AI kids story video generator: scenes of 2 to 30 seconds

Treat each page of the story as one scene and one request. Sume's video docs list wan-3.0 at 2 to 30 seconds per scene, so a short beat can be 2 seconds and a long one 30. Join the scenes into the finished story with Timeline 1.0, which takes 1 to 200 video slots.

Sources: Sume's video docs and Timeline docs, plus the Runway Dev changelog entry for WAN 3.0, all read 2026-10-01.

What scene lengths does each model allow?

Pick the model by the scene length your page needs. The docs state both ranges.

How do I keep the characters consistent?

Sume's docs say audio and video references are honored by Wan 3.0, and the model's catalog entry lists which input_references types it accepts. Runway's changelog describes WAN 3.0 as built for reference-driven shots, where you pass images, video or audio to keep characters and style consistent. Start from one character image and send it with every scene's request.

How do I join the scenes into one video?

Import each scene with POST /v1/media-imports, since Timeline accepts only this workspace's media.sume.com files. Put the narration or a music track on the audio spine (audio.duration_seconds 1 to 1800), and add one video[] slot per scene. Slots can be stills too, so a cover page can sit between scenes. Layout of a longer assembly: assemble a long-form video with Timeline.

What should I check before publishing?

Watch every scene yourself: generated video can change a character's look between requests, and this post makes no claim that a model suits a particular age group. Run the unbilled Timeline plan first so the segment count and duration match your page count.