Text-to-video
Use text for open composition. This path sends prompt, resolution, duration, and 16:9 or 9:16 framing.
Prepare a Wan 2.6 video task from text or one start image with mode-aware duration and framing controls.
Loading generator…
Model overview
The Wan 2.6 AI Video Generator on ZMS AI prepares one text or start-image task with duration, resolution, framing, and advanced controls.

Reference-video mode is hidden until its service path and a signed-in cycle pass verification.
Use the Wan model family to compare 2.6 and 2.7; visible ZMS controls define this route.
Input modes
Choose the input carrying the strongest visual constraint, then prompt only needed motion.
Use text for open composition. This path sends prompt, resolution, duration, and 16:9 or 9:16 framing.
Use one start image to anchor appearance and composition. Image mode sends prompt, resolution, and duration, not ratio.
This entry returns only after signed-in creation, completion, media, and charging pass together.
Input-path recipes
These original suggested recipes cover two text tasks, one start image, and one unavailable reference-video plan. Try preserves mode.
Current text and image paths plus one disclosed reference plan
4 starter recipesReview hands, geometry, eyelines, and one side track.
Medium-wide view inside a daylight bicycle workshop. One mechanic passes a polished steel wrench across a wooden bench to a second mechanic, who takes it with the right hand and sets it beside one blue toolbox. The camera makes one slow side track at bench height. Preserve both adult identities, four visible hands, wrench shape, toolbox count, bench edge, hanging wheels, and cool window-light direction. No speech, cut, zoom, extra person, duplicate tool, fused fingers, changing clothing, camera shake, or background redesign. End after the wrench rests beside the toolbox and both hands separate.
Inspect hands, wrench, contact, toolbox, eyelines, bench, track, and release.
Keep subject, place, and axis continuous across two beats.
Create one continuous ten-second sequence in the same sunlit pottery studio. Beat one: a ceramic artist carries one unfired vase from the worktable to a shelf in a steady medium side view. Beat two: without changing identity, vase, shelf, wardrobe, room layout, light, or camera axis, move to a closer view as the artist aligns the vase beside two existing bowls and removes both hands. Keep the progression physically continuous and the vase upright. No new person, duplicate vase, cut to another room, reversed travel direction, warped fingers, changing shelf count, text, or camera roll. End on the stable three-object arrangement.
Check beat continuity, identity, vase, shelf, direction, axis, and invented scenes.
Protect silhouette, controls, material, composition, and light from one image.
Use the uploaded first frame as the exact opening composition. Preserve the compact tabletop fan's cream metal cage, five blue blades, black speed knob, power cable, base proportions, desk edge, wall color, and daylight direction. Over six seconds, the blades begin turning slowly while the camera performs one restrained push-in at hub height. Keep the fan body completely still and fully inside frame. No hand, duplicate fan, changing blade count, bent cage, new label, cable movement, orbit, cut, focus jump, or background object. End with a steady slow rotation and the same framing relationship.
Compare cage, blades, knob, cable, base, crop, and light across frames.
Name borrowed motion and excluded identity, setting, and geometry.
Use the authorized reference video only for the timing of one spinning top: a two-second hand release, four seconds of stable rotation, a gentle wobble, then a stop. Create a different scene with one unbranded wooden top on a pale concrete table, side daylight, and a locked close camera at table height. Do not copy the reference person's identity, hand appearance, room, colors, objects, audio, logo, or camera angle. Preserve one top, its red painted ring, table edge, shadow direction, and rotation axis. No duplicate top, morph, jump cut, moving camera, extra hand after release, invented text, or background change.
Do not submit while hidden; later verify timing alone transfers.
Control boundaries
Check active inputs; settings do not transfer silently between modes.
| Control | Current ZMS request | Review note |
|---|---|---|
| 720p or 1080p | Use 720p for tests and 1080p after action and framing pass. | Resolution is a request, not delivery proof. |
| 2–15 seconds for text or image | Text and image tasks accept 2–15 seconds. | Give each action enough time to resolve. |
| Reference video is hidden | No reference-video duration or upload control is currently exposed. | Use text or one start image meanwhile. |
| No end frame or audio control | No final-frame upload or audio setting | Do not plan around either control. |
Archived Wan 2.6 example
Review movement, collision, framing, and continuity across this archived clip.
Prior internal archived Wan 2.6 example, not a current ZMS AI live-generation result or evidence of service quality. The source project does not include a separate rights-clearance ledger.Practical workflow
Keep one stable test and change one variable per revision.
Define subject, environment, action, camera, light, final beat, and protected details.
Use text for open composition or one image for a fixed opening.
Set resolution, duration, text-mode ratio, expansion, and multi-shot deliberately.
Record task, status, media, debit, motion, identity, geometry, continuity, and artifacts.
Advanced controls
Prompt expansion and multi-shot are linked request settings, not outcome guarantees.
Put subject, setting, action, camera, and final state in the base prompt.
Expansion is on by default. Disable it for exact wording and keep the setting stable during comparison.
Multi-shot depends on expansion. Outline each beat and keep the sequence within duration.
After return, verify coherent beats and stable identity; the control itself proves nothing.
Reference-video planning
The reference-video path is currently hidden, but teams can still prepare an honest brief for later verification. Use the board to separate the source motion worth studying from protected rights, the new scene, submitted controls, and the output checks that must pass together in review before any transfer is approved.

Describe only the timing, path, rhythm, or camera behavior that the authorized source should contribute to the new shot.
List people, products, rooms, logos, colors, audio, and distinctive details that must not transfer from the reference video.
Specify the new subject, setting, framing, duration, camera, final state, and protected geometry independently from the source clip.
Wait for reference mode to return, then verify signed-in task creation, returned media, source separation, motion transfer, rights, and debit together.
Product and source boundary
Upstream Wan documentation, the current ZMS request surface, the archived rescue clip, and the disabled reference path support different claims. This ledger preserves that separation and identifies the control, evidence, rights, account, and debit checks required before a production task.
| Decision | Current fact | Review action |
|---|---|---|
| Choose an input path | Text and one-start-image paths are visible. Reference video remains hidden after failed signed-in creation checks; image mode relies on the source composition rather than a submitted ratio. | Use the smallest available input, confirm mode-specific controls, and return to the Wan family or AI video generator if another route fits the brief. |
| Read the evidence | The rescue clip is prior internal archived Wan 2.6 evidence with duration and date. It is not a current live task and lacks a separate rights-clearance ledger. | Use it only for action and continuity review. Do not infer latency, reliability, prompt adherence, present service quality, or commercial clearance from the clip. |
| Approve production use | Client estimates list 20 credits per second at 720p and 30 at 1080p. Signed-in completion, returned media, and final debit remain unverified. | Run one account task, retain prompt and settings, inspect the full clip, confirm source rights, and check ZMS AI pricing before batching. |
FAQ
Answers about inputs, duration, resolution, advanced controls, credits, and service status.
It prepares text or one-image video tasks. Reference video is temporarily unavailable.
Use text or one start image. Reference video, end frame, and audio are unavailable.
Text and image controls allow 2–15 seconds; acceptance still depends on the service.
ZMS lists 720p/1080p. Text submits 16:9/9:16; image mode relies on source composition.
No. It remains hidden until its live path is restored and retested.
Multi-shot depends on prompt expansion. Treat both as request settings until returned media is reviewed.
The client estimates 20 credits/second at 720p and 30 at 1080p; signed-in debit is authoritative.