MiniMax H3 reference images: what you pay past the fifth

On Sume, a minimax-h3 reference-to-video request carries the first 5 reference images at no extra charge, and every image after the fifth adds $0.10 to the reserved amount. The per-second video price is separate, and minimax-h3-max has no such add-on in Sume's pricing code.

The figures come from Sume's pricing code, which applies the house multiplier of 1.25 to the provider's list add-on, and the Video generation docs, read 2026-09-29. Sume adds a 5.5% agent fee by default.

How many reference images can one request carry?

Up to 9 images, with up to 3 videos and 3 audio clips and no more than 12 files in total. The surcharge only counts images, so a request with 9 images pays for 4 extra.

What do 6 to 9 images add?

When does a request count as reference-to-video?

In current code a request is reference-to-video when it carries any reference video or audio, more than one reference image, or one reference image with no first or last frame. A single image sent as first_frame in frame_images is image-to-video and never pays the add-on.

How do I avoid paying for extra images?

Give the model five or fewer images, and assign each a job in the prompt. If a sixth image adds little, fold its content into a prompt line instead. If you need more identities in frame, weigh the add-on against a second clip; the clip price is in MiniMax H3 pricing per second.