Wan - AI Video Generator
Explore APIChoose an AI palette for artistic style.
Wan 2.6
Create Video with Character Reference
Nothing here yet.
Your assets will appear once published.Go To PlaygroundWan — AI Video Generator
3 specialized video generation models by Alibaba. Character reference videos, image animation with synced sound, and realistic character animation.
Wan 2.6
- UPLOAD REFERENCE (VIDEO 1)* — video upload, required
- UPLOAD REFERENCE (VIDEO 2) — optional second video reference
- PROMPT — up to 1000 characters
- ASPECT RATIO — Portrait or Landscape only
WAN 2.5
- REFERENCE IMAGE* — image upload, required
- PROMPT — up to 3000 characters
- DURATION — 5 sec or 10 sec dropdown
Wan 2.2 Animate
- UPLOAD CHARACTER IMAGE* — image upload, required
- UPLOAD REFERENCE VIDEO — optional video upload
- No prompt field — visual input only
Model Details & Workflows
Wan 2.6 — Two-Video Reference Workflow
Wan 2.6 accepts two reference videos (one required, one optional) and generates a new video based on visual patterns extracted from both. This dual-input approach allows for more nuanced character motion synthesis. You may also provide a text prompt (up to 1000 chars) to guide the generation. Aspect ratio is limited to Portrait or Landscape — Square output is not supported.
WAN 2.5 — Image-to-Video with Audio Sync
WAN 2.5 takes a static reference image and generates an animated video. You can provide a prompt (up to 3000 characters) and select output duration (5 or 10 seconds). This model is optimized for creating smooth animations from single-frame inputs with audio-synchronized motion where applicable.
Wan 2.2 Animate — Visual-Input-Only Generation
Wan 2.2 Animate works purely from visual inputs. Upload a character image (required) and optionally a reference video. This model does not use text prompts — all generation parameters come from image analysis and optional motion reference. This approach can produce highly consistent character animations without prompt interpretation overhead.
Input Guidelines
Reference Videos for Wan 2.6
Upload clear, well-lit video clips with a single subject against a simple background. Avoid fast cuts, scene changes, or multiple people. Keep reference videos under 10 seconds for best motion extraction quality. The reference person's proportions do not need to match your target character exactly, but extreme differences may produce unnatural results. Gross body motion (walking, gesturing, turning) transfers well; subtle expressions and micro-gestures often get lost.
Character Images for Wan 2.2 Animate and WAN 2.5
Start with clear, well-lit still images of your character. Transparent or solid-color backgrounds reduce generation artifacts. Wan analyzes pose, proportions, and visual features, so consistency in these elements produces the most reliable results. Full-body and upper-body poses both work well.
Aspect Ratio and Duration
Wan 2.6 supports Portrait (vertical) and Landscape (horizontal) aspect ratios only — Square is not available. WAN 2.5 can generate videos in either 5-second or 10-second durations. Choose based on your animation complexity and output requirements.
Prompt Guide (Wan 2.6 & WAN 2.5)
"The character walks forward confidently, arms swinging naturally, looking straight ahead" Clear directional movement with natural secondary motion. Both models handle walking animations well when direction and gait are described simply.
"The character raises their right hand and waves slowly, slight smile, head tilts to the left" Specific body part instructions with emotional cues. Including which hand and direction improves accuracy for upper-body animations.
"The character dances with energetic hip-hop movements, arms and legs in full motion" High-energy motion prompt. Models can generate dance movements but they tend to be generic interpretations. For precise choreography, use reference video instead.
Honest Limitations
- Character proportions may shift. During animation, limb lengths, head size, and body proportions can drift from the original character image, especially in longer or complex animations.
- Hands and fingers remain problematic. Like most current video generation models, Wan frequently produces malformed hands — extra fingers, merged digits, or anatomically impossible bending during motion.
- Wan 2.6 has no Square aspect ratio. This model only supports Portrait and Landscape output. You cannot generate square videos with Wan 2.6.
- Wan 2.2 Animate does not accept text prompts. This model relies entirely on image and optional reference video inputs. Text descriptions have no effect on generation for this model.
- Reference video extraction has limits. Fast movements, occlusions, and unclear footage produce poor motion extraction. Best results come from clean, single-person reference clips with steady lighting.
- No environmental interaction. Characters do not realistically interact with objects, furniture, or surfaces. A character attempting to sit may hover above a chair or clip through it.
- Multi-character scenes are unreliable. All Wan models are optimized for single-character animation. Scenes with two or more characters often produce merging artifacts, inconsistent motion, or one character being ignored entirely.

