AI children's story video: 8 Omni scenes for $8.00

A children's story video of about a minute is eight scenes of 8 seconds each, and on Sume it costs $8.00 at 720p with Gemini Omni Flash 1.1 ($0.125 per second). Keep the characters illustrated, not photographed: Google's Omni page, read 2026-10-08, says uploading and editing images that contain minors is not supported for users in the EEA, Switzerland and the UK, and that certain recognizable people are also refused.

The same character in every scene is the harder part. Omni on Sume takes up to 10 reference images per request, which is enough to carry one character sheet through all eight scenes.

The plan

Write the story as eight beats, one per request, and send the same character-sheet image as a reference in each. Refer to it in the prompt as <IMAGE_REF_0>; the tokens are 0-based and follow list order.

• Scene prompt: one action, one setting, one camera move, and No dialogue if you plan to add narration later.
• References: the character sheet first, a background plate second.
• Draft at 360p ($0.30 for 8 seconds), keep the scenes that hold the character, then re-run only those at 720p.
• Join: sequence the eight clips in Timeline 1.0 and lay a voiceover over them.

Cost of the whole book

The numbers assume exactly eight 8-second scenes and no re-rolls.

Where Veo 3.1 would differ

Sume does not list Veo 3.1. Google's Veo page allows up to 3 reference images, and says 8 seconds is required when reference images are used. Veo 3.1 Fast lists $0.10 per second at 720p on Google's pricing page, so eight 8-second scenes would be $6.40 direct, before any re-rolls. Veo's three-image limit is tighter than Omni's ten on Sume.

All Omni output has native audio, and the audio cannot be turned off. Music and effects in the prompt are fine for a picture book; for a read-aloud, ask for ambient sound only and keep the voice for the Timeline.

Before you share it

Check the clips for changed faces between scenes, extra fingers and garbled signs. Google says every Omni video carries SynthID, which is invisible and detectable in tools such as the SynthID Detector, so the file is marked as AI-made whether or not you say so.

Voice and consistency

Omni makes its own sound but takes no audio file, so narration is recorded separately and laid under the scenes. For the same character in every scene, send a drawn character sheet as a reference image in each request; Sume accepts up to 10 references per Omni call.

Submit with an Idempotency-Key header so a retry after a dropped connection returns the original job instead of a second charge; the same key with a different payload returns 409. Then poll GET /v1/jobs/:id/status and fetch GET /v1/jobs/:id/result when it finishes.