Wan 3.0 references or start/end frames, not both: two Sume calls

Krea's Wan 3.0 entry says a single job accepts reference media or start and end frames, not both. Sume's video docs also list those as separate input modes, so plan two requests: one guided by a start frame, one guided by references. On POST /v1/videos the docs say that if both frame_images and input_references are sent, frame_images takes precedence and the request is treated as image-to-video, so the references are not used; the video input-modes page does not describe a mix, so do not rely on one.

Krea facts are from its changelog (August 28, 2026); Sume facts are from the video input modes and video generation docs, read 2026-10-01.

What is the vendor rule?

Krea says Wan 3.0 can generate from a text prompt, animate a start image, or be guided by reference images, videos, and audio, at 480p, 720p, or 1080p. For a job, you attach reference media or you set start and end frames.

How does Sume list the modes?

Each mode is its own row in the docs, with its own fields.

What can I not send?

end_image_url requires image_url, so an end frame alone is invalid. reference_audio_urls requires at least one reference image or video. Wan 3.0 is among the models the docs list as honoring audio and video references; its duration range is 2 to 30 seconds.

How do I split one idea into two requests?

Pick the control that matters most for each clip. Use start and end frames when you need a specific opening and closing image. Use references when you need a character or style held across a shot. For a clip that needs both, make the reference-guided clip first and then use a frame from it as image_url on the second request; see frame images and input references together for what the docs say about combining them.