MiniMax H3 references on Sume: 9 images, 3 videos, 3 audio, 12 total

On Sume, minimax-h3 and minimax-h3-max accept at most 9 reference images, 3 reference videos and 3 reference audio files, with 12 files in total, and audio cannot be the only reference. Each reference video or audio clip must be 2 to 15 seconds, and the clips of one kind may add up to 15 seconds. Break one of these and the request returns a 400 naming the field, before any job is created.

MiniMax's announcement says H3 takes images, video and audio as context together; the count limits above are the Sume catalog's.

The limits at a glance

Counts are enforced by the validator; the duration and audio-only rules are the row's documented constraints.

Typical errors

The validator reports the first field that breaks a limit. Fix the field it names and resubmit with a new Idempotency-Key.

• reference_image_urls must not exceed 9 files.: send fewer images.
• reference_video_urls must not exceed 3 files.: keep the best three clips.
• Reference inputs must not exceed 12 total files.: you sent, say, 9 images, 3 videos and 1 audio file.
• Audio only: add at least one image or video.

The extra-image price on H3

On minimax-h3, Sume's catalog notes that the first five reference images are free and each additional image carries a $0.08 list charge, which Sume bills times 1.25. Nine images therefore add four paid images, about $0.40 at billed rate. minimax-h3-max bills output seconds only, and the catalog says the provider's reference-token overage is not reserved.

ok = {"images": 9, "videos": 3, "audio": 0}
bad = {"images": 9, "videos": 3, "audio": 1}

def within_limit(counts: dict) -> bool:
    return counts["images"] <= 9 and counts["videos"] <= 3 \
        and counts["audio"] <= 3 and sum(counts.values()) <= 12

print(within_limit(ok), within_limit(bad))

Where to read more

The row constraints are on the Video Router page, the request fields are in Video generation, and the vendor text is MiniMax's H3 announcement. The full error schema is in the API reference.