agentleFS
sanky369sanky369/vibe-building-skills/skills/creative/image-generation/SKILL.md

image-generation

Skill30 starsChanged 5 months agoMITView on GitHub

Turns any request for an AI-generated image into precise, generation-ready prompt specs and — when the FAL.ai automation is configured — the generated files themselves, using the single model fal-ai/nano-banana-pro. Use whenever the user asks to 'generate an image', 'make a picture of X', 'create a thumbnail / hero image / illustration / infographic', 'write me a prompt for this image', or any visual that isn't specifically product photography, a social-platform graphic, or a brand mark (hand those to the sibling skills). Produces one prompt-spec block per asset (subject, style, composition, lighting, mood, aspect ratio, negative constraints) plus file-naming and variant conventions, and runs docs/creative_cli.py to generate when a FAL_API_KEY is available.

  • Reads credentials

63 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use gift-evaluator, Podcast Generate and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • dav-niu474Skill

    gift-evaluator

    The PRIMARY tool for Spring Festival gift analysis and social interaction generation. Use this skill when users upload photos of gifts (alcohol, tea, supplements, etc.) to inquire about their value, authenticity, or how to respond socially. Integrates visual perception, market valuation, and HTML card generation.

    Also in 5 projects that use this

  • dav-niu474Skill

    Podcast Generate

    Generate podcast episodes from user-provided content or by searching the web for specified topics. If user uploads a text file/article, creates a dual-host dialogue podcast (or single-host upon request). If no content is provided, searches the web for information about the user-specified topic and generates a podcast. Duration scales with content size (3-20 minutes, ~240 chars/min). Uses z-ai-web-dev-sdk for LLM script generation and TTS audio synthesis. Outputs both a podcast script (Markdown) and a complete audio file (WAV).

    Also in 5 projects that use this

  • huang-x-hSkill

    Video Generation

    Implement AI-powered video generation capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to generate videos from text prompts or images, create video content programmatically, or build applications that produce video outputs. Supports asynchronous task management with status polling and result retrieval.

    Also in 6 projects that use this

  • huang-x-hSkill

    VLM

    Implement vision-based AI chat capabilities using the z-ai-web-dev-sdk. Use this skill when the user needs to analyze images, describe visual content, or create applications that combine image understanding with conversational AI. Supports image URLs and base64 encoded images for multimodal interactions.

    Also in 6 projects that use this

  • +1 more from huang-x-h
  • +7 more from dav-niu474

The file

We split this skill into the rules it gives an agent

11 rules. 0 are common, 6 are distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 54 edited versions we read keep these.

  • 6

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use for general-purpose images: hero images, illustrations, thumbnails, infographics, textures, concept art, any one-off visual.

    line 18
  2. Distinctive:

    Ask in one batch, only what's missing:

    line 31
  3. Very specific:

    Check for a style block: If the user has run skills/creative/creative-strategist, reuse its style block verbatim in every prompt.

    line 46
  4. Distinctive:

    Draft the prompt with the 5-part formula: Every prompt contains, in order:

    line 50
  5. Distinctive:

    Add negative constraints as plain prohibitions ("no text overlay", "no watermark", "no busy background").

    line 56
  6. Very specific:

    Decide variant strategy: If it's a throwaway draft → 1 image at 1K.

    line 58
  7. Very specific:

    Generate or deliver: If FAL_API_KEY (or FAL_KEY) is set:

    line 63
  8. Very specific:

    Use --web-search only when the image must reflect current real-world data (e.g.

    line 72
  9. Very specific:

    Read references/automation.md before writing any generation command — it has the full, verified parameter set, CLI flags, Python helpers, …

    line 81
  10. Distinctive:

    If images were generated, append the saved paths under Results and say which variant you recommend and why (one line).

    line 105
  11. Distinctive:

    Before delivering, verify every spec:

    line 110

“Common” is measured: kept by at least 6 in 10 of the 54 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Image Generation

Convert a fuzzy image request into prompt specs precise enough that any competent image model — and specifically fal-ai/nano-banana-pro on FAL.ai, the one model this repo's automation uses — renders what the user actually pictured. The prompt spec is the deliverable; generated files are the bonus when the automation is configured. Prime rule: never generate from a vague prompt. A weak prompt costs a generation credit and teaches you nothing; a sharp spec is reusable forever.

When to use / when not to

Use for general-purpose images: hero images, illustrations, thumbnails, infographics, textures, concept art, any one-off visual. Hand off instead when the request is really:

  • Product shots for commerce/marketing → skills/creative/product-photography
  • Platform-sized social content → skills/creative/social-graphics
  • Logos, icons, patterns, brand marks → skills/creative/brand-asset
  • No visual direction exists yet and consistency matters → run skills/creative/creative-strategist first and consume its style block
  • A whole multi-asset campaign → skills/creative/orchestrator routes it

Intake

Ask in one batch, only what's missing:

  1. Subject — what exactly is in the image? (object, scene, people, text?)
  2. Use — where does it go? (this decides aspect ratio and resolution)
  3. Style anchor — existing style guide / creative-strategist style block, reference images, or 3 adjectives for the mood?
  4. Automation — is FAL_API_KEY set so you can generate, or deliver specs only?

