Workflow decision guide
Choose the Right AI Video Workflow
Start from the material you already trust. A prompt, a frame, a reference set and an existing clip solve different production problems, so the best workflow is the one that gives the model the clearest evidence without unnecessary inputs.
Choose by the source you already have
Use text-to-video when the idea exists only as a shot brief. Use image-to-video when one composition or a first-and-last-frame pair must anchor the result. Use reference-to-video when identity, objects, environments, motion or audio need separate evidence. Use video edit when an existing clip must remain recognizable.
- Text only: begin with text-to-video.
- One composition or two endpoint frames: begin with image-to-video.
- Several identity, style, motion or audio sources: begin with reference-to-video.
- A clip whose timing or composition should survive: begin with video edit.
Add control only when it removes ambiguity
More files do not automatically produce a better result. Every source should answer a specific question the prompt cannot answer clearly. Conflicting references increase ambiguity, while a smaller role-based set makes it easier to diagnose what should change in the next attempt.
Match the workflow to an available model
WAN3 supports text, frame and mixed-reference creation. Wan 2.5 and 2.6 cover simpler text, image and source-video paths with fixed choices. Wan 2.7 adds flexible frame, reference and directed-edit workflows. Each detailed guide below states the exact input and output boundary rather than treating every model as interchangeable.
Plan one diagnostic first generation
Choose the shortest duration and resolution that can answer your creative question, review the exact credit quote in Workspace, and change one variable at a time. Once the composition, motion and continuity work, move to the final duration or 1080p output.