Wan 视频生成

What This Skill Does

The Wan 视频生成 (Wan Video Generation) skill leverages the advanced Wan 2.6 model to create videos from static images or generate new videos based on existing ones, ensuring character consistency. This skill is ideal for users who want to animate images, create videos from a textual description and an image, or maintain consistent subjects across multiple video clips.

It offers two primary functionalities:

1. Image-to-Video: Transforms a single image into a dynamic video. This is perfect for bringing static visuals to life.
2. Reference-to-Video (R2V): Generates new videos using 1-3 reference videos to maintain the consistency of subjects (people, animals, objects) across the generated output. This is crucial for scenarios where maintaining character identity is paramount.

The skill provides different model versions: wan/v2.6/image-to-video for standard, high-quality image-to-video generation, wan/v2.6/image-to-video/flash for a cost-effective and faster alternative, and wan/v2.6/reference-to-video specifically for the R2V functionality.

Billing is based on resolution and duration, with different pricing tiers for standard/R2V models and the Flash version.

Installation

To install this skill, use the following command:

clawhub install openclaw/skills/skills/hexiaochun/wan-video

This skill is part of the openclaw/skills source repository, authored by hexiaochun.

Use Cases

• Animated Photos: Bring family photos, product images, or artwork to life.
• Character Consistency: Create videos where the same character appears in different scenes with a consistent look.
• Concept Visualization: Generate video previews for ideas or storyboards based on a single image.
• Social Media Content: Quickly produce engaging video content for platforms like TikTok, Instagram Reels, or YouTube Shorts.
• Marketing Materials: Develop short, dynamic video ads or promotional clips.
• Artistic Expression: Explore creative possibilities by animating static art or generating abstract video sequences.
• Virtual Try-ons: Potentially generate videos showing how an outfit looks on a model based on an image.

Example Prompts

1. "Create a 5-second, 720p video of a cat jumping from a chair, using this image as a reference: [image URL]"
2. "Generate a 10-second video with 1080p resolution. I want @Video1 dancing in a futuristic city. Use this video as a reference for the character: [reference video URL]"
3. "Animate this image of a landscape [image URL] into a 15-second video, showing clouds moving across the sky. Use the Flash version."

Tips & Limitations

• Reference Video Quality: For the R2V model, ensure your reference videos have a frame rate of at least 16 FPS for optimal results.
• Prompt Engineering: Be descriptive in your prompts. For R2V, clearly indicate which video (@Video1, @Video2, etc.) contains the subject you want to replicate.
• Duration for R2V: The reference-to-video model strictly supports only 5-second or 10-second durations.
• Cost Consideration: The Flash version offers a cost-effective alternative to the standard image-to-video model.
• Resolution: Choose resolutions (720p or 1080p) based on your quality and cost requirements.
• Negative Prompts: Utilize negative_prompt to exclude unwanted elements or characteristics from your generated videos, improving output quality.
• Prompt Expansion: enable_prompt_expansion can help the model better interpret and enhance your prompts, especially for complex scenarios.