ComfyUI Better Gemini

Custom ComfyUI node(s) for generating images with Google Gemini via the official google-genai Python SDK.

Install

1. Clone into your ComfyUI custom_nodes directory:
   • cd /path/to/ComfyUI/custom_nodes
   • git clone https://github.com/<you>/ComfyUI-Better-Gemini.git
2. Install Python deps (network required):
   • pip install -r ComfyUI-Better-Gemini/requirements.txt
3. Set an API key (recommended):
   • export GOOGLE_API_KEY="..." (or GEMINI_API_KEY)
4. Restart ComfyUI.

Nodes

• Better Gemini (image)
  • Inputs: prompt, model (dropdown; populated via models.list()), prompt_images (optional), aspect ratio, resolution / width+height, temperature, top_p/top_k, max tokens, thinking difficulty, seed.
  • Outputs: IMAGE, STRING (any returned text / notes).

Example Workflow

Notes

• The node imports google-genai lazily so ComfyUI can still boot even if dependencies aren’t installed yet; execution will raise a clear error until installed.
• This extension uses ComfyUI’s V3 extension loader (comfy_entrypoint).
• The model dropdown is populated via client.models.list() (filtered to models supporting generateContent). It requires an API key via GOOGLE_API_KEY/GEMINI_API_KEY; otherwise it falls back to a single default model and logs a warning.
• Gemini requires seed to fit in an int32; larger ComfyUI seeds are deterministically folded via seed % 2**31.
• resolution/aspect_ratio are best-effort (model-dependent). For models that support it (e.g. gemini-3-pro-image-preview), the node sends generationConfig.imageConfig with imageSize/aspectRatio; it logs a warning if the returned size doesn’t match (no auto-resize).
• If Gemini returns no images (blocked or a text-only model), the node returns a blank placeholder image and includes a note in the STRING output (and logs a warning).

Dev

• Run unit tests: python -m unittest discover -s tests -p 'test_*.py' -v