Open-source image generation toolchain: ComfyUI and friends

For open-source image generation in 2026, treat ComfyUI as the workshop and hosted tools (Midjourney, Leonardo, etc.) as managed studios. Pick based on whether you need reproducible graphs or fastest pretty output.

Recommended local path

1. Install ComfyUI (see Image / open listings under Open-source / Self-host).
2. Start with one checkpoint + one upscaler. Resist downloading 40 LoRAs on day one.
3. Save a working workflow JSON before you experiment.
4. Add ControlNet / IP-Adapter only when you have a concrete composition problem.

Alternatives like Automatic1111-style UIs are fine for quick prompts; ComfyUI wins when you need repeatable pipelines.

Hosted vs local

Safety and licensing

• Model licenses are not interchangeable; some ban commercial use.
• Do not scrape living artists’ names as “styles” for public products without a policy.
• Keep NSFW and public demos on separate machines if that matches your risk tolerance.

Related

• Image
• Open-source / Self-host

Operational guide; model quality moves monthly.