MiniMax-H3 ref2va

Headlines · 2

1. Oct 21
   AnyGameRenderer brings entity-consistent rendering to simple game footage

   Ruicong Liu and four coauthors present AnyGameRenderer, an arXiv paper describing a video-to-video generative renderer that upgrades simple game footage with characters, motion detail and effects while preserving layout, camera movement, entity positions and timing. The system accepts plain video without engine buffers, per-object class labels or low-poly proxies. It is implemented through LoRA fine-tuning of the 33B-parameter MiniMax-H3 ref2va model and trained on about 10,000 paired clips from agent-made games exported through GEAR, Unreal Engine and real gameplay recordings.

   Hugging Face · Papers · 🔥 0
2. Oct 13
   Oxford, Monash and Bristol researchers unveil Oneira, an interactive video world model

   A team from the University of Oxford, Monash University and the University of Bristol has proposed Oneira, an interactive video world model that separates "what happens in the world" from "how the frame is rendered," according to a 机器之心 report. Oneira has four parts: a Coding Agent that maintains a World State Table of each object's position and interaction state, an Engine that renders that table into a simplified conditional video, and a Video Generation Model that fills in real appearance, motion and interaction detail. The authors built InteractionBench, with 100 interaction cases and 50 memory cases, and compared Oneira with MiniMax-H3 Ref2VA, LingBot-World-V2, YUME 1.5 and AlayaWorld; they self-report gains of 29, 19 and 29 percentage points on whether an interaction occurs, timing accuracy and correct object, plus memory-case PSNR and SSIM improvements of 93% and 83% and an 18% gain in state retention. A project page, paper and code links were included in the report; all figures are the team's own and have not been independently reproduced.

   机器之心 · 🔥 0

Experience and discussion from the community

Share my MiniMax-H3 ref2va experienceAsk about MiniMax-H3 ref2va