#MAGREF
LightNews LightNews
mindsetware.bsky.social
🎯 What it does:

✅ Same person in different locations
✅ Two reference images at once
✅ Products stay accurate across scenes
✅ Facial features don't drift

#MAGREF #MultiSubjectVideo #VideoAI
June 20, 2025 at 4:47 PM
mindsetware.bsky.social
Paper: arxiv.org/abs/2505.23742
Code: github.com/MAGREF-Video...

🛠️ Already has ComfyUI nodes ready.

Pro-tips:

- White background reference images work best
- Use Light X2V LoRa to speed things up
- Detail retention is really good
GitHub - MAGREF-Video/MAGREF: Official implementation of MAGREF: Masked Guidance for Any-Reference Video Generation
Official implementation of MAGREF: Masked Guidance for Any-Reference Video Generation - MAGREF-Video/MAGREF
github.com
June 20, 2025 at 4:47 PM
mindsetware.bsky.social
I just discovered MAGREF.

This fine-tuned Wan 2.1 model keeps faces and objects consistent across video scenes.

Single subject or multi-reference mode - both work surprisingly well.
This image shows a demonstration of AI video generation technology. The left side displays reference inputs organized into three categories: "Single ID" (featuring a Van Gogh self-portrait), "Multi-ID" (showing headshots of a smiling man and woman), and "ID + Object + Background" (containing images of a cat, outdoor scene, and another person). The right side shows "Generated Video Frames" with four sequential frames for each scenario. The top sequence shows a man in a blue button-up shirt sitting at a wooden cafe table, holding coffee while looking at his phone. The middle sequence depicts two people in an art studio - a woman in a blue jacket standing behind a seated man in white, both near an easel. The bottom sequence features a man in a blue shirt and straw hat sitting outdoors, interacting with a cat.
June 20, 2025 at 4:47 PM
mengli512.bsky.social
ByteDance's MAGREF lets you turn ANY photo into a dynamic video—consistent faces, multi-character scenes, no face-swaps!
🎥✨ Einstein riding a bike? Turing animated? All possible. How would YOU use this tech?
#MAGREF #ByteDance #AI
aidisruption.ai/p/bytedances...
ByteDance's New AI Tool Lets Anyone Shine in Videos
ByteDance's MAGREF AI generates multi-subject videos from a single image—no face-swapping, full control over characters & scenes.
aidisruption.ai
June 13, 2025 at 3:55 PM
researchtrend.ai
[2025-05-30] 📚 Updates in #VOS

(1) <a href="https://researchtrend.ai/papers/2505.23742" class="hover:underline text-blue-600 dark:text-sky-400 no-card-link" target="_blank" rel="noopener" data-link="bsky">MAGREF: Masked Guidance for Any-Reference Video Generation
(2) MAGREF: Masked Guidance for Any-Reference Video Generation

🔍 More at researchtrend.ai/communities/VOS
May 30, 2025 at 11:22 AM
cscv-bot.bsky.social
https://github.com/MAGREF-Video/MAGREF [7/7 of https://arxiv.org/abs/2505.23742v1]
May 30, 2025 at 6:20 AM
cscv-bot.bsky.social
faces significant challenges in maintaining multi-subject consistency and ensuring high generation quality. In this paper, we propose MAGREF, a unified framework for any-reference video generation that introduces masked guidance to enable coherent [2/7 of https://arxiv.org/abs/2505.23742v1]
May 30, 2025 at 6:20 AM
cscv-bot.bsky.social
Deng, Guo, Yin, Fang, Yang, Wang, Yuan, Wang, Liu, Huang, Ma: MAGREF: Masked Guidance for Any-Reference Video Generation https://arxiv.org/abs/2505.23742 https://arxiv.org/pdf/2505.23742 https://arxiv.org/html/2505.23742
May 30, 2025 at 6:20 AM
arxiv-cs-cv.bsky.social
Yufan Deng, Xun Guo, Yuanyang Yin, Jacob Zhiyuan Fang, Yiding Yang, Yizhi Wang, Shenghai Yuan, Angtian Wang, Bo Liu, Haibin Huang, Chongyang Ma
MAGREF: Masked Guidance for Any-Reference Video Generation
https://arxiv.org/abs/2505.23742
May 30, 2025 at 6:11 AM