Self-trained ref2vid model generates 15s clips with reference images in ~6.9s
Kyrannio · x · 2026-10-30
Developer Kyrannio reports his posttrained reference-to-video model now generates 15-second clips incorporating reference image media in about 6.9 seconds (8-GPU test, measured 6.94s). Jumbled dialogue still needs adjustment, but speed keeps improving; he notes compute is getting scarcer.
Related event: Self-Trained ref2vid Model Generates 15-Second Video in Seconds(2 posts)→
More from Multimodal
- AI act Stereo Drift hits 23k monthly Spotify listeners, new album coming — measure_plan · 2026-10-30
- HyperFrames + HeyGen: decent video from a basic prompt in ~10 minutes — toolstelegraph · 2026-10-30
- Hailuo MCP + Claude: chaining motion graphics generation through an agent — aziz4ai · 2026-10-30
- AI short film Nexus ep 7 cost $28,775 to make, creator shares full breakdown — NightsRadiant · 2026-10-30
- LayerShop: open-source Photoshop-style editor with ComfyUI local AI on layers — HonestTicket6013 · 2026-10-30
- AI video prompt goes viral: rally car racing a collapsing volcanic mountainside — umesh_ai · 2026-10-30