Pro专属WAN AI is Alibaba’s open video foundation model family for turning text prompts and images into short cinematic clips. The suite supports text-to-video, image-to-video, Wan-VAE encoding, bilingual visual text, consumer-GPU tiers, and audio or multi-shot workflows when the tool you use supports them.
12比例
分辨率
时长
3Watch the official Alibaba channel demo for WAN 2.3 motion, dance, and character movement before you draft your own text-to-video or image-to-video prompt.
These examples focus on the parts that matter in video: subject, action, camera movement, style, mood, and the details that should stay recognizable—whether you generate on wan.video, in VidMage, or on a self-hosted Wan stack.
A product-ad prompt for skincare film layouts—clean bottle hero, soft studio light, slow push-in.
A skincare bottle on a clean vanity, soft daylight, slow push-in camera, dewy highlights on glass, keep label shape and color stable.
A product motion prompt with controlled rotation, stable logo-free details, clean reflections, and a clear final frame.
A clean sneaker rotates on a studio turntable, soft shadow on warm paper backdrop, slow product reveal, keep the shape and color stable.
A beauty-editorial prompt with flowing gown, coastal wind, soft golden light, and stable face framing.
A fashion model in a white linen dress walks a coastal path at golden hour, soft wind in hair, slow push-in camera, warm skin tones, keep face and dress silhouette stable.
An action prompt with fast lateral camera motion, coral jacket locked, and a sharp face in the rain-soaked neon frame.
A young woman in a coral running jacket sprints through a rain-soaked neon city, dynamic tracking camera, keep jacket color and face framing stable.
A character and style prompt with outfit detail, fabric motion, and consistent face framing.
A model turns at the end of a runway, silver jacket catches light, fabric moves naturally, camera circles once, keep the outfit and hairstyle stable.
A mood-led city prompt with wet pavement reflections, soft handheld energy, and a smooth lateral tracking move.
A camera tracks along a rainlit city street, wet pavement reflections, soft teal ambience, distant headlights, handheld cinematic energy, keep exposure stable.
WAN AI works best when the input describes the whole shot, not just the subject. These features turn a loose idea into clearer video instructions: input type, open-weight encoding, visual text, camera movement, and audio or multi-shot intent.
Describe the subject, action, setting, mood, camera movement, and style. WAN can turn that written idea into a short video scene with visible motion.
Start with a product photo, character image, first frame, or visual reference. WAN can add motion while keeping the original subject and style recognizable.
The Wan 2.1 suite ships as open weights with a Wan-VAE stack built for efficient 1080P-capable encoding. Self-host when you need reproducible pipelines; use wan.video when you want a hosted workflow—check which version is active either way.
Wan 2.1 can render Chinese and English text inside the frame. Spell the exact words, language, placement, and contrast in the prompt—then review legibility in the output.
WAN is useful when motion is the main idea: tracking cameras, push-ins, orbit moves, and action beats. Name one clear camera path plus physical action so the clip reads as video.
Wan 2.5 and newer hosted tools support synchronized audio and multi-shot storytelling. Write ambience, dialogue, and beat-to-beat shot notes early—then confirm whether your tool exports sound and how many shots one run supports.
Choose WAN AI when a short clip needs open-weight flexibility or hosted wan.video access, bilingual visual-text prompts, clear camera direction, and enough room to match clip length and GPU limits to your setup.
Start from a written scene or animate a still frame when the subject, pose, product, or first-frame composition should guide the clip.
WAN is widely used when teams want GitHub weights for local pipelines or wan.video for a hosted generator—pick the path that matches your compliance and GPU story.
Use character or reference inputs when identity, outfit, product shape, or brand styling must stay readable while the scene moves.
Name on-screen words, language, and placement when signage or title cards matter—Wan’s bilingual text story only helps when the prompt is explicit.
Pick vertical, square, or wide ratios before generating so WAN composes for Reels, Shorts, feeds, or cinematic previews.
When ambience, performance, or multi-shot structure matters, write beats into the brief early and confirm the live wan.video or host options before production.
Not sure which Wan version fits your clip? Compare 2.1, 2.2, and 2.5 on video quality, clip length, wait time, on-screen text, and sound—then check which version your tool actually offers before you generate.
| Features | Wan 2.1 | Wan 2.2 | Wan 2.5 |
|---|---|---|---|
| When it launched | Feb 2025 — first open release you can download and run yourself | 2025 — Alibaba demo line with smoother motion | 2025–2026 — mainly on wan.video and partner sites |
| Sizes available | Small (1.3B) and large (14B) models for text or image to video | Large builds plus lighter mixture-of-experts options | Cloud plans on wan.video — no file to install |
| GPU memory (run at home) | About 8 GB for the small model | About 12 GB or more for large local builds | Not needed — renders on Alibaba’s servers |
| Max video quality | Best around 720p; can reach 1080p with the Wan video encoder | 720p to 1080p | Up to 1080p on hosted downloads |
| Frames per second | About 24 fps | About 24 fps | Up to 24 fps |
| How long each clip can be | About 5 s on a home GPU at 480p; longer clips possible with the full download | Usually 5–10 s per clip | Up to about 15 s, including multi-shot storyboards |
| How long it takes to generate | About 4 min for a 5 s 480p clip on an RTX 4090 (small model) | Often 3–8 min on cloud hosts at 720p–1080p | Often 2–5 min per 1080p clip in the cloud queue |
| On-screen text (Chinese + English) | Yes — write the exact words and where they should appear | Same as 2.1 | Yes — say which language the sign or title should use |
| Built-in sound & multiple shots | No built-in sound — add music and voice in your editor | Some demos include performance audio — check your tool | Synced sound and multiple shots in one run on hosted exports |
| Extra editing tools | Start/end frame control plus editing plugins in the open download | Community tools that extend the 2.1 download | Text-based edits inside hosted apps |
| Good for | Running Wan on your own GPU, open downloads, ComfyUI workflows | Dance, performance, and smoother character motion | wan.video without installing anything — longer clips with sound included |
Use VidMage to turn a WAN video idea into a clear brief: choose the input, describe the shot, add camera or audio notes when needed, generate the video, then refine the parts that drift.
Use Text to Video when you want a new scene from a written brief. Use Image to Video when a product still, character frame, start frame, or style reference should guide the clip.
Write the subject, action, camera movement, mood, lighting, on-screen text intent, clip length, aspect ratio, and the character or audio notes that should stay readable.
Generate the first clip, review what drifted in motion, character, or text legibility, then refine the prompt, reference image, or camera notes until the cut matches the scene.
WAN AI fits creators and teams that need open-weight video experiments or hosted wan.video delivery—with clear motion, visual-text prompts, and reusable direction across social clips, ads, and product scenes.
Turn quick ideas into vertical video concepts for TikTok, Reels, Shorts, effect trends, and mute-first caption edits.
Create product clips, launch visuals, ad concepts, and seasonal scenes with clear framing for 9:16, 1:1, or 16:9 campaign edits.
Explore alternate hero shots, campaign beats, product moments, social cuts, and reference-led variants before production starts.
Prototype action beats, camera movement, lighting ideas, transitions, and stylized shots before committing to a larger edit.
Test visual systems, mood, product styling, reference frames, camera language, and subject consistency before choosing the final direction.
Prepare explainers, demonstrations, scenario clips, and visual examples where the viewer needs to understand motion or cause and effect.
Describe the shot or upload a starting frame. Note how it should move, then hit Generate when you're ready.
