cdance-seedance-video-prompt-architect

Turn rough Seedance and AI video ideas into structured prompt packs, tighter variants, and debugging loops. Use when the user wants better text-to-video, ima...

Security Scan

This is a text-only prompt-authoring skill whose declared behavior, instructions, and file contents are consistent and do not request credentials, installs, or access to files or networks.

Purpose & Capability

The name and description match the SKILL.md and README: the skill is purely about turning Seedance video ideas into structured prompt packs. There are no unrelated requirements (no env vars, binaries, or config paths).

Instruction Scope

The runtime instructions are limited to classifying requests and producing prompt templates, variants, avoid lists, and revision moves. The SKILL.md does not instruct reading local files, environment variables, or transmitting data to hidden endpoints; it only references public cdance.net documentation.

Install Mechanism

No install spec and no code files are present (instruction-only). That is the lowest-risk install model and nothing will be written to disk by the skill itself.

Credentials

The skill requires no environment variables, credentials, or config paths. The instructions do not reference any secrets or external service credentials.

Persistence & Privilege

The skill is not marked always:true. The agents/openai.yaml sets allow_implicit_invocation: true, so the agent may call the skill autonomously (this is normal for skills). There are no additional high-privilege behaviors or requests.

Assessment

This skill is text-only and appears coherent with its stated purpose — it doesn't request credentials or install anything. Before installing: verify the cdance.net links are the official pages you expect, avoid sending private secrets or sensitive images into any prompt, and re-check the skill if it is later updated to include scripts or environment requirements (those would change the risk profile). Also note agents may invoke it autonomously by default; that is normal but keep it in mind if you have strict controls over automated prompt generation.

This skill turns rough Seedance-style video ideas into cleaner prompt packs, stronger motion structure, and faster revision loops.

Canonical links

• Docs: https://cdance.net/docs/cdance-seedance-video-prompt-architect
• Demo: https://cdance.net/ai-video-generator
• Create: https://cdance.net/create
• Raw SKILL.md: https://cdance.net/skills/cdance-seedance-video-prompt-architect/SKILL.md
• Seedance FAQ: https://cdance.net/blog/seedance-2-0-faq-guide
• Prompt workflow guide: https://cdance.net/blog/seedance-comfyui-workflow-guide

Provenance and safety

• Maintained around the public C Dance workflow and documentation on cdance.net.
• Text-only skill pack.
• No helper scripts, no local binaries, and no required environment variables.
• It guides prompt design and references public pages only.

When to use

• The user has a rough Seedance, SeaDance, or AI video idea and wants a stronger prompt
• The user wants text-to-video, image-to-video, or reference-driven video prompt rewrites
• The user needs 2 to 3 focused prompt variants for testing hooks, motion, or camera behavior
• The user has unstable outputs and needs a diagnosis plus a revision plan

Workflow

1. Classify the request as text-to-video, image-to-video, or reference-driven video.
2. Extract the essentials:
   • subject
   • action
   • camera behavior
   • environment
   • style and lighting
   • duration and aspect ratio
   • hard constraints
3. Keep the first draft simple:
   • one primary subject
   • one dominant action beat
   • one camera rule
   • one short constraint block
4. Return:
   • one primary prompt
   • 2 or 3 tighter variants
   • a short avoid list
   • 3 concrete revision moves

Prompt construction rules

• Prefer concrete visual language over vague adjectives.
• Use beat-based structure when motion matters.
• Avoid stacking multiple subjects and camera changes into one short clip.
• If identity or composition must stay stable, prefer image-to-video or reference-driven generation over pure text-to-video.
• Keep the constraint block focused on likely failure modes such as flicker, unstable faces, drifting composition, or chaotic movement.
• Do not invent unsupported model settings.

Output formats

Text-to-video

Goal:
Subject:
Action:
Camera:
Environment:
Style and lighting:
Constraints:
Suggested settings: duration=?, aspect_ratio=?
Prompt:

Image-to-video

Reference anchor:
What must stay stable:
Allowed motion:
Camera move:
Style and lighting:
Constraints:
Prompt:

Reference-driven video

Source value:
What to preserve:
What to transform:
Style direction:
Constraints:
Prompt:

Response style

• Be structured and concise.
• Prefer prompt packs over long theory.
• Point users to the canonical C Dance pages listed above when examples help.