comfly-seedance-tvc-video

name: comfly_seedance_tvc_video description: Generate 20-second storyboard-driven promo videos by first creating two polished 10-second storyboard board images with gpt-image-2 (with automatic fallback to nano-banana-2 when gpt-image-2 fails) and then turning those boards into two Seedance clips with doubao-seedance-2-0-fast-260128. Use when the user provides product references plus a storyboard-board style target and wants a premium TVC workflow for beauty, skincare, FMCG, tea, and other brand-heavy commercial videos. metadata: {"openclaw":{"emoji":"video","homepage":"https://ai.comfly.org"}}

Comfly Seedance TVC

Use this skill when the user wants a product to become:

• two polished storyboard board images
• each board covering about 10 seconds
• each board containing segment design + Chinese voice-over / copy
• two corresponding videos
• optionally merged into one 20-second ad

This skill is intentionally different from comfly_veo3_daihuo_video:

• it accepts multiple reference images
• it produces two complete storyboard board images rather than many single-frame stills
• each board is designed as a 10-second segment with Chinese copy and narration guidance
• it uses gpt-image-2 for the board images by default, and automatically falls back to nano-banana-2 if gpt-image-2 exhausts its retries (e.g. the upstream image=[url] reference path is unstable)
• it uses doubao-seedance-2-0-fast-260128 on the Seedance official-format endpoint for lower-cost testing

Confirmed API pattern

• Upload local images when needed: POST /v1/files
• Analyze reference storyboard image: POST /v1/chat/completions
• Render per-shot storyboard stills: POST /v1/images/generations
• Submit Seedance task: POST /seedance/v3/contents/generations/tasks
• Query Seedance task: GET /seedance/v3/contents/generations/tasks/{task_id}

Default model choices

• analysis model: gpt-4.1-mini
• storyboard board image model: gpt-image-2 (primary), nano-banana-2 (fallback)
• video model: doubao-seedance-2-0-fast-260128

Workflow

1. Upload the primary reference image plus any extra product / packaging / style / scene references.
2. Analyze them into structured board JSON:
   • product / brand consistency notes
   • overall visual style
   • exactly two storyboard boards
   • each board covering about 10 seconds
   • Chinese voice-over text for that 10-second segment
   • English board-image prompt
   • English Seedance motion prompt
3. Render two polished storyboard board images with gpt-image-2 (auto fallback to nano-banana-2 on failure).
4. Submit two Seedance tasks in parallel:
   • board image as first_frame
   • all uploaded references as reference_image
5. Poll until both complete.
6. Optionally merge the two clips into a 20-second ad.

Quality rules

• Each generated image is one complete 10-second storyboard board, not one single micro-shot.
• The board image should include time segmentation, sub-panels, Chinese notes, and Chinese voice-over/copy blocks.
• Use multiple references to keep product identity stable.
• The two videos should be launched in parallel.
• Testing stage defaults to the fast Seedance model; later it can switch to the higher-quality non-fast variant.

Pipeline entry

The Python runtime entry is:

• {baseDir}\scripts\comfly_seedance_storyboard_pipeline.py

It accepts:

• reference_image
• reference_images
• apikey
• base_url
• task_text
• platform
• country
• language
• analysis_model
• image_model
• image_model_fallback
• video_model
• storyboard_count / segment_count
• segment_duration_seconds
• total_duration_seconds
• merge_clips

Important notes

• This skill assumes the uploaded image is a storyboard board or visual shot brief, not just a plain product photo.
• The Seedance task body follows the official-format style:
  • model
  • content
  • ratio
  • duration
  • generate_audio
  • watermark
• For image-to-video, the generated still is sent as role: "first_frame".
• The original storyboard image is sent as role: "reference_image" to help preserve the intended product identity and art direction.

Safety

• Never hardcode API keys into committed files.
• If the user pasted a real key into chat, recommend rotating it after testing.