MiniMax H3 Max 1080p costs twice 768p: the three tiers priced

MiniMax H3 Max (minimax-h3-max) costs $0.0625 per second at 480p, $0.10 at 768p and $0.20 at 1080p on Sume, for 5 to 15 seconds. The docs describe 1080p as a latent refinement from native 768p, so the double price buys a refined output rather than a different native render.

Per clip

Sume bills the provider list price times 1.25 for H3 Max, as for every model, and reserves it at submit.

What it accepts

H3 Max does text-to-video, first and last frame image-to-video, and reference-to-video, with native stereo audio. It reports image, video and audio in supported_input_references.

Because 1080p is a refinement of the 768p render, a reasonable workflow is to review at 768p and spend the extra $0.10 per second only on the take you will publish.

Two-step request

Same prompt, only resolution changes.

# draft
curl -X POST https://api.sume.com/v1/videos -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"minimax-h3-max","prompt":"Skater at dusk, low angle","duration":10,"resolution":"768p"}'
# final
curl -X POST https://api.sume.com/v1/videos -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"minimax-h3-max","prompt":"Skater at dusk, low angle","duration":10,"resolution":"1080p"}'

Budget view

Fifty 10-second clips cost $50.00 at 768p and $100.00 at 1080p. A split that reviews fifty at 768p and re-renders five at 1080p costs $60.00.

Sume reserves the full price at submit, so the clip price must be free in the balance when you send the request. If it is not, the call fails with 402 insufficient_credits before any provider work starts.

H3 Max is also available as a lip-sync model on a separate Sume route with its own duration window; the table above is for the text, frame and reference generation route only.

Check the first frames of 1080p image-to-video

The Sume catalog notes that some 1080p image-to-video outputs from this model can have a baked crop jump after the first two frames. Look at the opening frames of a 1080p image-to-video clip before you assemble it. Only if you confirm the jump, start the clip at the first stable frame with source_in on the Timeline slot (2/24 of a second for the 24 fps clips observed), and keep enough source for any transition. Fit cover does not remove a baked transient.