Infer, don't ask: aspect ratio from the stated destination (thumbnail → 16:9, story → 9:16, feed → 1:1 or 4:5); resolution from stakes (2K default, 4K hero/print, 1K drafts). Don't stall: if subject and use are clear, state your assumptions for the rest and proceed.

Workflow

  1. Check for a style block. If the user has run skills/creative/creative-strategist, reuse its style block verbatim in every prompt. If not, and this is a one-off, derive a minimal one (style + palette + mood) from context and label it an assumption.
  2. Draft the prompt with the 5-part formula. Every prompt contains, in order: subject (the main thing, specific: "a luxury leather watch with gold accents", never "a watch") → descriptive detail (materials, colors, text to render) → style ("professional product photography", "flat modern illustration", "photorealistic") → technical treatment (lighting, composition, focus, "4K, highly detailed") → mood ("warm and inviting"). Add negative constraints as plain prohibitions ("no text overlay", "no watermark", "no busy background").
  3. Decide variant strategy. If it's a throwaway draft → 1 image at 1K. If it's a keeper asset → propose 3–4 genuinely distinct prompt specs (different composition or style angle, not synonym swaps), generate 1 of each or num_images 3–4 of the winner, and have the user pick. Never present one option for a hero asset.
  4. Generate or deliver. If FAL_API_KEY (or FAL_KEY) is set:

    python docs/creative_cli.py custom \
      --category "<category>" --name "<asset-name>" \
      --prompt "<full prompt>" \
      --aspect-ratio 16:9 --resolution 2K --num-images 3

    Use --web-search only when the image must reflect current real-world data (e.g. "2026 trends infographic"). If no key is set, deliver the prompt-spec blocks and the exact command the user can run later — do not fake results or claim images were generated.

  5. Review and iterate. Compare output to the spec's mood/style/negative constraints. Fix misses by editing the prompt (more specific subject, explicit lighting, stronger prohibitions) — there are no other model knobs. One iteration round maximum before checking in with the user.

Read references/automation.md before writing any generation command — it has the full, verified parameter set, CLI flags, Python helpers, and output layout.

Required output format

Deliver one block per asset (this exact structure, whether or not you generate):

## Prompt Spec — [asset name]
- **Use / destination:** [where it will live]
- **Subject:** [the main thing, concretely]
- **Style:** [artistic treatment]
- **Composition:** [framing — centered / rule of thirds / negative space / overhead]
- **Lighting:** [studio / natural golden hour / dramatic rim / high-key]
- **Mood:** [2–3 adjectives]
- **Aspect ratio:** [one of 21:9 16:9 3:2 4:3 5:4 1:1 4:5 3:4 2:3 9:16] · **Resolution:** [1K/2K/4K] · **Format:** [png/jpeg/webp]
- **Negative constraints:** [what must NOT appear]
- **Prompt (final, paste-ready):**
  > [single flowing prompt assembled from the fields above]
- **Variants:** [n] — [what varies between them]
- **File naming:** assets/<category>/<asset-name>/<asset_name>_<n>_<timestamp>.png (automation default)
- **Generate with:** `python docs/creative_cli.py custom --category ... --name ... --prompt "..." --aspect-ratio ... --resolution ... --num-images n`

If images were generated, append the saved paths under Results and say which variant you recommend and why (one line).

Quality bar

Before delivering, verify every spec:

  • Subject is concrete enough that two strangers would picture the same image
  • Style, lighting, composition, and mood are each explicitly stated — none left to model default
  • Aspect ratio matches the stated destination (no 1:1 "because default")
  • Negative constraints listed (at minimum: no watermark; state text policy explicitly — models mangle long text)
  • No invented parameters — only the ones in references/automation.md; the model is always fal-ai/nano-banana-pro
  • Keeper assets got 3+ distinct variants, not one take
  • If automation wasn't run, the deliverable says so and includes runnable commands

Hard don'ts: never claim an image was generated when it wasn't; never put brand color hex codes in quotes and hope — name the colors in words too ("deep navy #004E89"); never render paragraphs of text inside an image.

Integration

  • skills/creative/creative-strategist → feeds this skill its style block (paste into every prompt's style/mood/lighting fields).
  • skills/creative/product-photography, skills/creative/social-graphics, skills/creative/brand-asset → specialized front-ends that produce their own prompt specs and use the same automation; route there when the request fits.
  • skills/creative/product-video and skills/creative/remotion-script-writer → consume generated stills as source frames / assets.
  • references/automation.md — full FAL.ai client, CLI, and Python-helper reference. Read it whenever you're about to run a generation command.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.