Vidu Q3 ad reference-to-video vs Sume reference images

QwenCloud lists vidu/viduq3-ad_reference2video, a Vidu reference-to-video model for advertising where you upload product images. Vidu is not among Sume's public video ids. If you want product images as references on Sume, use a model whose catalog entry lists the input reference type, and send input_references on POST /v1/videos.

For another reference-driven workflow see Wan 3 reference limits.

What does the Vidu Q3 ad model do?

QwenCloud describes it as a Vidu reference-to-video model designed for advertising, with marketing-grade smart scene cuts, camera movements, and direct audio output. You upload product images to generate ad videos. Sume has no matching id, so none of those behaviors are claimed here.

Which Sume models take reference images?

The legacy Video 1.0 shape takes 1 to 9 reference_image_urls. On POST /v1/videos, input_references is the reference-to-video field. Only models whose supported_input_references lists a type accept that type.

How do product references behave?

The docs say the model uses references as visual guidance rather than exact frames. If both frame_images and input_references are sent, frame_images takes precedence and the request is treated as image-to-video.

What should I do for an ad clip?

Pick a model that lists reference support in GET /v1/video-router/models, send your product images as references, and review the result for product accuracy. Do not assume ad-specific scene cuts exist on Sume.