Wan 3.0 document to video: what the model reads and what Sume passes

Wan 3.0 can take a document as input on the vendor's own API, but Sume's wan-3.0 does not expose that field today. Alibaba Cloud describes a document-to-video feature that parses a file and generates a video from it; Sume's catalog lists wan-3.0 for text, first and last frames, and image, video and audio references, and notes that file_url, web_url and enable_thinking are not exposed in v1.

The Alibaba details are from its Wan3.0 guide and API reference; fal's Wan 3 page says documents need thinking mode. The Sume side is from Video generation and the current video catalog. All read 2026-09-29.

What can Wan 3.0 do with a document?

Per Alibaba, you provide one document or one public link and the model parses the content to build the video. The limits it publishes:

Can I send a document to wan-3.0 on Sume?

No. The /v1/videos request has prompt, frame_images and input_references; a reference is an image_url, video_url or audio_url. There is no field for a document, so a PDF or a slide deck has nowhere to go on this route.

How do I get a video from a document on Sume, then?

Read the document yourself and turn it into a prompt. Extract the text, pick the three or four scenes worth showing, and write one prompt per scene; export the charts or pages you want on screen as images and pass them as input_references or a first_frame. PDF to video AI shows the agent route, where the extracted text goes in input.

Which route when?

• You want a clip per scene, with control of each prompt: wan-3.0 on POST /v1/videos, 2 to 30 seconds each.
• You want one call from a brief and material to a finished video: a Format or an Agent Completion, with the text in input.
• You need the vendor's own document parsing: that is Alibaba's API, which Sume does not front today.