built a CLI-based image generation agent that takes vague prompts like:
“a warrior in a forest”
And automatically:
Expands it into a detailed cinematic description (lighting, mood, camera angle, art style)
Routes it to the appropriate model via OpenRouter
Downloads the generated image locally
Tech stack: React Ink (CLI UI) TypeScript Modular subagent architecture OpenRouter (free-tier models supported) One thing I realized while building this: You don’t actually need expensive API credits to experiment with AI agents.
There’s a surprisingly strong free stack available: OpenRouter (multiple free-tier models) Antigravity (Claude, GPT variants, Gemini access) OpenCode (GLM-5, MiniMax) Local models via LM Studio / Ollama / ComfyUI
The biggest barrier isn’t cost — it’s clarity of architecture.