How Do I Use Image-to-Video API with a Reference Image?

Overview

Use an image-to-video API in four steps: prepare an authorized still, upload it if local, place its returned URL in the selected route’s image input, and POST a short motion brief. Save the prediction ID, then GET the result. The reference starts the shot but does not guarantee unchanged product geometry in every frame.

Crop the still to the target composition and remove distracting elements. In WaveSpeedAI’s current upload flow, POST filename and byte size for a ticket, PUT the file bytes to its signed URL, then use download_url as the model’s image input. The selected image-to-video page owns the exact field name; an example MiniMax H3 route uses image, prompt, duration and resolution. Describe subject action and camera movement rather than repeating what the still already shows.

Review the clip against the reference: does the product label stay readable, do faces drift, and does the shot work at the intended crop? For catalog videos, a simple camera move can be more useful than ambitious scene changes. If exact identity is mandatory, compare another route or use conventional animation for the sensitive element.

Treat the uploaded still and generated clip as separate assets with their own permissions and storage records. The API handles generation; your workflow still owns consent, approval and long-term delivery.

Retain the original crop and request settings beside the job record so a reviewer can reproduce the intended shot. If the motion brief changes, create a new tracked attempt rather than overwriting the history of the approved source.

Source: WaveSpeedAI image-to-video guide; WaveSpeedAI upload API; WaveSpeedAI result API. Source check: 2026-09-29.