Community

482 community available in the Stable Diffusion directory

Google's new AI algorithm reduces memory 6x and increases speed 8x

https://arstechnica.com/ai/2026/03/google-says-new-turboquant-compression-can-lower-ai-memory-usage-without-sacrificing-quality/

pheonis2

1,569

QR Code ControlNet

Why has no one created a QR Monster ControlNet for any of the newer models? I feel like this was the best ControlNet. Canny and depth are just not the same.

flasticpeet

1,365

Thank you Chinese devs for providing for the community if it not for them we'll be still stuck at stable diffusion 1.5

dead-supernova

1,300

Never forget…

ShadowBoxingBabies

2,230

Here it is boys, Z Base

Link: https://huggingface.co/Tongyi-MAI/Z-Image Comfy https://huggingface.co/Comfy-Org/z\_image/tree/main/split\_files/diffusion\_models

Altruistic_Heat_9531

1,576

Z-Image + Qwen Image Edit 2511 + Wan 2.2 + MMAudio

https://youtu.be/54IxX6FtKg8 A year ago, I never imagined I’d be able to generate a video like this on my own computer. (5070ti gpu) It’s still rough around the edges, but I wanted to share it anyway. All sound effects, excluding the background music, were generated with MMAudio, and the video was upscaled from 720p to 1080p using SeedVR2.

Budget_Stop9989

1,406

LTX-2 vs. Wan 2.2 - The Anime Series

theNivda

1,403

Surgical masking with Wan 2.2 Animate in ComfyUI

Surgical masking lets you preserve the original scene’s performance and image quality, keeping everything intact while only generating the new object, in this case Wolverine's mask. For this, I used Kijai’s workflow and added an input video node into the Blockify masking node with my mask. https://github.com/kijai/ComfyUI-WanVideoWrapper/blob/main/example\_workflows/wanvideo\_WanAnimate\_preprocess\_example\_02.json

enigmatic_e

2,217

LTX-2 I2V isn't perfect, but it's still awesome. (My specs: 16 GB VRAM, 64 GB RAM)

Hey guys, ever since LTX-2 dropped I’ve tried pretty much every workflow out there, but my results were always either just a slowly zooming image (with sound), or a video with that weird white grid all over it. I finally managed to find a setup that actually works for me, and hopefully it’ll work for you too if you give it a try. All you need to do is add --novram to the run\_nvidia\_gpu.bat file and then run my workflow. It’s an I2V workflow and I’m using the fp8 version of the model. All the start images I used to generate the videos were made with Z-Image Turbo. My impressions of LTX-2: Honestly, I’m kind of shocked by how good it is. It’s fast (Full HD + 8s or HD + 15s takes around 7–8 minutes on my setup), the motion feels natural, lip sync is great, and the fact that I can sometimes generate Full HD quality on my own PC is something I never even dreamed of. But… :D There’s still plenty of room for improvement. Face consistency is pretty weak. Actually, consistency in general is weak across the board. The audio can occasionally surprise you, but most of the time it doesn’t sound very good. With faster motion, morphing is clearly visible, and fine details (like teeth) are almost always ugly and deformed. Even so, I love this model, and we can only be grateful that we get to play with it. By the way, the shots in my video are cherry-picked. I wanted to show the very best results I managed to get, and prove that this level of output is possible. Workflow: https://drive.google.com/file/d/1VYrKf7jq52BIi43mZpsP8QCypr9oHtCO/view?usp=sharing

yanokusnir

2,539

I’m the Co-founder & CEO of Lightricks. We just open-sourced LTX-2, a production-ready audio-video AI model. AMA.

