MiniMax H3 reference-to-video: five images free, then $0.08 each

On Sume's minimax-h3 id, a reference-to-video request is priced per output second plus a list add-on of $0.08 for each reference image after the first five. The catalog notes read: first 5 reference images free, each additional image $0.08 list. minimax-h3-max has no such line; its notes say reference-to-video is billed as output seconds at list × 1.25, and that fal's reference-token overage is not reserved.

That difference matters because both ids accept up to nine reference images, so a nine-image brief on minimax-h3 carries four paid images while the same brief on minimax-h3-max does not. The numbers below come from the Sume catalog constraints and the pricing code in the repository; check GET /v1/video-router/models/minimax-h3 for the live rate card before you budget.

What counts as an extra image

The pricing code counts the strings in reference_image_urls for the request and subtracts five. The fee applies only when the request is reference-to-video, meaning images, reference clips or reference audio drive the generation rather than a first or last frame. A request with fewer than six reference images pays only the per-second rate.

Limits stay the same on both MiniMax ids: at most 9 reference images, 3 reference videos and 3 reference audio files, and at most 12 reference files combined. The Video Router docs also say minimax-h3 accepts 5 to 15 seconds at native 480p or 768p, with 768p treated as first-class rather than as 720p.

Ten-second totals at 768p

The minimax-h3 list rate is $0.06 per second at 768p, per the catalog line dated 2026-08-25. Ten seconds is $0.60 of video. Each extra image adds $0.08. Sume bills list × 1.25 on every model, so the last column applies that multiplier to the whole list amount; if you budget conservatively, treat that column as the figure to plan around.

The arithmetic: 7 images means 2 extra, 2 × $0.08 = $0.16, so $0.60 + $0.16 = $0.76 list. 9 images means 4 extra, 4 × $0.08 = $0.32, so $0.92 list. Times 1.25 gives $0.75, $0.95 and $1.15.

When to switch ids

If the brief really needs seven or nine reference photos, compare minimax-h3-max at 768p. Its list rate is $0.08 per second at 768p (catalog line dated 2026-09-08), so ten seconds is $0.80 list and $1.00 at ×1.25 with no image add-on. The nine-image minimax-h3 clip at $1.15 is more expensive than that, while a five-image clip on minimax-h3 at $0.75 is cheaper.

Quality and speed are separate questions. The docs describe H3 Max as the faster 768p variant with a 1080p latent refinement, so test your own brief rather than choosing on price alone.

Trimming references is the cheapest fix: if two of the seven photos are near-duplicates, dropping them returns the clip to the free five-image tier on minimax-h3.

Request shape

A reference-to-video request on the Video Router sends the images as reference_image_urls, public HTTPS links:

``json { "model": "minimax-h3", "prompt": "The same sneaker rotating on a concrete plinth, soft studio light", "reference_image_urls": [ "https://example.com/shoe-1.jpg", "https://example.com/shoe-2.jpg", "https://example.com/shoe-3.jpg", "https://example.com/shoe-4.jpg", "https://example.com/shoe-5.jpg", "https://example.com/shoe-6.jpg" ], "resolution": "768p", "duration": 10 } ``

Six images means one paid image, $0.08 list, on top of the $0.60 for ten seconds.