Symphony Agent takes prompts, images or example videos: Sume's inputs

Each of the three inputs reported for Symphony Agent has a matching field on Sume's video endpoint, with caps. A search-result snippet from Stackmatix (read 2026-10-02; snippet only) says Symphony Agent analyses trends and produces promotions from prompts, images or example videos. Sume's Video route takes a prompt, an image_url, and reference images and videos. What Sume does not do is analyse trends inside the same call.

Which Sume field matches which input?

The field names come from the Video 1.0 page, which is a compatibility alias for Video Router Auto. New integrations use POST /v1/videos with model: "sume/auto", which the same page points to.

What happens to a reference video?

It guides generation; it is not copied. The docs describe reference_video_urls as reference-guided input, and URLs must be public HTTPS. Use footage you have the right to use as a reference. If you want to read a reference clip before planning a remix, video inspect returns probe facts and stills at no charge for the probe.

Where does trend analysis happen on Sume?

Not inside the video call. Sume has a separate trending-videos search for TikTok research, covered in the stored posts below, and you decide which trend to follow. A bundled agent that picks the trend for you removes that step; with the API it is your step.

Pick a trend, write the prompt, pass a reference if you have the rights, and set aspect_ratio to 9:16 for a vertical clip. Then label the result if it shows a synthetic face or a photorealistic scene.

What does a minimal request look like?

A prompt alone is a valid request. Add image_url when you want a first frame, end_image_url for an end frame, and reference arrays when you want the look of existing assets. Choose duration from the integers the catalog lists for the model, since supported lengths differ by model and GET /v1/videos style catalog reads show them.

Run one test clip per input type before a batch. A prompt-only clip, an image-led clip and a reference-led clip give you three samples of how each input behaves, and you can then pick the one that suits your product.