🖼️ LLMs_Vison_Gemini

You've found a workflow referencing 🖼️ LLMs_Vison_Gemini and ComfyUI is telling you the node is missing. Here's the thing: it's not missing, it's gone. This was the pack's dedicated Google Gemini vision node, and the ComfyUI-LLMs 1.0.0 refactor removed it along with the other per-provider vision nodes, folding them all into one LLMs Vision Unified. Gemini support is still there - it's just one dropdown value now.

What it was

The old node was the Gemini half of a three-way vision split (Ali, Gemini, GLM4) that predates the unified node. Its inputs tell you a lot about how vision nodes looked in early 2024: prompt, model_name (gemini-pro-vision), a stream toggle, an api_key field right on the node, and an optional image input. Output was a STRING named text. Mechanically it called Google's google-generativeai SDK - model.generate_content([prompt, image]) - which is exactly what the unified node still does under the hood.

Two things about that old design are worth remembering, because they age badly:

• The api_key widget was a trap. A key typed into the node lives in the workflow JSON, which gets embedded in the PNG metadata and shared on Civitai and Discord. That's how people leak Gemini keys in the wild. Today the key lives in settings.yaml, not the graph - strictly better.
• gemini-pro-vision is ancient. Google has moved on multiple model generations since this pack shipped. The old default is now a footnote; when you use the replacement, pick the current vision model from your config.

The replacement

Install the pack and use LLMs Vision Unified instead: model_type = gemini, model = your Gemini vision model, image and prompt wired in. You'll get the same STRING output, minus the stream toggle and the on-node key.

cd ComfyUI/custom_nodes
git clone https://github.com/leoleelxh/ComfyUI-LLMs
cd ComfyUI-LLMs
pip install -r requirements.txt
cp settings.yaml.sample settings.yaml

Then in settings.yaml → vision_models → gemini, drop in your key (makersuite.google.com / AI Studio). The README's clone URL is misspelled as leoleexh - use the one above or ComfyUI Manager ("ComfyUI-LLMs"), and run python validate_config.py to confirm the provider loads.

Gotchas

• Placeholder key. The sample ships AIxxxxxx for Gemini - that's a marker, not a key.
• Proxy, depending on where you are. The README's proxy warning is most relevant for the Chinese APIs, but Gemini is Google, so the reverse applies if you're behind a firewall that likes to block it.
• Old workflows still do the key-widget thing. If a friend's workflow asks you to paste a Gemini key into a node widget, that's the pre-refactor pattern - and it's a smell. Move it to settings.yaml instead of continuing the leak.