If you are deciding whether Wan 2.7 belongs in a repeatable video pipeline, make the first decision about the CometAPI workflow—not about a feature checklist. The supplied CometAPI guide documents model ID wan2.7 behind an asynchronous POST /v1/videos route. A usable asset requires the rest of the path too: preserve the returned task ID, poll GET /v1/videos/{video_id}, handle a terminal failure, and request the MP4 from GET /v1/videos/{video_id}/content only after success.
Start with the guide’s small validation case: 5 seconds at 1280x720. Treat task creation as a checkpoint, not a completed deliverable. Your acceptance test should record the task ID, use a bounded polling window, retain the complete failure payload when the task fails, and verify that the downloaded response is a non-empty video rather than an error document carrying an .mp4 filename. The guide’s example is operational documentation, not evidence that your account, prompts, or production assets will succeed.
This sequence also gives the budget a concrete starting point. The guide lists $0.08 per generated second for 720p and $0.12 for 1080p, so one 5-second attempt is an estimated $0.40 or $0.60 respectively. Those are source-dated listed generation rates, not a promise of current pricing or a complete production budget. Track retries and divide total generation spend by accepted clips; a cheap failed or rejected clip is not a production saving.
Only after the 720p path is observable end to end should you test 1080p, image references, or other controls. The guide explicitly warns that native Wan-family features—such as first/last-frame control, continuation, audio guidance, and editing—may not map one-to-one onto CometAPI’s unified schema. Confirm the route’s current fields and media limits individually instead of inferring support from the model family. Also confirm which duration values the specific route accepts before treating native 2–15-second guidance as a contract.
The practical rollout decision is therefore staged: keep Wan 2.7 at a pilot gate until CometAPI task persistence, bounded polling, terminal-state handling, and validated MP4 retrieval work for your workload; then compare resolution and advanced inputs using cost per accepted clip and your own creative acceptance criteria. This guide supports that test plan, but it does not supply latency, success-rate, quality, or production-readiness measurements.