Hi everyone. I’m Zeev Farbman, Co-founder & CEO of Lightricks. I’ve spent the last few years working closely with our team on LTX-2, a production-ready audio–video foundation model. This week, we did a full open-source release of LTX-2, including weights, code, a trainer, benchmarks, LoRAs, and documentation. Open releases of multimodal models are rare, and when they do happen, they’re often hard to run or hard to reproduce. We built LTX-2 to be something you can actually use: it runs locally on consumer GPUs and powers real products at Lightricks. I’m here to answer questions about: Why we decided to open-source LTX-2 What it took ship an open, production-ready AI model Tradeoffs around quality, efficiency, and control Where we think open multimodal models are going next Roadmap and plans Ask me anything! I’ll answer as many questions as I can, with some help from the LTX-2 team. Verification: Lightricks CEO Zeev Farbman >The volume of questions was beyond all expectations! Closing this down so we have a chance to catch up on the remaining ones. > >Thanks everyone for all your great questions and feedback. More to come soon!

ltx_model

1,730

SVI 2.0 Pro for Wan 2.2 is amazing, allowing infinite length videos with no visible transitions. This took only 340 seconds to generate, 1280x720 continuous 20 seconds long video, fully open source. Someone tell James Cameron he can get Avatar 4 done sooner and cheaper.

I used workflow and custom nodes from wallen0322: https://github.com/wallen0322/ComfyUI-Wan22FMLF/blob/main/example\_workflows/SVI%20pro.json

Fresh_Diffusor

2,171

New implementation for long videos on wan 2.2 preview

UPDATE: Its out now: Github: https://github.com/shootthesound/comfyUI-LongLook Tutorial: https://www.youtube.com/watch?v=wZgoklsVplc I should I’ll be able to get this all up on GitHub tomorrow (27th December) with this workflow and docs and credits to the scientific paper I used to help me - Happy Christmas all - Pete

shootthesound

1,527

Former 3D Animator trying out AI, Is the consistency getting there?

Attempting to merge 3D models/animation with AI realism. Greetings from my workspace. I come from a background of traditional 3D modeling. Lately, I have been dedicating my time to a new experiment. This video is a complex mix of tools, not only ComfyUI. To achieve this result, I fed my own 3D renders into the system to train a custom LoRA. My goal is to keep the "soul" of the 3D character while giving her the realism of AI. I am trying to bridge the gap between these two worlds. Honest feedback is appreciated. Does she move like a human? Or does the illusion break? (Edit: some like my work, wants to see more, well look im into ai like 3months only, i will post but in moderation, for now i just started posting i have not much social precence but it seems people like the style, below are the social media if i post) IG : https://www.instagram.com/bankruptkyun/ X/twitter : https://x.com/BankruptKyun All Social: https://linktr.ee/BankruptKyun (personally i dont want my 3D+Ai Projects to be labeled as a slop, as such i will post in bit moderation. Quality>Qunatity) As for workflow 1. pose: i use my 3d models as a reference to feed the ai the exact pose i want. 2. skin: i feed skin texture references from my offline library (i have about 20tb of hyperrealistic texture maps i collected). 3. style: i mix comfyui with qwen to draw out the "anime-ish" feel. 4. face/hair: i use a custom anime-style lora here. this takes a lot of iterations to get right. 5. refinement: i regenerate the face and clothing many times using specific cosplay & videogame references. 6. video: this is the hardest part. i am using a home-brewed lora on comfyui for movement, but as you can see, i can only manage stable clips of about 6 seconds right now, which i merged together. i am still learning things and mixing things that works in simple manner, i was not very confident to post this but posted still on a whim. People loved it, ans asked for a workflow well i dont have a workflow as per say its just 3D model + ai LORA of anime&custom female models+ Personalised 20TB of Hyper realistic Skin Textures + My colour grading skills = good outcome.) Thanks to all who are liking it or Loved it. Last update to clearify my noob behvirial workflow.https://www.reddit.com/r/StableDiffusion/comments/1pwlt52/former\_3d\_animator\_here\_again\_clearing\_up\_some/

BankruptKun

4,542

Time-to-Move + Wan 2.2 Test

Made this using mickmumpitz's ComfyUI workflow that lets you animate movement by manually shifting objects or images in the scene. I tested both my higher quality camera and my iPhone, and for this demo I chose the lower quality footage with imperfect lighting. That roughness made it feel more grounded, almost like the movement was captured naturally in real life. I might do another version with higher quality footage later, just to try a different approach. Here's mickmumpitz's tutorial if anyone is interested: https://youtu.be/pUb58eAZ3pc?si=EEcF3XPBRyXPH1BX

enigmatic_e

6,053

Z-Image + SCAIL (Multi-Char)

I noticed SCAIL poses feel genuinely 3D, not flat. Depth and body orientation hold up way better than Wan Animate or SteadyDancer, 385f @ 736×1280, 6 steps took around 26 min on RTX 5090 ..

