Google AI - Image Generator (Imagen 4)

GoogleAI_ImageNode turns a text prompt into a finished image using Google's hosted Imagen 4 models - no local checkpoint, no VRAM, no controlnet setup, just a prompt and an API key. It outputs a proper ComfyUI IMAGE tensor, so everything downstream (upscalers, VAE decoding for inspection, image save nodes) just works. For someone who wants a cloud generator as one step in a ComfyUI pipeline rather than a separate app, this is the least-friction route this pack offers.

What it's for

Two situations. First, you want Google-quality text-to-image in your graph without installing a huge model - Imagen 4 is genuinely good at following long prompts and rendering text in images, which is where local models often stumble. Second, you want a fallback or comparison generator: prompt the local model and Imagen side by side, and pick the result you like before committing a batch. It's a paid API call per image, so think of it as the premium step in a workflow, not the default one.

How it works

Pure REST to Google's :generateImages endpoint - no SDKs. Your prompt, aspect ratio, seed, and (if provided) negative prompt go up; the first returned image comes back decoded into a [1, H, W, C] tensor. The seed and randomize_seed inputs behave exactly like you'd expect from local generation: fix the seed for reproducible images, or let it roll. Key resolution follows the pack standard (node field → GEMINI_API_KEY env var → .env). And true to this pack's design, API errors don't crash the workflow - they come back as a placeholder error image you'll see on the canvas.

Inputs and outputs that matter

• prompt (STRING, multiline, default "A beautiful cinematic portrait") - the actual ask. Imagen rewards detailed, structured prompts.
• model (COMBO, default imagen-4.0-generate-001) - pick between the standard, ultra, and fast variants in the combo.
• aspect_ratio (COMBO, default 1:1) - the usual set: 1:1, 16:9, 9:16, 4:3, 3:4.
• seed (INT) and randomize_seed (BOOLEAN) - reproducibility dials.
• api_key, negative_prompt (STRING) - key override and an optional negative prompt (Imagen's support is thinner than diffusion's, so keep it simple or leave it empty).
• image (IMAGE) - the generated image, ready to feed into a save node, upscaler, or img2img pass.

Installing it

One of the 15 Google AI nodes in COMFYUI_PROMPTMODELS (PromptModels Studio in Manager):

cd ComfyUI/custom_nodes
git clone https://github.com/cdanielp/COMFYUI_PROMPTMODELS

Put GEMINI_API_KEY=AI... in the pack's .env and restart. Needs ComfyUI 0.26.0+. No model downloads.

Common issues

Money and keys, in that order. Each image is a billed API call, and "preview" model IDs eventually get retired, so if a previously working combo starts returning errors, check whether the model ID changed. Safety filters are real on Google's side: heavily restricted subject matter gets blocked and you'll see an error image, not a generation - that's the API being conservative, not the node being broken. And keep the key in .env; in the node field it ships along in workflow JSON and PNG metadata.