How to Create, Edit, and Extend Videos with Wan 3.0

How to use Wan 3.0 on KreadoAI: write a timed prompt, add image, video, and audio references, then edit or extend the clip. Frame mode stays separate.

This tutorial shows how to create, edit, and extend a video with Wan 3.0 on KreadoAI. You will open the model, pick one generation mode, paste a prompt, and check a real clip. The walkthrough is one perfume spot, then a reference generate, a clothing edit, a greenhouse extend, and a separate first-and-last-frame request.

The version differences are in explore the new features in Wan 3.0. Stay on this page for the steps, the uploads, and the prompts you can paste.

Access Wan 3.0 on KreadoAI

Open the Wan 3.0 AI Video Generator and start a video from the model picker. Choose Wan 3.0 Standard (wan3.0-video) while you are still changing the prompt. Choose Wan 3.0 Prime (wan3.0-video-prime) when the prompt is stable and you want the next takes faster. Prime is a speed setting. It does not add editing, audio, or a longer ceiling.

Set the ratio before you generate. 16:9 is the right first choice for the examples below. 9:16 is the right choice when the delivery is a phone feed. Resolution is 480P, 720P, or 1080P. Use 720P until the shot is the one you will publish. Length is anywhere from 2 to 30 seconds. The examples use 6, 8, and 10 seconds because a short failed take is cheaper to replace.

Labels in the picker can move. Follow the names on the screen: text or no references, reference uploads, edit, extend, and first and last frames.

Choose the right generation mode

Each mode has its own upload. Use the row that matches the job before you attach files.

Mode What you upload What you do not upload in that same request
Text to video Prompt only Images, videos, or a first/last pair
Multimodal reference Images, videos, and audio. Caps: 10 images, 5 videos, 5 audio, 20 total A first-frame and last-frame pair
Edit One existing video, plus the change you want A new last-frame still as if it were frame mode
Extend One existing video, plus where time should continue A stack of unrelated reference clips
First frame, or first and last frame One still, or two stills Reference videos, extra images, or audio

Official Wan 3.0 usage separates frame mode from multimodal reference mode. The asset rules are different. If you need the product photo and a motion clip, use reference mode. If you need an approved opening frame and an approved closing frame, use frame mode. Run a second generation for the other job. Do not combine them and hope the model sorts the files.

Write an effective Wan 3.0 prompt

A usable prompt has six parts: the overall description, a citation for each uploaded file, the timeline, the spoken line, the sound, and what to avoid. Cite files as Image 1, Video 1, and Audio 1, in the order you uploaded them. The perfume prompt in the next section is the filled-in text-to-video version.

Text-to-video template:

```text Create a [duration]-second [style] video.

Scene: [Location, light, and atmosphere.]

Subject: [The product or person, and the details that must stay fixed.]

Shot 1 (00:00–00:05): [Camera, action, and what is in frame.]

Shot 2 (00:05–00:10): [Next camera, next action.]

Audio: [One spoken line, the music, and the effects.]

Visual style: [Lens, color, and mood.]

Avoid: [Extra products, subtitles, logos, watermark.] ```

Edit template:

```text Edit Video 1: Replace [the specific element] with [the desired element].

Keep: [What must not change: layout, product, camera path.]

Change: [Light, weather, clothing, or the spoken line.]

Maintain the original scene composition and camera movement. ```

Extend template:

```text Extend Video 1 forward by [X] seconds.

Starting from the final moment of the original clip: [Where the subject is, and what it is doing.]

Continue with: [The next movement and the camera direction.]

End with: [The last frame you want.] ```

The original clip plus the extension must stay within 30 seconds total. If the source is already 10 seconds, ask for 5 more seconds, not 25.

Generate a video from text

This pass has no uploads. The bottle has to come from the words, so the prompt repeats the same frosted glass and gold cap in every shot and gives the voice one line.

Paste this, set 10 seconds, 16:9, 720P, and turn audio generation on:

```text Create a 10-second beauty commercial for one frosted glass perfume bottle with a pale gold cap. The bottle has no label, no letters, no logo, and no brand name. The bottle design stays identical in every shot.

Shot 1 (00:00-00:04): The bottle stands on warm linen beside a bright window. Slow push-in. Soft morning light.

Shot 2 (00:04-00:07): Macro of the gold cap and the frosted glass. A fine mist hangs in the light. Do not add text anywhere.

Shot 3 (00:07-00:10): A hand sets the same bottle down and leaves the frame. The bottle stays centered.

Audio: a soft glass tap, a quiet mist, and one calm voice saying exactly: "One drop. All evening." No subtitles, no on-screen text, no watermark. ```

The 10-second result below was generated on Wan 3.0 Standard at 720P with audio generation on. Watch it with sound. The line to check is "One drop. All evening." The picture to check is whether the bottle stays one object, and whether any letters appeared on the glass. If a logo shows up, add "no text, no brand name, and no letters on the glass" and generate again.

If the bottle changes between shots, do not add more adjectives. Repeat one physical description in Shot 2 and Shot 3, and cut any second product. If the voice is missing, confirm audio generation is on and that the Audio block contains one short sentence, not a slogan list.

Create videos with reference materials

Text will redesign a real product. Reference mode is the fix. Upload the files first, then name each one's job.

