comfyui-tooling-nodes

ComfyUI exchanges images via the filesystem. This requires a multi-step process (upload images, prompt, download images), which invites a whole class of potential issues you might not want to deal with. It's also unclear at which point those images will get cleaned up if ComfyUI is used via external tools.

Load Image (Base64)

Loads an image from a PNG embedded into the prompt as base64 string.

• Inputs: base64 encoded binary data of a PNG image
• Outputs: image (RGB) and mask (alpha) if present

Load Mask (Base64)

Loads a mask (single channel) from a PNG embedded into the prompt as base64 string.

• Inputs: base64 encoded binary data of a PNG image
• Outputs: the first channel of the image as mask

Send Image (WebSocket)

Sends an output image over the client WebSocket connection as PNG binary data.

• Inputs: the image (RGB or RGBA), supports batches

This will first send one binary message for each image in the batch via WebSocket:

12<PNG-data>

That is two 32-bit integers (big endian) with values 1 and 2 followed by the PNG binary data. There is also a JSON message afterwards:

{'type': 'executed', 'data': {'node': '<node ID>', 'output': {'images': [{'source': 'websocket', 'content-type': 'image/png', 'type': 'output'}, ...]}, 'prompt_id': '<prompt ID>}}

Load Image from Cache

Loads an image or mask that has been uploaded previously into the workflow. Uploaded images are temporarily stored in RAM rather than written to disk. This method has less overhead compared to embedding images as base64 into the prompt, but is more complex to implement.

• Inputs: id of an image that was uploaded previously
• Outputs: image (RGB) and mask (A of RGBA input, or first channel if no alpha present).

To upload an image, upload the bytes of a PNG via a HTTP PUT request to /api/etn/image/{id}. JPEG or other formats also work. Choose any id which does not clash with other images you upload, and reference it in the node. The request returns 201 if the image was uploaded and 200 if it was already cached.

Save Image to Cache

Stores an output image in RAM temporarily and allows retrieval over HTTP. This is typically faster than WebSocket, especially for large images.

• Inputs: the image (RGB or RGBA). Batches are supported.

This node will send a JSON message over WebSocket when an image is ready:

{
  "type": "executed",
  "data": {
    "node": "<node ID>",
    "output": {
      "images": [
        {"source": "http", "id": "<image ID>", "content-type": "image/png", "type": "output"}
      ]
    },
    "prompt_id": "prompt ID"
  }
}

To download the images, send a HTTP GET request to /api/etn/image/{id} with the image IDs from the message. Images will be cached for a few minutes.

Regions

These nodes implement attention masking for arbitrary number of image regions. Text prompts only apply to the masked area. In contrast to condition masking, this method is less "forceful", but leads to more natural image compositions.

Regions Attention Mask Workflow: region_attention_mask.json

Background Region

This node starts a list of regions. It takes a prompt, but no mask. The prompt is assigned to all image areas which are not covered by another region mask in the list.

Define Region

Appends a new region to a region list (or starts a new list). Takes a prompt, and mask which defines the area in the image

Verifiziert 11. Sept.