Video Generation

Video generation — coming soon

Short-form text-to-video (2–10 second clips) plus image-to-video animation. Candidate models under evaluation: Wan 2.1, Mochi 1, Hunyuan Video. Longest horizon of all the roadmap items because it needs substantial validator GPU memory.

Planned capabilities
- Text-to-video (720p, 5–10s)
- Image-to-video (animate a single frame with a motion prompt)
- Video editing (prompt-based style change)

Availability: on the roadmap — video needs substantial validator GPU memory, so we'll light up a preview model first once capacity is there.