Zhenzhen Gemini API

Most of the nodes in this pack lock you into one model with a dropdown. ComflyGeminiAPI is the opposite: the model field is a free-form text box, defaulting to gemini-2.0-flash-exp-image but happy to take whatever Gemini image model the relay currently carries. It's the "if the dedicated node for this model doesn't exist yet, use this one" Swiss Army knife - and for a relay that adds new Google models constantly (the changelog rotates through gemini-3-pro-image, gemini-3.1-flash-image, nano-banana variants, you name it), that flexibility is genuinely useful.

It's also the oldest and most barebones node in the family, so temper expectations: no async task queue, no multi-image editing like the dedicated S2A nodes. What it does, it does simply - one prompt, one model, sensible sampling controls.

What you'd actually touch

• prompt - the generation prompt.
• model - a text box. This is the point. Type a model name the relay supports and it calls it; if the name's wrong you'll get a clear error back.
• resolution - presets from 512×512 up to 2048×2048, plus three special entries: object_image size, subject_image size, and scene_image size. Those pair with the optional object_image, subject_image, and scene_image IMAGE inputs - the "composite a subject into this scene" pattern. Pick the matching size entry when you connect the corresponding image.
• num_images - 1 to 4 per call, each one billed.
• temperature and top_p - sampling knobs (0–2 and 0–1). Unlike most image API nodes these are exposed, which is nice when you want a model to get more or less creative. Defaults of 1.0 / 0.95 are fine.
• seed and timeout - seed for reproducibility; timeout (10–600s) for how long you'll wait before calling it dead.

Outputs: generated_images (IMAGE tensor), response (raw JSON), and image_url (the hosted image).

Install and key

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, or ComfyUI Manager → "Comfyui-zhenzhen". API key from ai.t8star.org into the api_key field.

When to reach for it, and when not to

Reach for it when the pack hasn't shipped a dedicated node for the Google model you want - you can often beat the changelog. Don't reach for it when you need reference-image editing or 4K stability; the dedicated Comfly_gemini_3_1_flash_image_edit_S2A node does that properly with async task handling. Same shared caveats as always: premium token groups stabilize the Gemini image models (if you get failures at higher resolutions, that's your key group, not the node), 500s are usually upstream and a re-run clears them, and this is a metered paid API - the credit meter runs per image, not per conversation.