dreamactor-video

What This Skill Does

Dreamactor-video is a powerful AI agent skill designed for high-fidelity motion transfer. It utilizes the ByteDance Dreamactor V2 model to seamlessly animate a static character image using a driving video. The AI analyzes the motion, facial expressions, and lip-sync information from the input video and applies those patterns to the subject in the reference image while preserving the subject's identity and background integrity. This skill is optimized for both human subjects and non-human characters, including pets and animated figures.

Installation

To integrate this skill into your OpenClaw agent, use the CLI command: clawhub install openclaw/skills/skills/hexiaochun/dreamactor-video

Use Cases

• Personalized Content Creation: Transform static portrait photos into talking or dancing videos for social media engagement.
• Animation and Gaming: Quickly generate animations for character sprites or 3D character avatars using reference footage.
• Marketing & Social Media: Create dynamic product or mascot presentations by having static images "perform" promotional video content.
• Accessibility & Communication: Help individuals or characters "speak" or express emotions in videos through mouth-sync and facial expression mapping.

Example Prompts

1. "I have an image of my dog and a video of a dancing cat; can you use the dreamactor skill to make my dog dance like the cat?"
2. "Please use this portrait photo and this video of a person talking to create a new clip where the person in the photo is saying what is in the video."
3. "Apply the movements and expressions from this dance video to the character in my reference image using Dreamactor."

Tips & Limitations

• Input Quality: Always provide high-resolution, clear images where the subject is easily distinguishable from the background. Image files should not exceed 4.7MB.
• Video Duration: The skill supports videos up to 30 seconds. Keep in mind that longer videos require more processing time and compute credits.
• Transitions: We recommend keeping trim_first_second set to true to ensure the final output video begins smoothly without initial transition artifacts.
• Performance: While the model is robust for various subjects, ensure the driving video contains clear, distinct movements for optimal results. Complex, blurry, or low-light video inputs may lead to reduced accuracy in motion transfer. Remember that billing is consumption-based per second of output generated.