MiniMax H3 reference images: first five free, then $0.10 each on Sume

On Sume, MiniMax H3 (minimax-h3) reference-to-video includes the first five reference images in the per-second price and charges $0.08 provider list for each image after that. With the 1.25 margin that is $0.10 per extra image. The cap is nine reference images, so the most you can add is four extras: $0.40 billed. A 5-second 768p clip with nine references costs $0.775, against $0.375 with five or fewer.

This add-on comes from the Sume pricing package and applies only to MiniMax H3 reference-to-video requests, as of 2026-10-08. MiniMax H3 Max does not carry it.

The arithmetic

The video part is the same as a plain clip: 5 seconds x $0.06 list = $0.30 at 768p. Each extra image adds $0.08 list. With nine images, the list is $0.30 + 4 x $0.08 = $0.62, and $0.62 x 1.25 = $0.775.

At 480p the base is 5 x $0.05 = $0.25 list, so nine images give $0.25 + $0.32 = $0.57 list and $0.7125 billed.

Why H3 Max is different

For MiniMax H3 Max the catalog says reference-to-video is billed as output seconds at list x 1.25, and that the provider's reference-token overage is not reserved. The same nine-image, 5-second 768p job on H3 Max is 5 x $0.08 x 1.25 = $0.50. That is cheaper than H3 with nine images ($0.775) even though H3 Max has the higher per-second rate, so the extras flip the order.

With five or fewer images the order is the usual one: H3 is $0.375 and H3 Max is $0.50.

Batch example

A batch of 50 clips with nine references each on H3 at 768p costs 50 x $0.775 = $38.75. The same batch capped at five references per clip costs 50 x $0.375 = $18.75. The four extra images cost $20.00 across the batch, which is worth it only when the extra images change the result.

Reference limits that apply

The vendor description of H3 Max on fal calls it a post-trained variant of MiniMax H3 (fal page, read 2026-10-08). Check the live price for your key with GET /v1/video-router/models before you plan a large batch.

• Up to 9 reference images, up to 3 reference videos (2 to 15 s each, 15 s combined) and up to 3 reference audio clips (2 to 15 s each, 15 s combined).
• Images, videos and audio together: 12 or fewer. Audio cannot be the only reference.
• Duration 5 to 15 seconds. Native resolutions 480p and 768p.
• Reference images of H3 are counted by the pricing code only on reference-to-video requests, not on plain image-to-video.