Lip sync on a 2.76:1 widescreen still: crop to 2.5:1 first

A 2.76:1 widescreen still is outside the aspect-ratio range Sume lists for MiniMax H3 Max Lip Sync, which is 0.4 to 2.5 (width divided by height) in the API reference read on 2026-10-09. A frame of 2760 x 1000 pixels has a ratio of 2.76. Crop it to 2500 x 1000 (ratio 2.5) and it fits, at the cost of 260 pixels of width.

Checking the ratio before you submit

Divide width by height. The route accepts values from 0.4 to 2.5 inclusive as written in the schema. Common sizes fall in or out as the table shows.

Where to crop

Keep the face centered and leave some margin above the head and below the chin rather than cropping tight. A widescreen still usually has the face in one third of the frame, so crop around the face and drop the dead space on the other side. You are not losing content that the model would use.

Send the cropped image at a public HTTPS URL as image_url, or use a ready avatar by handle. Never send both. The audio must be Sume-hosted, 5-14.8 seconds and under 10 MB.

What the clip costs

The cost is independent of the still. It depends on the audio length and resolution: $0.0625 per second at 480p, $0.10 at 768p, $0.20 at 1080p, billed on ceil of the audio second. A 12-second line at 768p is 12 x 0.10 = $1.20. Re-cropping a still changes nothing in the price.

If you need the widescreen framing in the final video, lip sync the cropped face and place it in a wider scene afterwards with the timeline compose route, which Sume lists at $0.02 per job for still-plus-video composition.

• Range: 0.4 to 2.5.
• Audio: 5-14.8 s, Sume-hosted, under 10 MB.
• Exactly one visual source.

A worked example

Say your source is a 2760 x 1000 frame with the speaker's face around x = 1500. Crop a 2500-pixel-wide window from x = 250 to x = 2750 so the face sits near the center, giving 2500 x 1000, ratio exactly 2.5. If you want a margin for rounding, go to 2480 x 1000 (2.48). Then upload the crop, put its public HTTPS URL in image_url, and send a 9-second line at 768p: 9 x 0.10 = $0.90.

The same crop logic works for tall stills. A 1000 x 3000 frame is 0.33, below the 0.4 floor, so crop it to 1000 x 2500 (0.4) or 1000 x 2000 (0.5). Start from the face and work outward.