Vidu Q3 drama character consistency vs Sume reference images

QwenCloud lists vidu/viduq3-drama_reference2video for drama and AI comic series, citing strong character consistency. Vidu is not in Sume's public video ids. On Sume, reference images guide a clip rather than fix exact frames, and if you also send frame_images, those take precedence and the request becomes image-to-video.

For a series workflow see microdrama episodes with consistent characters.

What does the vendor claim?

The changelog says the drama model features strong character consistency, refined motion effects, and authentic emotional expressions, ideal for story-driven content creation. This post does not test or restate those claims for Sume.

How do references work on Sume?

input_references provides style or content reference images for reference-to-video. The model uses these as visual guidance rather than exact frames. frame_images specify first or last frame images for image-to-video.

Which models accept audio and video references too?

Audio and video references are honored by the Seedance 2.x models, Wan 3.0, MiniMax H3, and MiniMax H3 Max, per the docs. Check supported_input_references for the model you choose.

How do I keep a character steady across shots?

Reuse the same reference images in each request, and send them as input_references without frame_images unless you want a specific first or last frame. Review each clip, since references guide rather than lock the result.