Griptape Driver: OpenAI Compatible Image Generation

"OpenAI Compatible" in a node's name usually means one thing: it works with OpenAI's own API and with every service that cloned it. Griptape Driver: OpenAI Compatible Image Generation is that - an image-generation driver whose base_url you can repoint at OpenAI, a proxy, a hosted clone, or any provider that speaks the OpenAI image API dialect.

It's the flexible pick of the image drivers. If you're not sure which provider your project will use, or you're routing through a gateway, this node lets you keep the same graph and just change the URL. The tradeoff: because "compatible" means it has to work with the lowest common denominator, you don't get provider-specific extras.

How it works

Configuration node, standard pattern: the DRIVER output feeds a Griptape Run: Prompt Image Generation Task, which sends the prompt to whatever the base URL points at and returns the image.

The inputs that actually matter:

• image_generation_model - dall-e-2, dall-e-3, or gpt-image-1. The default is dall-e-3; pick the model your endpoint actually serves.
• size - 256, 512, 1024 square, plus 1024x1792 and 1792x1024. Note some models won't accept all of these; DALL-E 2, for instance, only does the three small ones.
• base_url - defaults to https://api.openai.com/v1. This is the field that makes the node "compatible": repoint it at your proxy or compatible provider and the same graph keeps working.
• api_key_env_var - env-var name for the key, OPENAI_API_KEY by default, not the literal key.

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: griptape[all], openai, python-dotenv, plus git-hosted extensions. And the pack-wide torch caveat: if Griptape's torch install mismatches ComfyUI's build, reinstall torch with the CUDA index as the README describes.

Gotchas

The compatibility that's a feature is also the trap. A service can call itself "OpenAI compatible" and still not implement every model or every size - a 404 or "model not found" on a valid-looking size means your endpoint doesn't serve it, not that the node is broken. Also check the auth model: some compatible providers want a different header shape or key, which the node won't magically fix. If you're pointing at OpenAI proper, this node is functionally the same as the plain OpenAI image driver - prefer that one for the default case and keep this for the heterogenous setups.