MiniMax H3 reference images on Sume: first 5 free, then $0.10 each

On Sume, minimax-h3 accepts reference images, videos and audio in one request, and the first five reference images are free. Each image after the fifth bills $0.10, which is the $0.08 list price times Sume's 1.25 multiplier. The request limits are 9 images, 3 videos and 3 audios, with at most 12 references in total.

What reference-to-video is for

Sora-style reference work usually means a character or product held constant across clips. The H3 pair is the Sume family built for that, with native stereo audio and 5 to 15 second clips.

Choose references with care since they are the lever you control. Five clean images of the same subject from different angles usually do more than nine near-duplicates, and staying at five keeps the extra fee at zero.

Limits and fees

The image fee in practice

The extra fee applies per image beyond the fifth, so 5 images add nothing, 7 images add $0.20, and the maximum 9 add $0.40. These fees come on top of the per-second generation price, which depends on resolution; read it from the pricing table in the video docs.

If you need more than five, count the fee into your quote: the figure is per request, so a batch of twenty clips that each send eight images carries 20 x 3 x $0.10 = $6.00 in image fees on top of generation.

Combining the limits

The cap of 12 is shared. Nine images leave room for three more items across video and audio, so a 9 image, 3 video, 3 audio request is rejected for being 15 in total.

• Five or fewer images: no reference fee.
• Six images: one $0.10 charge.
• Nine images, two videos, one audio: twelve in total, $0.40 in image fees.

Audio and alternatives

Audio is always on for the H3 pair, so there is no generate_audio switch to forget. If you need a silent clip, kling-3 with audio off is the id with that control. For a larger model with a 1080p refinement, read the H3 versus H3 Max post.