A 30-second ad from three 10 s MiniMax H3 clips: cost and joins

MiniMax H3 and H3 Max cap a clip at 15 seconds on Sume, so a 30-second ad is two or three clips joined. Three 10-second clips cost $2.25 at 768p on minimax-h3, $3.00 at 768p on minimax-h3-max, and $6.00 at 1080p on Max.

The harder part is continuity: each clip is a separate generation with its own native stereo audio, so plan the cuts and the sound before you submit.

The cost

The Video Router docs give 5 to 15 seconds and the fal list per second. With Sume's 1.25 multiple, 10 seconds at 768p is $0.75 on H3 and $1.00 on H3 Max, and 10 seconds at 1080p on Max is $2.00. Two 15-second clips are cheaper to coordinate than three 10-second ones and cost the same, since the price is per second.

Keeping shots linked

Both rows support text-to-video, first/last-frame image-to-video and reference-to-video. The first/last-frame mode is the useful one for chains: end clip one on a frame, export that frame as a still, and use it as the first frame of clip two. The docs list start/end-frame image-to-video as supported on both ids, so each shot can be pinned at both ends.

For a product ad, a simpler method is to reuse the same reference images in each clip, so the product looks the same across the cuts. On minimax-h3, remember the extra $0.10 per reference after the fifth image on every clip, and that three clips triple the surcharge.

Sound across three clips

The MiniMax announcement (read 2026-10-05) says H3 generates native stereo sound, and the Sume docs say there is no toggle, so each clip carries its own sound. Dialogue and music are generated per clip. If you want one music bed across the ad, the Sume docs describe a Timeline API with an audio spine for assembling clips with a single track, which is where to join and replace the sound.

Do not expect the ambience of clip two to start where clip one ended. Prompt each clip's sound explicitly and keep the style words the same.

• Prompt the same style, lighting and sound words in each clip.
• Use the last frame of the previous clip as the next first frame.
• Join on a timeline and replace the music bed if the joins are audible.

When to use one longer model instead

If you need 30 continuous seconds with no join, a different model may fit better: the Video Router lists seedance-2.5 at 4 to 30 seconds and wan-3.0 at 2 to 30 seconds. They are not MiniMax rows and their prices differ, so compare before you switch.

A job plan

A simple plan has four steps. First, write three shot prompts with the same subject, lighting and style words. Second, render clip one at 480p on the row you will finish on, at $0.63, and check it. Third, render the three clips at 768p. Fourth, join them on a timeline, add a single music bed if needed, and deliver.

At 768p on H3 the three clips are $2.25, and the 480p check adds $0.63. Replace one weak clip and you pay $0.75 again. A 30-second ad with one re-render still costs under $4.00 on H3, and under $5.00 on H3 Max.