Griptape Driver: Amazon Bedrock Titan

Amazon's Titan image model isn't the model anyone hypes, and that's kind of the point of this node. Griptape Driver: Amazon Bedrock Titan points your Griptape image task at Amazon's own image generator - the option you pick when you're already on AWS and want image generation inside your existing Bedrock setup rather than bolting on another vendor.

It's a "good enough, boring, and billed to your AWS account" choice. Don't expect it to out-draw the community's local checkpoints on realism or style; expect it to work, be accessible through the same credentials you already have, and not eat your VRAM.

How it works

Same driver pattern as the rest of the pack: you configure it here, and the DRIVER output feeds a Griptape Run: Prompt Image Generation Task node, which sends your prompt to the API and returns the image. This node itself never renders anything.

What you set:

• image_generation_model - defaults to amazon.titan-image-generator-v1; it's a freeform string, so you can point at a newer Titan image model if Amazon ships one and you have access.
• size - a two-choice dropdown: 512x512 or 1024x1024. That's all Titan's image endpoint offers here; don't go hunting for a portrait mode, it doesn't exist.
• seed - for reproducible runs.

Then the AWS trio: aws_access_key_id_env_var, aws_secret_access_key_env_var, and aws_default_region_env_var. As with the Stable Diffusion Bedrock driver, these take environment-variable names, not the keys themselves - the pack reads the vars at runtime. Set them in your shell before launching ComfyUI, or in Settings → Griptape.

Installing

Ships in the ComfyUI Griptape Nodes pack:

• ComfyUI Manager: search "Griptape" → install ComfyUI-Griptape.
• Manual: cd ComfyUI/custom_nodes && git clone https://github.com/griptape-ai/ComfyUI-Griptape, then restart.

Pack dependencies are griptape[all], openai, python-dotenv, plus git-hosted extensions. The torch quirk applies pack-wide: Griptape installs torch and can fight ComfyUI's version - the README's fix is pip uninstall torch then reinstall with --extra-index-url https://download.pytorch.org/whl/cu121.

Gotchas

The two big AWS-side failures: the model isn't granted in your region (check Bedrock → Model access, request it, wait for approval), or the region env var is missing - Titan image generation is not available in every Bedrock region, so pick one that lists it. Also keep expectations honest: 1024x1024 max, square-only, and Amazon's aesthetic. If your brief calls for a specific art style or aspect ratio, Titan is the wrong tool - that's what the style presets on the Bedrock Stable Diffusion driver are for.