Seedance Creator

The Core Problem: Complexity in AI Media Creation

Using Seedance 2.0 on ByteDance's Jimeng platform for high-quality image and video generation presents two primary challenges: crafting structured prompts that precisely control cinematography, style, and duration, and efficiently executing batch or customized generation tasks via command-line tools.

How It Works: Dual Roles and Core Capabilities

Seedance Creator acts as a dual-role assistant: a Prompt Engineer and a CLI Executor.

Core Capability Matrix

This skill directly encapsulates five core generation modes of Seedance 2.0:
* Text-to-Image (text2image): Generates images from Chinese descriptions, supporting specified ratios and resolutions.
* Text-to-Video (text2video): Creates videos lasting 4-15 seconds at various resolutions.
* Image-to-Image (image2image): Performs style transfer or content modification on a local image.
* Image-to-Video (image2video): Generates dynamic video content with a specified image as the first frame.
* Video Extension: Smoothly extends an existing video in segments.

Key Workflow and Prompt Engineering

The skill interacts with the Jimeng API via the dreamina CLI, verifying the environment and login status beforehand. Its prompt generation adheres to Seedance 2.0 platform specifications:
* Cinematography Lexicon: Offers standard keywords for shot scales (e.g., "extreme wide shot"), camera movements (e.g., "dolly shot"), and rhythms (e.g., "slow motion").
* Style Vocabulary: Includes terms for visual quality (e.g., "cinematic look"), artistic styles (e.g., "cyberpunk neon", "Chinese ink painting").
* Timestamp Storyboarding: For longer videos, it uses timestamps to plan shots precisely, e.g., [00:00 - 00:05] Wide shot, celestial mountains in mist.

All commands and prompts are written in Chinese. After generation, the skill parses the returned JSON results and directly provides downloadable image or video links.

Applicability and Key Considerations

• Platform Dependency: Deeply integrated with the Jimeng platform and Seedance 2.0 model. Generation quality is limited by the platform's model capabilities.
• Input Constraints: Image uploads support formats like jpeg/png, with a maximum of 9 per request. Single video generation is capped at 15 seconds; longer content requires a segmented strategy.
• Content Moderation: The platform automatically blocks uploads containing photorealistic human faces. This limitation must be circumvented.
• Dual-Mode Output: The skill can either act as an "executor" to complete tasks directly via CLI, or only as a "prompt engineer" to output structured prompts for manual use on the Jimeng web interface, catering to different scenarios.