• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI
    Announcement

    LoGo aims to improve 3D consistency in long AI-generated videos

    A researcher describes a spatially localized 3D reward model designed to show where generated scenes go wrong.

    GG
    ZM
    4 Sources, ,

    TLDR

    LoGo was introduced as a World Labs internship project focused on post-training world models. Its creator says reward design matters greatly for 3D consistency in long generated videos. Another researcher says scenes can morph and objects can change or disappear as videos get longer; they describe a spatially localized 3D reward model intended to show the video model where things went wrong.

    Combined views

    7.8K

    4 Sources, first seen 2h ago

    Combined views

    7.8K

    4 Sources, first seen 2h ago

    102 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    2h ago
    first seen 2h ago
    102 likes
    2 comments
    55 saves
    15 reposts
    2 comments
    55 saves
    15 reposts

    4 Sources

    @ziqi__maVery excited to share LoGo, my internship project at @theworldlabs on post-training world models! We found that reward design matters a great deal in post-training long-horizon video gen for 3D consistency, echoing what we see in other domains, e.g. LLM reasoning. Check out more 👉 https://ziqi-ma.github.io/logo-website/2h
    @georgiagkioxariRT @ziqi__ma: Very excited to share LoGo, my internship project at @theworldlabs on post-training world models! We found that reward design…2h
    @gowthami_sMost RL techniques in image/video land are very naive, assigning same reward for all the tokens in the trajectory. This is suboptimal and doesn’t really address local artifacts. We address this with our awesome Local-Global (LoGo) reward, simple idea that works and we tested it on various models and different RL algos (diffusionNFT and GRPO). Once again simple intuition driven ideas work. 🚀2h
    Featured Source

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    4 Sources

    @ziqi__maVery excited to share LoGo, my internship project at @theworldlabs on post-training world models! We found that reward design matters a great deal in post-training long-horizon video gen for 3D consistency, echoing what we see in other domains, e.g. LLM reasoning. Check out more 👉 https://ziqi-ma.github.io/logo-website/2h
    @georgiagkioxariRT @ziqi__ma: Very excited to share LoGo, my internship project at @theworldlabs on post-training world models! We found that reward design…2h
    @gowthami_sMost RL techniques in image/video land are very naive, assigning same reward for all the tokens in the trajectory. This is suboptimal and doesn’t really address local artifacts. We address this with our awesome Local-Global (LoGo) reward, simple idea that works and we tested it on various models and different RL algos (diffusionNFT and GRPO). Once again simple intuition driven ideas work. 🚀2h