YT · YOUTUBE// 4h agoMODEL RELEASE

Motif-Video 2B tops open video benchmarks

Motif-Video 2B is a compact text-to-video and image-to-video diffusion transformer that aims to win on architecture and training efficiency instead of brute-force scale. The release emphasizes a micro-budget training recipe, 720p generation, and top-tier open-source benchmark performance, making it a notable entry in the open video-generation race.

// ANALYSIS

The interesting part is that this is not just another bigger video model; it is a deliberately small release that still claims state-of-the-art open-source results.

–Trained on fewer than 10M clips and under 100,000 H200 GPU-hours.
–Uses a 2.0B-parameter flow-matching diffusion transformer with role-separated stages and shared cross-attention.
–Reports 83.76% on VBench, described as the highest score among open-source video generators.
–Supports both text-to-video and image-to-video in one checkpoint, with output up to 720p and 121 frames.
–The release is framed as open and reproducible, with weights and code promised to follow.

// TAGS

video-genopen-sourcebenchmarkresearchmultimodal

DISCOVERED

4h ago

2026-04-19

PUBLISHED

4h ago

2026-04-19

RELEVANCE

8/ 10

AUTHOR

AI Search