Griptape Create: Image Variation

Griptape Create: Image Variation takes an image you've got, pairs it with a text prompt, and generates a variation of that image through Griptape's image-generation drivers. It's the image-to-image sibling of Create Image from Text - same driver system, same temp-file output, but it starts from an actual picture instead of empty canvas. If your workflow describes an image with one node and then wants a re-draw of it, this is the stage that does the re-draw.

What it's for

The honest use case: you have an image flowing through your graph (from an image query/description node, from another task, from a Load Image) and you want a reimagined version driven by a text direction. It's img2img-flavored, done through an LLM provider's variation API rather than a local diffusion sampler. That means it's an API call, with all the quality caveats that come with letting a driver handle it - you're not dialing in denoising strength or seeds here.

How it works

It runs a VariationImageGenerationTask. The input image is converted to base64, sent along with your prompt to the driver, and the resulting file is loaded back as a ComfyUI IMAGE tensor plus a file path. No driver attached? It falls back to OpenAI dall-e-2, which needs OPENAI_API_KEY.

The driver model matters a lot here. If you're using a Black Forest / Flux driver, the node checks the model against an explicit allowlist (flux-pro-1.0-canny, flux-pro-1.0-depth, flux-pro-1.1, flux-pro-1.1-ultra, etc.) and raises a clear error if the model isn't variation-capable. The README also flags the whole variation-task family as "In Beta" - expect rough edges.

Inputs

• STRING (required, multiline) - the text direction for the variation.
• image (IMAGE, required) - the source image. Wire it from Load Image or any node producing an IMAGE.
• input_string (STRING) - appended to STRING with a newline, for dynamic text.
• key_value_replacement (DICT) - {{ key }} placeholder substitution.
• driver (DRIVER) - pick your provider/model. Defaults to OpenAI dall-e-2 without one.

Outputs

• IMAGE - the variation, as a ComfyUI tensor.
• FILE_PATH (STRING) - where the generated file was written (temp dir).

Installation

Part of ComfyUI Griptape Nodes:

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

Same shared pack install and driver/API-key requirements as the other image task nodes.

Gotchas

• Beta node. The README literally brackets variation generation as in beta. Test with one cheap call before building a workflow on it.
• Model compatibility is the biggest trap: dall-e-2 works, some drivers flat-out reject variation, and the Black Forest models are restricted to the allowlist. Read the driver you're using.
• Paid API calls again - each variation costs money. No local rendering here.

For a graph that needs "describe this, then re-imagine it," it's the missing link - just keep your driver expectations realistic.