generate-visual-prompt

Generate Visual Prompt

Transform a visual concept into a detailed, model-optimized prompt for image or video generation with 2–3 variations and model-specific guidance.

When to use

• User has selected a visual from ideation and wants a generation prompt
• User wants to refine a prompt for a specific generation model (Flux, SDXL, video)
• User wants multiple prompt variations to explore different interpretations

Inputs to gather

• Visual concept — Description of the visual to create (or reference to ideation.md)
• Project context — Project name (if not active) and any style/brand constraints from project.yaml
• Preferred model — Flux | SDXL | video | user preference
• Technical specs — Desired resolution, aspect ratio, format
• Iteration constraints — Any elements to emphasize or avoid

Procedure

1. Load project context. Check for an active project; if not, ask the user to specify. Read project.yaml for style preferences, target platform, and resolution guidelines.

2. Clarify the visual concept. If the user references an idea from ideation.md, read that file. If providing a new concept, confirm the visual type, mood, and purpose.

3. Determine generation method. Identify whether this is text-to-image, image-to-image, or video generation based on the concept and user intent.

4. Structure the prompt. Use this framework: [Subject] + [Style] + [Composition] + [Lighting/Mood] + [Technical Specs]

   • Subject — Main focus, key elements, actions or states
   • Style — Artistic style (photorealistic, illustration, 3D render, etc.); color palette; reference artists if appropriate
   • Composition — Framing (close-up, wide, etc.); perspective; layout and focal points
   • Lighting/Mood — Lighting type; atmosphere; emotional tone
   • Technical — Aspect ratio, resolution, any platform-specific needs

5. Generate 2–3 prompt variations. Create variations that:

   • Explore different style approaches or moods
   • Vary composition or emphasis
   • Test model-specific keyword strategies

6. Add negative prompts. If relevant, include negative prompts to avoid unwanted elements (blurry, low quality, watermarks, oversaturated, stock photo feel, etc.).

7. Provide model-specific tips. Recommend:

   • Flux — Excellent at detailed prompts, photorealistic and artistic styles, handles text reasonably
   • SDXL — Strong for architecture and product imagery, benefits from style keywords
   • Video — Keep motion simple, specify camera movement, shorter prompts often work better

8. Save the prompt YAML. Write to ${CLAUDE_PLUGIN_DATA:-$HOME/.local/share/claude-plugins}/visual-communications/projects/<project-name>/prompts/<slug>.yaml:

   name: <Visual Name>
   created: <date>
   model: <flux | sdxl | video>
   prompt: |
     <full generation prompt>
   negative_prompt: |
     <optional negative prompt>
   settings:
     aspect_ratio: <ratio>
     style: <style>
     resolution: <resolution>
   variations:
     - prompt: |
         <variation 2>
     - prompt: |
         <variation 3>
   generated_files: []
   notes: |
     <any additional guidance or iteration notes>
   

9. Present to user. Show all prompts with recommended settings. Solicit feedback, iterate if needed, and finalize.

Output / side effects

• Prompt YAML saved to projects/<project-name>/prompts/<slug>.yaml
• 2–3 prompt variations documented with model recommendations
• generated_files list initialized (to be populated after generation)
• Ready for /run-fal-generation to execute

Safety / constraints

• Follow target-platform resolution guidelines from project.yaml.
• Do not include sensitive or private information in prompts.
• Test model-specific tips; some models may benefit from more or fewer descriptors.