ImageWanVideoGen
Multimodal video studio

Generate Wan 3.0 video from any idea

This page is the studio for Wan 3.0: write a prompt, drop stills, lock a first and last frame, or attach motion, voice, a brief, or a public URL. Stay here from draft to download. Choose resolution, aspect ratio, length, and whether the clip should carry native sound—then generate without bouncing between separate tools.

0 / 20,000

One-click examples

480P: 16 credits/s. Billed as (input video seconds + output seconds) × rate. Failed jobs refund automatically.

Inspiration

A gallery of finished clips

Cinematic, fashion, portrait, and story moments for mood-boarding. These clips show lighting, camera energy, and pacing you can aim for in the generator above—not a catalog of locked templates.

How Wan 3.0 generation works

WanVideoGen is a browser studio for short cinematic clips on the current hosted model. You describe the shot, optionally attach references, set format, and download when the job finishes. Credits are reserved up front and returned automatically if the upstream render fails.

What this model is for

Treat it as a shot engine, not a full editor. It is strongest on a single continuous moment: a hero product orbit, a portrait with wind and breath, a street walk that turns a corner, or a two-beat explainer with titles and voice. It will not replace timeline cutting, color finishing, or licensed stock libraries. Write what should happen in time—subject, camera, lighting, and sound—rather than a mood board of adjectives. If identity or wardrobe must hold, attach a still. If the ending is as important as the opening, use first and last frames. Open-weight Wan video research is published by the Wan-Video team; this site serves Wan 3.0 as a hosted generation with a credit meter and six input modes.

From prompt to download

Sign in with Google, pick a mode, and fill only what that mode needs. Text-only is enough for atmosphere tests. Image mode animates a still while trying to keep face and clothes. Frame mode interpolates a journey between two compositions. Reference mode can mix stills, motion clips, and audio so character, energy, and rhythm travel together. Document and webpage modes read a file or a public URL and turn the main points into a short visual summary. Then choose 480P, 720P, or 1080P, an aspect ratio for feed or landscape, a length from 2 to 30 seconds, and whether native audio should be generated. A seed repeats a look; leave it random to explore. Jobs usually take a few minutes. Stay on the page until the player appears.

Who gets the most from it

Solo creators use it for social drafts before a shoot. Product marketers test packaging and lighting without a studio day. Educators turn a one-pager into a spoken visual. Agencies pitch motion while the live-action brief is still open. If you already have a plate or a voice note, start in reference mode instead of hoping a text prompt invents your brand. Browse the inspiration wall for pacing, then write your own beat sheet in the box above. When you are ready to buy volume, jump to pricing; when you publish, you still own the rights to every asset you uploaded. For field background on text-to-video systems, Wikipedia keeps a public overview.

Wan-Video on GitHub · Text-to-video models (Wikipedia) · See current plans and per-second rates

One model, every way to begin

Keep drafting, referencing, and exporting in one place instead of hopping between a writer, an image tool, an editor, and a separate audio pass.

Multimodal references

Guide identity, wardrobe, camera energy, and rhythm with up to ten images, five motion clips, and five audio files. Use a face still for likeness, a short clip for walk cycle or handheld shake, and a bed of room tone or footsteps so the soundtrack is not invented from nothing. Mix only what you need; extra unused slots are ignored.

First and last frames

Upload an opening composition and a closing one when the destination matters as much as the start. The model has to invent the path between them, so keep geography and wardrobe compatible unless you want a deliberate time jump. This mode is the right default for before/after, travel, and product-to-lifestyle transitions.

Native audiovisual creation

Dialogue, ambience, score, and hit points can be generated with the picture instead of being bolted on later. Say in the prompt who speaks, how loud the room is, and whether music should swell or stay under. You can still turn audio off for a silent plate you will mix in an editor.

Document to video

Drop a PDF, deck, sheet, or text brief when the source of truth is already written. The model reads structure and turns headings into scenes. Keep the file focused: one narrative beats a 40-page dump. Spell out the order of points in the prompt if the document is messy.

Webpage to video

Paste a public URL for explainers and product stories when you do not want to retype the page. Private, login-walled, or blocked sites will not work. Ask for the main idea first, then two supporting details, and say whether you want a narrator or titles only.

Production controls

Pick 480P for cheap iteration, 720P for most social drafts, or 1080P when the frame will be shown large. Six aspect ratios cover stories, landscape, and classic photo crops. Length runs from two to thirty seconds. A numeric seed helps you retry lighting without rewriting the whole prompt.

Plans and per-second rates

480P is $0.08 per output second, 720P $0.16, and 1080P $0.32. Reference clips add their duration to the bill. Credits lock when a Wan 3.0 job starts and return if the provider fails.

Basic

$14.90/mo

For creators getting started

  • 3,000 credits / month
  • All Wan 3.0 generation modes
  • Email support

Premium

$32.90/mo

For power users & teams

  • 6,500 credits / month
  • 480P, 720P, and 1080P with native audio
  • Priority generation queue

Questions before you generate

What can I use as input?

Start from a written prompt, a single still, a first and last frame pair, mixed image/video/audio references, a supported document, or one publicly reachable webpage. You do not need every slot. Empty modes stay empty. Private drive links and login-only pages are not fetched.

Which formats and durations are supported?

Aspect ratios include adaptive, 16:9, 4:3, 1:1, 3:4, and 9:16. Resolution is 480P, 720P, or 1080P. Output length is any whole second from 2 through 30. Adaptive lets the model pick a crop when your references disagree.

Does the clip include sound?

Audio is on by default so dialogue, music, and room tone can land with the picture. Describe who speaks and how dense the mix should be. Switch it off when you only need a silent plate for an editor.

Can I use the files commercially?

Publishing depends on your plan and on rights you already hold to prompts, faces, music, and logos you upload. We do not grant you third-party likeness or trademark clearance. Read the terms before client delivery.

How are credits counted?

You pay for output seconds plus any reference video seconds, multiplied by the resolution rate. A ten-second 720P shot with no motion reference costs ten times the 720P rate. Failed jobs refund the reservation. Unused credits on a subscription follow the plan rules on the pricing section.

Will a person or product stay on-model?

Likeness is much more stable when you attach a clear still and say what must not change: face, outfit, logo placement. Text-only prompts invent a new subject each run. Seeds help you retry lighting; they do not replace a reference image.

How is this different from a timeline editor?

You get one generated take, not a multi-track project. Trim, captions, brand kits, and color boards still belong in the tool you already use after download. Use this page to invent the moving plate, then finish elsewhere.

How long does a job take?

Most renders finish in about one to five minutes depending on length, resolution, and queue. Keep the tab open so the player can attach when the file is ready. If a job errors, credits return and you can resubmit.