Image-to-Video From a Product Still: End Frames and Grok Limits

A product photo is the safest first frame you have: the label, the color and the proportions are already right. What differs is what each model lets you do next. Some take only the first frame; some take a last frame so the clip can return to its start.

What each model takes

On Sume, kling-3, wan-3.0, minimax-h3, minimax-h3-max and gemini-omni-flash-1.1 accept a first frame plus an end frame. The Seedance family accepts first and last frames plus references. grok-imagine-video-1.5 is image-to-video only, requires image_url or first_frame_url, and has no end frame and no audio toggle; it runs 4 to 15 seconds at 480p or 720p for $0.0125 per second. xAI's own guide describes text, image and video-edit input, so Sume's row is narrower than the vendor's feature set.

A request that uses both frames

On POST /v1/videos, each entry in frame_images carries a frame_type of first_frame or last_frame. If you also send input_references, the frame images win and the request is treated as image-to-video. Google's Veo page says first and last frame are supported for Veo too, with the 8-second rule applying to 1080p and 4K.

Use the end frame for loops and for reveals where the final pose is the packshot. Skip it for open-ended motion, where forcing a destination often makes the middle of the clip stiff.

Budget approach

Draft on Grok at $0.0125 per second to check the framing and the motion direction: a 5-second test costs about 6 cents. Move to a model with an end frame only when the shot needs one. Remember that Sume exposes no seed on any video model, so a draft cannot be reproduced exactly at higher quality; the final is a new generation.