What makes it different
Core ideas
Character memory
Characters persist across segments so a face or costume doesn’t drift shot to shot — the hardest problem in multi-shot AI video.
Reference-driven
References steer look and subject rather than relying on prompt wording alone.
Segment generation
Each segment generates independently, so you can regenerate one without redoing the episode.
Stitched export
Approved segments assemble into a single video.
Working in it
1
Start a project
Projects are saved and resumable — episodic work spans sessions.
2
Establish references and characters
Set these before generating in bulk. Consistency comes from the references, so getting them right first saves the most credits.
3
Generate segment by segment
Review each one. Regenerate the weak ones individually rather than restarting.
4
Export the episode
Stitch the approved segments into the finished video.
Credits
Every segment is a video generation and costs accordingly — an episode is the sum of its segments. Regenerating one segment costs one segment.Where it fits
- Cinema Studio — plan shots as a storyboard before generating
- Video Editor — a real timeline for precise trimming, captions, and layered edits
- Generate — single clips with full control
Autonomous orchestration and final assembly aren’t available through the API — Video Agent projects can be persisted programmatically, but the end-to-end run happens in the app. See what you can automate.
