Animate 12 storybook pages as 5-second clips: Omni or MiniMax H3

For a 12-page picture book with a 5-second clip per page, MiniMax H3 at 768p costs $4.50 on Sume and Gemini Omni Flash 1.1 at 720p costs $7.50. Both take a still image as the first frame, and both fit 5 seconds: Omni accepts 3 to 10 seconds and H3 accepts 5 to 15, so 5 seconds is H3's minimum.

What each model takes

Both models are catalog ids on the Video Router. Omni takes image_url (plus an optional end_image_url) and reference images, up to 10, for 3 to 10 seconds at 360p, 720p, 1080p or 4K in 16:9 or 9:16. H3 is 5 to 15 seconds at native 480p or 768p; the Sume docs say 768p is first-class and 720p is not a value for it. Both render native audio that cannot be turned off.

Why page art needs a motion prompt

Google's docs say that a vague prompt such as "make it move" gives weaker results than a description of camera movement, subject motion and environmental effects. For a storybook page, write one sentence for each: a slow push toward the fox, the fox's ears twitching, leaves drifting across the foreground. Keep it to what the illustration already contains.

Keep the book consistent

Pick one tier for the whole book. Mixing 360p drafts and 720p finals in one reading experience shows as a change in sharpness from page to page. A good split is to run all 12 pages at 360p ($2.25) as a motion test, redo only the weak ones, then run the full set at the final tier.

Audio is the other consistency risk. Because each clip comes with its own generated sound, twelve clips will have twelve different soundscapes. For a read-aloud book, record or generate one narration track and put it on the timeline's audio spine, with the page clips as ordered video slots at $0.10 per output minute.

Where the time goes

Twelve requests are twelve separate jobs. Send them with the async mode and a distinct Idempotency-Key per page, such as book-page-01 to book-page-12, so a retry after a network error does not create and bill a duplicate clip. Keep the page number in your own records next to each request id.

Review the clips as a set before you pay for finals. A page where the animal drifts off the illustration style is cheaper to catch at $0.1875 than at $0.625.

Pick by the picture, not only the price

Illustrations with several characters that must stay recognizable benefit from Omni's reference images, which accept up to 10 stills and are addressed as <IMAGE_REF_0> and so on. For single-subject pages with no need for references, the cheaper H3 768p is the practical default. See the H3 versus H3 Max guide if you also need the faster variant, and turning a drawing into live-action footage for the opposite brief.