Text-to-video scene composer
Describe a scene as a sequence of visible events, then choose camera language, pacing, duration and delivery format. The generator submits a real Wan 2.7 video task and returns a playable result when the asynchronous render completes.
Compose a video
The luminous moth storyboard is an editorial concept image, not a frame extracted from a generated video. Actual results vary and should be reviewed for continuity, rights and safety before publishing.
The workspace focuses on the variables that determine whether a short clip feels intentional: what changes, how the camera witnesses it, how quickly the action develops and where the final beat lands.
Scene composer
From 80 credits · 720p · the 8-second option costs 120 · failed provider tasks are refunded
Write what changes across time. The completed video will play here without sending you to a separate demo page.
A video prompt needs verbs and sequence. State the opening condition, the physical change and the final state. Describing only a beautiful subject leaves the model to invent motion, which often produces drifting cameras, unstable geometry or an action that never resolves.
Camera instructions should support the action instead of competing with it. A locked frame reveals transformation; a slow push increases attention; a lateral track follows travel. Stacking orbit, zoom, crane and handheld language in one short clip usually reduces continuity.
Five seconds suits one reveal, gesture or product moment. Eight seconds gives a two-stage transformation room to breathe. Longer is not automatically better: every additional beat increases cost and creates more opportunities for identity drift or unwanted objects.
Watch once for the story, then scrub through the clip for shape changes, duplicate objects, flicker, unsafe details and abrupt camera jumps. Generated video is a production asset only after the subject, claims, likeness rights, audio and final destination have been checked.
It is a real text-to-video workflow. The form submits an authenticated Wan 2.7 task through the server, polls the provider queue and displays the resulting video when complete.
The page offers 16:9, 9:16 and 1:1 formats with controlled 5 or 8 second durations at 720p. These choices cover common landscape, vertical and square placements while keeping costs predictable.
The model must create many temporally consistent frames rather than one image. Queue load and duration affect render time, so the task remains asynchronous and can take several minutes.
A failed task is marked as failed and its consumed credits are restored by the existing transactional task system. Temporary status errors remain retryable before the final timeout.
Only use people, characters, music and source material you have the right to use. Do not create deceptive impersonation, non-consensual intimate material or misleading evidence, and review provider and destination-platform policies.