For this perfume spot the uploads are:

  • Image 1: the frosted bottle with the gold cap, the object that must survive into the final frame
  • Image 2: the dressing table, used only as the room
  • Video 1: a hand releasing a mist, used for the motion
  • Audio 1: a short mist and glass tap, used for the sound

Prompt:

```text Create a 10-second beauty commercial.

Image 1 is the frosted glass perfume bottle. The bottle in every shot must match Image 1: same glass, same gold cap, no label. Image 2 is the dressing table. Use Image 2 only for the room, the window light, and the table. Video 1 is the spray motion. Follow the hand and the mist in Video 1. Audio 1 is the mist and the glass tap. Use Audio 1 under the picture.

Shot 1 (00:00-00:04): The hand from Video 1 releases a mist beside the bottle from Image 1. Shot 2 (00:04-00:07): The bottle from Image 1 stands on the table from Image 2. Slow push-in. Shot 3 (00:07-00:10): The same bottle stays centered as the hand leaves the frame.

No letters, no logos, no subtitles, no watermark. ```

Set reference mode, 10 seconds, 16:9, 720P, audio on. The clip below is that job.

Check three things before you call it done. The gold-cap bottle from Image 1 should still be in frame. The hand should release a mist. The mist sound should sit under the picture. In this take the spray bottle from Video 1 also entered the shot, because the prompt did not forbid it. Add one line: "Do not bring the bottle from Video 1 into the frame. Use only the hand and the mist." If the room wins and the gold-cap bottle changes, say again that Image 1 controls the product and Image 2 controls only the background.

Do not add a last-frame still to this request. If you also need a locked ending frame, generate that as its own frame-mode job.

Edit an existing video

Edit mode takes Video 1 as the footage to change. On this workflow the duration follows the source clip, so you are rewriting the picture, not asking for a new length.

Source, a daylight walk in a beige coat:

Prompt:

```text Edit Video 1: Replace the beige wool coat with a black leather jacket.

Keep: The woman's face, hair, walk, the sidewalk, and the camera path.

Change: Only the coat. Beige wool becomes a fitted black leather jacket.

Maintain the original scene composition and camera movement. No new street, no rain, no subtitles, no watermark. ```

Result:

The same walk and the same sidewalk, with the coat swapped. Ask for the jacket, the weather, and a new line in one prompt and you usually get one of the three.

Extend a video

Extend mode needs the source video and a description of the moment you are leaving. Say which direction. Forward continues after the last frame. Backward adds time before the first frame. Both directions add time on each side. Keep the sum of the original and the new seconds at or under 30.

Source, a slow glide through a greenhouse:

Prompt:

```text Extend Video 1 forward.

Starting from the final moment of the original clip: The camera is still moving forward at eye level between the same rows of ferns.

Continue with: Keep gliding down the same aisle. No cut to a new room, no people.

End with: A slightly wider view of the same greenhouse aisle. ```

Extended clip:

Judge the join, not a new location. A correct forward extend looks like more of the same shot. If you wanted a different ending, that is an edit, or a new text-to-video, not an extend.

Use first and last frames

Frame mode is for a move you have already art-directed at both ends. Upload the opening still as the first frame and the closing still as the last frame. Write only the motion between them.

text
The closed taupe box on the linen opens, and the frosted perfume bottle with the gold cap rises to stand in front of the open box. Same box, same bottle, no letters, no logo. Slow, one continuous move.

The 8-second clip below is that request: closed box as the first frame, open box with the bottle as the last frame, 16:9, 720P. No reference video and no audio file were attached.

If the ending frame is ignored, you probably attached it inside a reference job. Remove the extra files and run frame mode alone.

Choose Standard or Prime

  Standard Prime
When You are still rewriting the prompt The prompt is stable and you want the next takes sooner
K-coins per second at 480P / 720P / 1080P 6 / 12 / 24 9 / 18 / 36
Audio, references, edit, extend, frames Available Available

A 10-second 720P Standard test is the right first spend. Generate 1080P after the shot, the line, and the product all survive a preview.

Tips for better results

Keep one mode in each request. Frame stills and reference videos do not share a job.

Change one thing per edit. A prompt that changes the jacket, the street, and the dialogue at once usually keeps one of the three.

Repeat the same product description in every shot of a text-to-video. A new adjective in Shot 3 becomes a second object.

Make the clock ranges add up to the duration you set. A 10-second job cannot start Shot 3 at 00:20.

Turn audio generation on only when the prompt has a line or an effect you want in the file. Turn it off when you only want the picture.

Keep the original clip plus the new seconds within 30. If the source is already 24 seconds, ask for a few more seconds, not another 20.

Generate the first Wan 3.0 clip

Pick one mode. Paste one of the prompts above. Set 720P and a length you can watch twice. Change the line that failed, then generate again.

Try Wan 3.0 on KreadoAI

For the version differences and where each mode fits a campaign, explore the new features in Wan 3.0.

FAQ

How do I use Wan 3.0 on KreadoAI?

Open the Wan 3.0 model, choose Standard or Prime, pick one mode, set the ratio, resolution, and a length from 2 to 30 seconds, then generate. Turn audio on only when the prompt contains a line or an effect you want in the file.

What belongs in a Wan 3.0 prompt?

A scene, a subject that stays fixed, timed shots, a citation for every upload, one spoken line if you need a voice, and a short avoid list. Cite uploads as Image 1, Video 1, and Audio 1.

How does Wan 3.0 image to video work?

Upload the still in reference mode, or upload one still as the first frame. Say which file controls the product. Do not also attach a last frame and a reference video in that same request.

How does Wan 3.0 video editing work?

Upload the clip as Video 1 in edit mode. Name the element to replace and the elements to keep. Change one thing per generate. The edit follows the source length.

How long can a Wan 3.0 extension be?

The original clip plus the new seconds must stay within 30 seconds. Say whether the extension runs forward, backward, or both, and describe the frame you are continuing from.