// Use when creating or editing raster images, visual concepts, product graphics, launch assets, diagrams, posters, thumbnails, reference sheets, UI mock images, or image prompts that need source-grounded claims, readable text, brand direction, and post-generation verification.
Use when creating Product Hunt launch carousel images for developer tools, AI tools, open-source projects, GitHub launches, or technical products. Research comparable launches, inspect the project, extract verified proof points, then generate 4-5 launch-ready images with consistent branding, product surfaces, readable copy, and source-grounded claims.
Use when generating a single dense technical architecture, reference poster, or system diagram image for a software project. Produces a source-grounded bitmap diagram with topology, primitives, ingest/process flow, core logic, state scopes, tiers/modules, surfaces, runtime cards, and strict semantic color rules.
| name | image-generation |
| description | Use when creating or editing raster images, visual concepts, product graphics, launch assets, diagrams, posters, thumbnails, reference sheets, UI mock images, or image prompts that need source-grounded claims, readable text, brand direction, and post-generation verification. |
Use this skill for image work where the output should be a generated or edited bitmap image. This is the broad image workflow; use more specific skills when the request clearly asks for Product Hunt carousel images or dense technical diagram images.
producthunt-launch-images for multi-image launch carousels.technical-diagram-image for one dense architecture/reference diagram image.Capture:
If the image includes factual claims, inspect sources before prompting.
Use this when the image refers to a product, repo, company, event, person, or technical system:
Subject:
Audience:
Use case:
Required dimensions/aspect:
Brand colors:
Allowed logo/visual identity:
Visible headline:
Visible subtext:
Verified facts:
Commands/URLs/ports:
Metrics/counts/dates:
Source links/files:
Unknowns to omit:
Negative constraints:
Never invent metrics, dates, commands, screenshots, logos, claims, or product UI. If a fact is not verified, omit it or label it as conceptual.
| Type | Best structure |
|---|---|
| Product graphic | headline, product surface, proof chips, command strip |
| Social card | one hook, one visual metaphor, one verified fact |
| Thumbnail | large readable phrase, one focal object, high contrast |
| Hero image | realistic subject/product, atmospheric background, no text-heavy layout |
| Technical visual | topology, labels, arrows, source-grounded nodes |
| Poster | compact header, sections, tables, diagrams, legend |
| Edit | preserve key subject, list exact changes, list what must remain unchanged |
A strong prompt includes:
Template:
Create <number/type> image for <subject>.
Use case: <launch/docs/social/reference/thumbnail/etc>.
Audience: <who should understand it>.
Format: <aspect ratio/dimensions if known>.
Style: <specific visual style, product surface, screenshot, poster, editorial, technical, etc>.
Composition: <foreground/midground/background, panels, cards, grid, table, diagram, focal object>.
Visible text: "<exact text>".
Facts to include: <verified facts only>.
Brand/color rules: <colors and semantic meanings>.
Negative constraints: no fake metrics, no malformed commands, no stock art, no unwanted logos, no tiny unreadable text, no unsupported claims.
Verification priority: <what must be correct>.
After generation, inspect the output for:
Regenerate only the failed image when possible. Make the next prompt shorter, more explicit, and more constrained around the failure.
For image edits:
Edit prompt template:
Edit the provided image.
Change: <specific changes>.
Preserve: <faces/products/text/layout/colors/background/items>.
Style match: <lighting, texture, camera, illustration style>.
Avoid: <unwanted changes>.
Image models can mangle text. Keep visible text:
For text-heavy needs, prefer fewer words and more structure. If a dense table is required, consider a diagram-specific skill and verify carefully.
Use semantic color when the image explains a system:
State the semantic rule in the prompt when colors carry meaning.
The image should:
For a single image, return the absolute path:
/absolute/path/to/image.png
For multiple images, return ordered paths:
1. /absolute/path/01-name.png
2. /absolute/path/02-name.png
3. /absolute/path/03-name.png