Wan Alpha

Wan Alpha — Text-to-Video Model with Alpha Channel

What is Wan Alpha?

Wan Alpha is a text-to-video model that generates clips with a real alpha channel, so your subject arrives already cut out on a transparent background — no green screen, no rotoscoping. From a single prompt it returns a WebM video or animated WebP that carries true transparency, ready to drop onto any layer in an editor or browser.

Built on the Wan2.1-T2V-14B foundation by the WeChat AI Computer Vision team, Wan Alpha learns the RGB and alpha channels jointly, so the matte lines up pixel-for-pixel with the color. That design is what makes it reliable for the elements traditional keying struggles with: smoke, glass, glow, and fine hair strands.

Key Features

• •True alpha output (RGBA): WebM or animated WebP with a real transparency channel, ready to composite.
• •Semi-transparent detail: renders smoke, glass, glowing effects, and fine hair edges cleanly.
• •Prompt-driven keying: name the subject, add "transparent background," a shot type, and a style tag.
• •Flexible delivery: 480p and 720p in landscape or portrait, plus 512x512 square presets.
• •Control: frame counts from 81 to 121, playback fps from 5 to 30, quality, and seed for reproducibility.

Best Use Cases

Wan Alpha fits any pipeline that needs transparent video assets. Use it for VFX and compositing, game effects and overlays, AR filters, motion graphics, broadcast lower-thirds, and web animations. Because the alpha matte is generated with the color, edges stay clean and semi-transparency is preserved — situations where chroma-key and manual rotoscoping break down. The transparent clip drops straight onto any background, so creators and developers skip masking and move faster from prompt to finished shot.

Prompt Tips and Output Quality

Always include the words "transparent background" — the model needs it to key the subject. Add a shot type (close-up, medium shot, wide shot) and a style tag such as "realistic style." Keep subjects relatively simple for the cleanest edges; busy scenes can show minor artifacts. Start at 480p and 81 frames to iterate, then move to 720p for delivery. Choose WebM for the widest editor and browser support, and raise output quality toward 95 when you need maximum edge detail.

How to Use Wan Alpha for Transparent Video

Wan Alpha turns a text prompt into a clip with a real alpha channel, so the subject arrives on a transparent background ready to composite. Getting clean results comes down to how you write the prompt and which delivery settings you pick.

Writing the prompt

Structure every prompt around four parts: the subject and its action, the phrase transparent background, a shot type (close-up, medium shot, or wide shot), and a style tag (for example, "realistic style"). The transparent-background phrase is what tells the model to key the subject, so never leave it out. Keep the subject relatively simple for the sharpest edges — a single character, object, or effect keys more cleanly than a crowded scene. Use the negative prompt to steer away from halos or background clutter, or leave it blank to use the model default.

Choosing resolution and length

Pick an aspect that matches your timeline: 832x480 or 1280x720 for landscape, 480x832 or 720x1280 for portrait, or a 512x512 square preset. Start at 480p while you iterate on the prompt, then switch to 720p for final delivery. Frame count runs from 81 to 121 in steps of 4; at the default 16 fps, 81 frames is about 5 seconds and 121 about 7.5 seconds. Raising fps makes playback smoother without changing how many frames are generated.

Recommended Settings

• •Fast iteration: resolution 832x480, num_frames 81, fps 16, output_quality 85, output_format webm.
• •Final delivery: resolution 1280x720, num_frames 81-121, fps 24, output_quality 95, output_format webm.
• •Fine edges (smoke, glass, hair): keep output_quality at 90-95 and prefer WebM so the alpha channel composites directly.
• •Reproducibility: set a fixed seed to lock a look; use -1 for a fresh random result.
• •Edge cleanup: if the first or last frame shows keying artifacts, enable remove_first_frame or remove_last_frame.

Choose animated WebP only when you specifically need that format; WebM composites in the widest range of editors and browsers.

Do I have to write "transparent background" in the prompt?

Yes. It is required for reliable keying; without it the subject may render on a solid background.

What output formats does Wan Alpha produce?

WebM video or animated WebP, both carrying a real alpha channel. WebM composites directly in most editors and browsers.

Can it render semi-transparent effects like smoke or glass?

Yes — semi-transparent objects, glow, and hair detail are a core strength of the model.

How long are the clips?

Frame count ranges from 81 to 121 in steps of 4; at 16 fps, 81 frames is about 5 seconds and 121 about 7.5 seconds.

What resolutions are available?

832x480 and 480x832 (480p), 1280x720 and 720x1280 (720p), and two 512x512 square presets.

Does it support non-English prompts?

Yes, the underlying model accepts both English and Chinese prompts.