Better-Interview-793

1,818

This is your ai girlfriend

Ha8lpo321

3,954

Wan SCAIL is TOP!!

3d pose following and camera

smereces

1,389

Z-Image on 3060, 30 sec per gen. I'm impressed

Z-Image + WAN for video

Mobile_Vegetable7632

2,336

when an upscaler is so good it feels illegal

I'm absolutely in love with SeedVR2 and the FP16 model. Honestly, it's the best upscaler I've ever used. It keeps the image exactly as it is. no weird artifacts, no distortion, nothing. Just super clean results. I tried GGUF before, but it messed with the skin a lot. FP8 didn’t work for me either because it added those tiling grids to the image. Since the models get downloaded directly through the workflow, you don’t have to grab anything manually. Just be aware that the first image will take a bit longer. I'm just using the standard SeedVR2 workflow here, nothing fancy. I only added an extra node so I can upscale multiple images in a row. The base image was generated with Z-Image, and I'm running this on a 5090, so I can’t say how well it performs on other GPUs. For me, it takes about 38 seconds to upscale an image. Here’s the workflow: https://pastebin.com/V45m29sF Test image: https://imgur.com/a/test-image-JZxyeGd Model if you want to manually download it: https://huggingface.co/numz/SeedVR2\_comfyUI/blob/main/seedvr2\_ema\_7b\_fp16.safetensors Custom nodes: for the vram cache nodes (It doesn't need to be installed, but I would recommend it, especially if you work in batches) https://github.com/yolain/ComfyUI-Easy-Use.git Seedvr2 Nodes https://github.com/numz/ComfyUI-SeedVR2\_VideoUpscaler.git For the "imagelist\_from\_dir" node https://github.com/ltdrdata/ComfyUI-Inspire-Pack

Ok-Page5607

2,144

I did all this using 4GB VRAM and 16 GB RAM

Hello, I was wondering what can be done with AI these days on a low-end computer, so I tested it on my older laptop with 4GB VRAM (NVIDIA Geforce GTX 1050 Ti) and 16 GB RAM (Intel Core i7-8750H). I used Z-Image Turbo to generate the images. At first I was using the gguf version (Q3) and the images looked good, but then I came across an all-in-one model (https://huggingface.co/SeeSee21/Z-Image-Turbo-AIO) that generated better quality and faster - thanks to the author for his work. I generated images of size 1024 x 576 px and it took a little over 2 minutes per image. (\~02:06) My workflow (Z-Image Turbo AIO fp8): https://drive.google.com/file/d/1CdATmuiiJYgJLz8qdlcDzosWGNMdsCWj/view?usp=sharing I used Wan 2.2 5b to generate the videos. It was a real struggle until I figured out how to set it up properly so that the videos didn't just have slow motion and so that the generation didn't take forever. The 5b model is weird, sometimes it can surprise, sometimes the result is crap. But maybe I just still haven't figured out the right settings yet. Anyway, I used the fp16 model version in combination with two loras from Kijai (may God bless you, sir). Thanks to that, 4 steps were enough, but 1 video (1024 x 576 px; 97 frames) took 29 minutes to generate (decoding process alone took 17 minutes of that time). Honestly, I don't recommend trying it. :D You don't want to wait 30 minutes for a video to be generated, especially if maybe only 1 out of 3 attempts is usable. I did this to show that even with poor performance, it's possible to create something interesting. :) My workflow (Wan 2.2 5b fp16): https://drive.google.com/file/d/1JeHqlBDd49svq1BmVJyvspHYS11Yz0mU/view?usp=sharing Please share your experiences too. Thank you! :)

yanokusnir

2,865

It's your choice at end

dead-supernova

2,934

Z Image Turbo ControlNet released by Alibaba on HF

https://huggingface.co/alibaba-pai/Z-Image-Turbo-Fun-Controlnet-Union

an303042

1,858

No hard feelings

dead-supernova

1,890

Z-Image-Base and Z-Image-Edit are coming soon!

Z-Image-Base and Z-Image-Edit are coming soon! https://x.com/modelscope2022/status/1994315184840822880?s=46

tanzim31

1,351