Gemini Model Selector

Most Gemini nodes in this pack carry their own model dropdown, which is convenient until you have six text nodes and want to switch them all from gemini-2.5-flash to a new preview in one move. That's the job of the Gemini Model Selector: it's a single node holding a dropdown of 55+ model IDs - text, image, Imagen, TTS, Veo, Lyria, embeddings - that outputs a plain string you can wire into any of the pack's custom_model inputs. Change it once, and every node fed by it follows.

But the selector earns its keep with the second input, which is genuinely clever: category_validate. Set it to text and the node will throw an error at runtime if the selected model isn't a text model. Pick veo and it refuses to pass a Lyria music model through. It's a guardrail against the dumbest failure in this whole category - wiring an imagen-... ID into a node that's about to call generate_content and watching it 404.

How it works

The mechanism is almost insultingly simple: it's a value pass-through. The dropdown maps to a fixed list of model IDs (the same verified list the whole pack shares), and the node returns the selected string as its model output. The interesting part is the validation - when category_validate isn't any, it compares the model's category against what you requested and raises a clear error if they don't match. So you get the shared dropdown ergonomics and a runtime check that's effectively free.

Inputs and outputs

• model - the big dropdown. 55 choices spanning every service in the Gemini module, from gemini-3-pro-preview and gemini-2.5-flash down through the specialized IDs like the robotics and computer-use previews. Defaults to gemini-pro-latest, the safe alias.
• category_validate - any, text, image_gen, imagen, tts, veo, lyria, or embedding. Set this to the kind of node it feeds. any disables the check.
• custom_model - the escape hatch. Type any model ID and it takes priority over the dropdown, so you're not stuck if Google ships something new before the pack updates.

One output: model (STRING). Feed it into another node's custom_model input - that's the intended wiring.

How to install it

It ships in ComfyUI-API-Toolkit, so:

cd ComfyUI/custom_nodes
git clone https://github.com/IxMxAMAR/ComfyUI-API-Toolkit
cd ComfyUI-API-Toolkit
pip install -r requirements.txt

Or install "API Toolkit" from ComfyUI Manager. The Gemini service needs google-genai>=0.8.0. No key needed for this node itself (it doesn't call anything) - the downstream node does the calling.

Common issues

The main trap is expecting it to do more than it does. It's a pass-through with a validator, not a router that can auto-pick models or know what your account can access. If the dropdown contains a model your key can't reach, the error will come from the node doing the actual call, not here - run the pack's Gemini List Available Models node to see what's real for your account, then use custom_model for anything not in the list.

Also worth knowing: because category_validate checks at execution time, a mismatch won't red-flag until you hit Queue. That's by design - it's a runtime guard, not a canvas-time linter - but it means your first run of a mis-wired workflow will error rather than silently produce garbage, which is honestly the better failure mode.