Using Wan AI is a seven-step workflow: choose the input that gives the right control, select a model by task and cost, describe motion, submit with visible credits, wait for the provider, diagnose the result, and download or revise.
Use text-to-video when the model can invent the composition. Use image-to-video when a product, character, artwork, or first frame must remain recognizable. Use video-to-video when an existing short clip should be edited, restyled, or continued.
Wan 2.2 is the low-cost visual draft. Wan 2.5 adds native audio in the hosted workflow. Wan 2.7 adds newer text/image controls, longer provider durations, continuation, and edit-video. Starting expensive does not improve an unclear instruction.
Name one subject, an observable action, one camera move, setting, light, sound if needed, and explicit constraints. For image input, do not repeat what the model can already see.
The selector shows the server catalog's charge. Resolution and duration are tied to that choice and locked again by the server. If your balance is short, the provider is not called.
Video inference is asynchronous. The task may be queued, processing, successful, or failed. The page polls the server, while the server—not the browser—owns timeout settlement and refunds.
Keep the prompt, model, input mode, resolution, and result together. A reusable recipe is more valuable than a lucky output that cannot be reproduced.
Open the Wan AI video generator
Last checked: August 12, 2026.
No. This hosted workflow runs in a browser. ComfyUI is an alternative for compatible open-weight versions when you own suitable hardware and want local control.
An account provides a credit balance, task history, refund destination, and an abuse-control identity. Draft prompt text remains in the browser when the sign-in dialog opens.
Use the Wan AI prompts page for general formulas and use-case recipes. Version pages explain limits and link back to the correct preselected generator model.