Imagen Image🧸

Imagen Image calls Google's Imagen 3 text-to-image model and returns the result into ComfyUI. Imagen 3 is Google's flagship image model, and it's very good at photorealism and clean text. Unlike the Gemini Image node in this same pack, this one is single-purpose: text in, image out. No multimodal input, no text output.

The catch that separates this from every other node in Plush is the key. Imagen runs on Google Cloud's Vertex AI, and getting access is a proper setup chore, not a one-line thing.

How it works

You write a prompt, pick an aspect ratio and a count, and the node hits Vertex AI's Imagen endpoint. The image comes back as a standard ComfyUI IMAGE. That's the whole loop. It's a thin, clean wrapper around a paid Google endpoint.

The inputs that matter

• prompt (multiline): your text. Imagen handles descriptive, natural-language prompts well.
• aspect_ratio: 1:1, 3:4, 4:3, 9:16, or 16:9. This is how you set orientation; there's no pixel-size field.
• number_of_images (1 to 8): how many to generate per run. Straight cost multiplier.
• Image_model: the Imagen model to use (imagen-3.0-generate-002).
• Custom_ApiKey (optional): because Vertex uses a custom setup, you'll typically store your key in a custom-named environment variable (say VERTEX_API_KEY) and read it in through the Custom API Key node.

Outputs are image, plus help and troubleshooting.

The setup wall

This is the node that needs the most work to turn on, so know it going in. To use Imagen you need a Google Cloud account with an active project, a payment method attached, Vertex AI enabled on that project, and the Imagen model activated. Then you generate a Vertex/Google Cloud key, stash it in an environment variable, and pull it into the node via Custom API Key. None of this is Plush's doing; it's Google Cloud's onboarding. If you already have a GCP project this is twenty minutes. If you don't, budget an afternoon.

Installing it

Install Plush-for-ComfyUI via ComfyUI Manager (search "Plush", install Plush-for-ComfyUI, restart), or manually with cd ComfyUI/custom_nodes && git clone https://github.com/glibsonoran/Plush-for-ComfyUI.git, then pip install -r requirements.txt and restart.

Common issues

Almost every problem here traces back to the Google Cloud side. The key isn't authorized because Vertex or the Imagen model wasn't activated on the project, or billing isn't attached, or the environment variable holding the key isn't visible yet (env vars usually need a machine reboot). Like the other Google models in this pack, Imagen also enforces safety filters that can reject prompts, so a benign prompt that keeps failing may be tripping a policy check. The troubleshooting output is your friend for all of these; wire a text display to it and it'll tell you which wall you hit. And set a spending limit on the Cloud project so a runaway batch can't surprise you on the bill.