MiniMax H3 on Sume: 9 reference images, first 5 free, 4 extra $0.08

MiniMax H3 on Sume accepts up to 9 reference images in one clip. The catalog note says the first 5 reference images are free and each additional image is $0.08 at list, so 9 images carry 4 x $0.08 = $0.32 of reference list charge on top of the clip. Vidu announced Q4 Preview on 2026-10-07 with up to 15 image references, which is more than the 9 that Sume lists for H3.

The limits, side by side

Sume reads these limits from the catalog constraints of each model. The Vidu row comes from the vendor's launch release, not from Sume.

The H3 rules in full

For H3 and H3 Max, the sum of images, videos and audios must be 12 or less, and audio cannot be the only reference. Reference videos are 2 to 15 seconds each with a combined limit of 15 seconds; the same combined limit applies to audios. A clip is 5 to 15 seconds long, at native 480p or 768p.

The free-image note belongs to H3. For H3 Max, the catalog says that the reference-token overage from the provider is not reserved by Sume, and it bills output seconds at list x 1.25. This post does not state a Sume price for H3 extra images, because the catalog gives the $0.08 as a list figure only.

• 9 images: 5 free + 4 extra, 4 x 0.08 = $0.32 at list.
• 5 images or fewer: no extra reference charge at list.
• Mixed set of 6 images, 3 videos, 3 audios: 12 files, at the combined limit.

Why the free count matters

Reference images are how you hold a character steady across shots. A set of five costs nothing extra at list, and most single characters need three or four: a front view, a side view, the outfit and one expression. Use the free slots for the character, and spend extra slots only on a second person or a prop that must match.

For a cast of three, 9 images allow three per person. A cast of four does not fit without trimming. Vidu Q4 would allow 15, but Sume does not list Vidu, so there is no Sume route to those 15; the comparison is only a reference point for the limit.

The clip itself

A 5-second H3 clip at 768p is 38 cents. The arithmetic: list $0.06 per second x 5 = $0.30, times the 1.25 margin = $0.375, rounded up to 38 cents. H3 Max at 768p is $0.08 x 5 = $0.40, x 1.25 = $0.50, so 50 cents. Both are billable prices from the Sume catalog formula.

If your character set has more than 9 faces, split it. Pick the faces that appear in a given shot and send only those. A tighter set also helps the model, since each reference competes for attention. For the full shot-splitting method with Wan 3.0, see the related post.