SD Generator CLI: No More Cargo Cult with Stable Diffusion

I’m open-sourcing SD Generator CLI today, a tool to stop fighting with Stable Diffusion workflows.

The Problem

Everyone starts SD the same way:

• Copy-paste “magic prompts” found on CivitAI
• Tweak seeds randomly
• Lose the parameters that produced our best results

Cargo cult engineering - and it doesn’t scale.

The Solution

I built a CLI that transforms this into a reproducible process:

• YAML templates with inheritance and imports
• Combinatorial or random generation for systematic testing
• Manifests to track what produced what
• Automatic metadata on generated images

As a bonus, there’s a small local Web UI to quickly browse your run results.

Technical Specs

• Python 3.10+
• Strict mypy
• Pytest tests
• Poetry monorepo
• MIT licensed

It’s a side project, 115 commits, it won’t save the world. But if it helps a few people spend less time finding their settings and more time generating, that’s cool.

Repo: github.com/oinant/local-sd-generator

If you’re struggling with your SD workflows, check it out. If you find it useful, a star is appreciated. PRs are welcome.