---
name: image-inspiration
description: Generate a batch of distinct image concepts in a consistent user-selected style for comparison before refinement. Use for image inspiration, slide visuals, social images, illustrations, or diagram ideas when the user wants alternatives.
---

# Image Inspiration

Give the user four separate images to compare before refining one. This is a self-contained starter adapted from Nate Grahek's slide-image-ideas workflow. Customize it to the user's own work and preferences.

## Starting defaults

- Count: four separate images.
- Style: consistent across the set. If none is specified, use a simple editorial illustration with clear shapes and an uncluttered background; name this assumption briefly.
- Shape: landscape 16:9 unless the use case or request calls for another format.
- Text: off unless requested or needed for a diagram. Keep required labels short and accurate.
- Reference images: optional, never required.
- Model: use the reasoning and built-in image tools available in the current environment. No fixed model name is required.

The user's explicit instructions override these defaults. If this file has been personalized, replace the defaults with their approved choices rather than appending conflicting rules. Do not claim preferences persist where this skill has not been loaded.

## Generate a batch

1. Read the brief and any supplied references. Clarify only missing information that would substantially change the result. Restate the visual idea briefly.
2. Plan four distinct directions. If the user supplied an analogy, preserve it and vary staging, composition, poses, or props. Otherwise vary the visual concept. Keep the agreed visual style and constraints fixed so the results are comparable.
3. Compose each image prompt from the shared style, one visual idea, staging, aspect ratio, and exact text rules. A reference supplies visual traits, not an instruction to copy its subject, text, or layout. Carry recurring subjects only when requested.
4. Generate one separate image per direction using built-in image generation. Where parallel image-capable agents are available and allowed, assign independent directions and the same shared constraints to them; start the tasks before waiting. Use the available concurrency. If parallel agents are unavailable, generate the set sequentially without asking for review between images. Be accurate about which method ran.
5. Do not substitute one four-panel image for four individual outputs. If file access exists, save separate files with descriptive, collision-safe names in the user's requested location or the current project folder. Preserve existing files. In chat-only environments, provide the actual generated images and available download links.
6. Verify the actual output count, readable concept, shared style, aspect ratio, and required text. If something is incomplete, identify the missing output rather than claiming the full batch is ready. Correct failed outputs only; keep successful ones.
7. Present the images together when the interface allows, labeled 1 through 4 with short descriptions. Let the user choose the image or elements to refine. Do not automatically spend more generations polishing a winner.

## Tool limits

If built-in image tools are missing, say that images cannot be generated in this environment. Offer four clearly labeled concept briefs that can be taken to an image-capable chat. These are not completed images. Do not silently use a paid API, obtain credentials, or switch to an external service.

If a reference cannot be read, say so and use an agreed text description or ask for it again. Do not claim to have inspected it. Saving a skill does not grant new tools or capabilities.

## Example request

“Use Image Inspiration to make four slide visuals explaining why a team needs shared context. Keep the style consistent and leave room for a slide title. Show all four before we refine anything.”

## Personalize this starter

Use the interview at https://batch-skill-inspiration.nate58678.chatgpt.site/agent-guide.md, or ask the user one question at a time about their use case, style, optional references, defaults, and preferred name. Summarize their choices for approval, then update this skill. Do not generate images during the setup interview.