b1rdmania Art Skill — Image Generation
A CLI tool for generating images from text prompts using multiple AI models. Used by the brand-skill in Phase 2 (Visual Direction) and Phase 3 (Mark Development) for reference image generation.
The Tool
Tools/Generate.ts — A multi-model image generation CLI that supports:
| Model | Flag | API Key | Best for |
|---|---|---|---|
| Gemini (nano-banana-pro) | --model nano-banana-pro |
GOOGLE_API_KEY |
Best quality, text rendering |
| Replicate (nano-banana) | --model nano-banana |
REPLICATE_API_TOKEN |
Faster iteration |
| Flux | --model flux |
REPLICATE_API_TOKEN |
High quality, stylistic variety |
| GPT Image | --model gpt-image-1 |
OPENAI_API_KEY |
Alternative generation |
Setup
- Install bun runtime:
curl -fsSL https://bun.sh/install | bash - Set at least one API key as an environment variable
- Run with:
bun run Tools/Generate.ts --model [MODEL] --prompt "[PROMPT]" --output ~/Downloads/output.png
Usage
# Basic generation
bun run Tools/Generate.ts \
--model nano-banana-pro \
--prompt "Abstract minimalist logo concept: [description]. Clean vector style, dark background. No text." \
--size 2K \
--aspect-ratio 1:1 \
--output ~/Downloads/brand-ref-1.png
# With background removal (useful for marks)
bun run Tools/Generate.ts \
--model nano-banana-pro \
--prompt "[PROMPT]" \
--size 2K \
--remove-bg \
--output ~/Downloads/mark-reference.png
# With reference image for style consistency
bun run Tools/Generate.ts \
--model nano-banana-pro \
--prompt "[PROMPT]" \
--reference-image existing-mark.png \
--size 2K \
--output ~/Downloads/variation.png
Flags
| Flag | Options | Default | Purpose |
|---|---|---|---|
--model |
nano-banana-pro, nano-banana, flux, gpt-image-1 | nano-banana-pro | Image generation model |
--prompt |
text | (required) | The generation prompt |
--size |
1K, 2K, 4K (or aspect like 16:9) | 16:9 | Output resolution or aspect ratio |
--aspect-ratio |
1:1, 16:9, 9:16, 3:2, 21:9 | 16:9 | Aspect ratio (Gemini nano-banana-pro) |
--output |
path | ~/Downloads/generated-image.png | Output file path |
--remove-bg |
(flag) | off | Remove background (requires REMOVEBG_API_KEY) |
--thumbnail |
(flag) | off | Generate both transparent + background versions (enables --remove-bg) |
--reference-image |
path (repeatable) | none | Style/content reference (nano-banana-pro only) |
--creative-variations |
number | 1 | Generate multiple variations (-v1, -v2, etc.) |
In the Brand Process
Phase 2 — Visual Direction (Mode A):
Generate 3-4 reference images exploring different visual interpretations of the brand concept. Use --aspect-ratio 1:1 for logo references (default is 16:9). High resolution, abstract/minimalist prompts.
Phase 3 — Mark Development (Path A):
Generate high-contrast reference images for bitmap tracing with vtracer. Use --remove-bg for clean marks. Trace the result to SVG.
Tips
- Always output to
~/Downloads/first for preview - Square (1:1) aspect ratio works best for logo references
- Include "no text" in prompts for logo/mark references
- Use
--remove-bgwhen generating marks for tracing - Keep prompts abstract: "minimalist logo concept" not "a logo for my company"
nano-banana-prohandles text rendering best if you need labels