Google Gemini, now wired directly into your ComfyUI graphs

What it does

This pack of ComfyUI custom nodes surfaces Google Gemini—Pro, Pro Vision, and 1.5 Pro—inside node-based Stable Diffusion workflows. It lets you generate text prompts from images, carry on multi-turn conversations with context, upload single files (images, audio, PDFs, TXT), and apply system instructions to steer outputs. Nodes come in two security modes: implicit API key (read from environment variables or config.json, safe to share) and explicit (key baked into the node, which the README sensibly warns against distributing).

The interesting bit

Gemini 1.5 Pro turns the graph into a multimodal reasoning pipeline: it accepts inputs up to 1,048,576 tokens and single files up to 20 GB, so your workflow can ingest audio or documents alongside images. The trade-off is a hard throttle of two requests per minute and 1,000 per day.

Key highlights

• Three model tiers: Gemini-pro, Gemini-pro-vision, and Gemini 1.5 Pro, with the latter handling text, image, and single-file inputs.
• Chat nodes keep conversation context; advanced variants accept system instructions.
• Two API key modes: hidden (share-safe) and exposed (node-level, for private use only).
• Pre-built workflows include a DALL·E 3-style pipeline and an all-in-one LoRA training preprocessor with auto-tagging.
• Helper nodes DisplayText_Zho and ConcatText_Zho handle string formatting inside the graph.

Caveats

• Single-file upload only; video and multi-file uploads are marked as not yet supported.
• Gemini 1.5 Pro rate limits are strict: two requests per minute and 1,000 per day.
• The README warns that local users may have connectivity issues to Google’s API and recommends Colab or Kaggle.

Verdict

Useful if you want an LLM co-pilot inside ComfyUI for prompt engineering or multimodal preprocessing. Avoid if you need high-throughput generation or fully offline operation.