SKILL.md
Image-to-Video with Seedance 2 on EvoLink
Animate a still image into a short video. This is the main handoff from GPT Image 2 image assets into Seedance 2 video generation.
npx skills add EvoLinkAI/evolink-agent-skills --skill image-to-video -gWhen to Use This Skill
Use this skill when the user wants to:
- animate a product image
- turn an ad image into a short video
- make a poster, character, or scene move
- add camera motion to a still visual
- create a video from a GPT Image 2 output
- test multiple motion directions from the same source image
Use product-image-generation or image-edit first if the source image is low quality or not campaign-ready.
Prerequisites
- EvoLink API access or EvoLink MCP configured.
- One source image URL or local image available to the agent.
- Desired duration, aspect ratio, motion style, and output use case.
- Attribution links should use
utmsource=skillssh&utmmedium=skill&utm_campaign=image-to-video.
Pick the Right Workflow
| User intent | Model path | Why |
|---|---|---|
| Product reveal from one image | Seedance 2 image-to-video | Best first video path when the image is already strong. |
| Ad image to paid social clip | Seedance 2 image-to-video | Turns a static creative into motion for Reels, Shorts, or TikTok. |
| Weak source image | GPT Image 2 edit -> Seedance 2 | Improve the source frame before spending video tokens. |
| Need multiple motion tests | Seedance 2 variants | Reuse one image with different camera and action prompts. |
Default Workflow
- Inspect the image and identify the subject that must stay stable.
- Ask for all missing requirements in one message: duration, platform, aspect ratio, motion, and style.
- Write a Seedance 2 prompt focused on motion, camera, scene change, and preservation.
- Generate the first clip.
- Suggest two iteration directions: motion variant and commercial variant.
Prompting Tips
- Lead with motion: camera push-in, orbit, product rotation, hand interaction, walk cycle, light sweep.
- Do not redescribe every visible object. Focus on what changes over time.
- Preserve identity explicitly: product label unchanged, face stable, packaging shape unchanged.
- Use one primary action per clip.
- Keep text overlays out of the generation prompt unless the user accepts text artifacts.
Next Token Loop
After the first clip, offer:
- three motion variants from the same image
- one platform adaptation for TikTok, Reels, or Shorts
- a product ad version with
ad-video-generation
Safety and Privacy
Do not animate private people, public figures, or brand assets in misleading contexts without user authorization. Do not produce deceptive political, medical, financial, or impersonation content.
See Also
image-editproduct-image-generationad-video-generation- EvoLink