).jpg (2.07 MB, 2048x3072)

2.07 MB JPG

>>108297135
yah it's mainly kneeling or feet behind kind of poses that it seems to struggle with

File: 00082-comfyui-cfg=1.0-ste(...).jpg (240 KB, 1200x1800)

240 KB JPG

>>108297221
ah. Let me try a few with that pose.

File: 00000013-1004335600395132(...).jpg (1.13 MB, 2048x3072)

1.13 MB JPG

>>108297334
i admit it's difficult for most models, but z-image gives up too early

File: 00104-comfyui-cfg=1.0-ste(...).jpg (557 KB, 3600x1800)

557 KB JPG

>>108297367
Yeah, they sometimes turnout weird.

File: 00000028-1102400426908865(...).jpg (1.33 MB, 2048x3072)

1.33 MB JPG

>battlesuit turns vaguely egyptian
heh

File: 00000030-330329461752538-(...).jpg (1.34 MB, 3072x2048)

1.34 MB JPG

me in the back

File: 1772682843454_a9c9757b-51(...).png (1.54 MB, 1152x768)

1.54 MB PNG

Quokka and Wolfuokka

File: 00000039-166727297408429-(...).jpg (1.22 MB, 3072x2048)

1.22 MB JPG

>>108298329
tasmanian wolf?

new
>>108298416
>>108298416
>>108298416

File: 1772684546636_5899b235-49(...).png (1.1 MB, 1152x768)

1.1 MB PNG

>>108298419
Thanks :)