Use the Wan 2.7 AI Video Generator to prepare a text-to-video or one-image-to-video task with current ZMS controls.
Loading generator…
Model overview
The Wan 2.7 AI Video Generator in ZMS AI prepares one short task from text or one start image. It keeps prompt, resolution, duration, and text-mode framing visible so the same brief can be reviewed against the exact settings used for a test.
The current driver exposes 720p or 1080p and 5, 10, or 15 seconds; text mode also sends 16:9, 9:16, 4:3, or 3:4. Image mode does not currently send ratio. Compare versions on the Wan AI model family, and verify one signed-in task before treating the registered adapter or credit estimate as delivered production behavior.
Current control boundary
Read the active controls as a request definition, not a guarantee of a completed generation. The four boundaries below describe the local ZMS implementation rather than every capability documented for the upstream Wan model.
Start from a written brief when the composition is open. Describe one subject, one primary action, the setting, camera behavior, and a clear ending state. Text mode sends the selected ratio, resolution, and duration with the prompt.
Upload one start image when identity, product shape, layout, or art direction must be anchored. The prompt should explain how that still evolves over time. There is no current end frame or multi-image reference control on this ZMS page.
Choose 720p or 1080p and design the action for 5, 10, or 15 seconds. A longer setting gives a camera move more time to resolve, but it also creates more frames in which identity, geometry, or background continuity can drift.
The current Wan 2.7 interface does not expose an end frame, reference video, audio switch, or visible negative-prompt field. Upstream documentation may describe broader modes; those features are not promises about this independent ZMS integration.
Prompt settings and review
The Wan family page retains the four generated videos and their full evidence cards. This version page reuses those exact prompts without copying the gallery, pairing each one with the 720p, 5-second, 16:9 run settings and a shot-specific acceptance check.
Generated-pair prompts from the Wan family ledger
4 starter recipesProtect small-object geometry while hands, liquid, steam, and a restrained camera remain readable in one product-scale action.
Close shot of hands pouring tea from a clay pot into a small cup on a bamboo terrace, steam curling in the light. The camera holds steady at table height as the cup fills. Dappled afternoon sun through leaves, warm earth tones, shallow depth of field, no camera shake.
Inspect pot, cup, fingers, liquid path, and fill level across the whole clip; reject fused handles, changing vessel shapes, or a late camera wobble.
Use one receding vehicle and a low tracking move to test spatial continuity, horizon stability, and the behavior of dust over time.
A dust-covered pickup truck drives away down an empty desert highway as heat shimmer distorts the horizon. The camera tracks slowly from behind at low height, sand drifting across the asphalt. Late afternoon sun, ochre and pale blue palette, long shadows, steady motion.
Check vehicle proportions, lane position, wheel contact, horizon level, and dust direction; the truck should recede continuously without duplicating or jumping in scale.
Separate the worker's deliberate action from fast sparks while preserving protective equipment, steel geometry, and one motivated push-in.
A welder in a leather apron lowers a helmet visor and strikes an arc against steel tubing in a dim workshop, sparks scattering across the concrete floor. The camera pushes in slowly from a three-quarter angle. High contrast between the arc light and deep shadow, no camera shake.
Review visor, hands, torch contact, tubing lines, and spark origin frame by frame; reject bent steel, exposed facial drift, or sparks that detach from the arc.
Hold a locked composition while independent foreground droplets, distant traffic, reflections, and a glass remain spatially coherent.
Rain runs down a cafe window as blurred traffic lights pass on the street beyond, a half-full glass of water on the sill catching the light. The camera holds a locked medium shot, focus on the droplets. Cool evening blues with warm interior spill, quiet atmosphere, no camera shake.
Confirm the camera stays locked, droplets move on the glass plane, traffic remains behind it, and the water glass keeps one stable rim and fill level.
Practical workflow
Build one reviewable shot before planning a sequence. Holding the mode and output settings steady makes it easier to tell whether a prompt revision actually improved motion, continuity, or composition.
State where the clip will be used, what the viewer must notice, and which single event should happen. Separate unrelated locations, time jumps, or actions into different tasks instead of forcing a full story into one generation.
Use text when the model can interpret the initial composition. Use one start image when a visual fact must remain authoritative. Upload only assets you are authorized to use, and avoid implying that a still can control details never stated in the prompt.
In text mode, select landscape, portrait, 4:3, or 3:4 before submitting. Choose 720p for a lighter test or 1080p when the review needs more output detail. Pick 5, 10, or 15 seconds based on the action’s real timing.
After a signed-in submission is accepted, record task status, final media, duration, resolution, framing, and account charge. Until that full path is tested, describe the workspace as configured and avoid making claims about speed, reliability, or output quality.
Archived motion evidence
Watch the whole shot before revising the prompt. Check whether the subject remains recognizable, physical forms stay stable, movement follows a readable path, the camera has a clear purpose, and the final beat resolves instead of stopping abruptly.
Prior internal archived Wan 2.7 example, not a current ZMS AI live-generation test. The source project does not include a separate license or rights-clearance ledger.Production review
Keep upstream documentation, the current ZMS request surface, and the two published evidence lanes separate. The family gallery proves four prompt-to-video pairs; this page's archived clip supports motion review but does not prove current service delivery.
| Decision | Current fact | Review action |
|---|---|---|
| Set the request | ZMS exposes text or one start image, 720p or 1080p, and 5, 10, or 15 seconds. Only text mode currently submits the selected aspect ratio. | Confirm the active mode and visible controls, keep the format stable, and use the Wan 2.7 workflow guide when the shot needs a fuller preparation sequence. |
| Read the evidence | The family page holds four current Wan 2.7 prompt-to-video pairs. This page separately retains an eight-second archived example with incomplete original task and rights records. | Use current pairs to compare prompt, settings, and result. Use the archived clip only to inspect motion continuity; do not infer current latency, reliability, or commercial clearance. |
| Approve a batch | The client estimates 20 credits per second at 720p and 30 at 1080p, while signed-in completion and final account deduction remain unverified. | Run one account task, retain the request record, inspect the full clip, check ZMS AI pricing, and move to the AI video generator if another input path fits better. |
FAQ
Clear answers about inputs, output controls, image-mode framing, credits, live verification, and the archived example shown on this page.
The Wan 2.7 AI Video Generator is ZMS AI’s focused workspace for preparing a Wan 2.7 task from text or one starting image. The local driver is registered, but a signed-in live task has not yet been verified.
The current ZMS interface exposes text-to-video and image-to-video with one start image. It does not expose an end frame, reference video, audio control, or visible negative-prompt field for Wan 2.7.
Text mode lists 16:9, 9:16, 4:3, and 3:4. The image-to-video request currently does not submit an aspect-ratio value, even though the shared interface can display ratio controls, so ZMS does not claim reliable image-mode ratio control.
The Wan 2.7 workspace lists 720p and 1080p plus 5, 10, and 15 second durations. Those options are confirmed in the local request configuration; successful delivery still depends on a signed-in live task being accepted and completed.
The client estimates 20 credits per second at 720p and 30 credits per second at 1080p. The resulting account charge has not been verified through a signed-in live task, so the service remains authoritative for the final debit.
No. It is a prior internal archived Wan 2.7 example, not evidence of current ZMS quality, latency, reliability, or task completion. The source project has no separate license or rights-clearance ledger, so reuse still needs project-level review.