Gemini 3.5

Every few months Google drops a newer Gemini, and this pack tracks along. GeminiNode35 is the current flagship of the text nodes: a wrapper for gemini-3.5-flash on Vertex AI that returns plain text. If you've already met the 2.5 and 3.1 versions, the layout will feel like home - same inputs, same string output, newer model under the hood. The reason to reach for this one over its siblings is simple: it's the newest and best at prompt engineering, captioning, and anything that requires a model to actually think about what your workflow just made.

What it does

Prompt in, generated_output (STRING) out, with the same multimodal file-path inputs as the rest of the family: image_file_path, video_file_path, audio_file_path plus their MIME type dropdowns. Point it at a generated file and it can describe it, critique it, or turn it into a better prompt. One notable difference from the other Gemini text nodes: the model dropdown has exactly one option, GEMINI_35_FLASH. No Pro tier, no Lite tier, no choice - which is either simplicity or a constraint depending on how you look at it. For now, this is the model Google wants you using in this slot.

Inputs that matter

• prompt - the instruction. Default "Describe the content in detail." is a strong hint at intended use.
• system_instruction - optional persona/rules layer; the difference between a generic chat node and a reusable workflow tool lives here.
• temperature / top_p / top_k - sampling controls, defaults are fine.
• max_output_tokens (max 8192), candidate_count (up to 8) - length and alternatives.
• stop_sequences, response_mime_type - the latter is the sleeper: application/json gives you parseable structured output for downstream nodes.
• Four safety-threshold dropdowns, medium-and-above default.

Installing it

Shared pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/GoogleCloudPlatform/comfyui-google-genmedia-custom-nodes
pip install -r comfyui-google-genmedia-custom-nodes/requirements.txt

Or ComfyUI Manager → search comfyui-google-genmedia-custom-nodes → Install. Restart, look under Google AI/Gemini. Set gcp_project_id and gcp_region (default global), authenticate with gcloud auth application-default login, and remember every call is billed Vertex AI usage.

The honest take

It's the same dependable formula as the 2.5 and 3.1 nodes with a shinier model, and that's fine. If you're building a new workflow and want the best output for your prompt-enhancement stage, this is the one. If you're maintaining an existing graph, don't feel obligated to migrate - the older nodes still work and the interface hasn't changed. The one real caveat: a single-model dropdown means you can't drop to a cheaper tier to save money on high-volume calls, which the 3.1 node's three-way choice does give you. Budget-conscious tinkerers should keep that in mind.