comfyui-Karga_RemoteWorkflows

Karga Remote Workflow

A standalone ComfyUI custom node that sends workflows to a remote ComfyUI instance and returns the output image — all in a single node. Part of the Karga ecosystem, maintained as its own separate repository.

Features

• Workflow picker — drop any workflow_api.json into the workflows/ folder and select it from a dropdown
• Remote execution — queues the job on a remote ComfyUI instance over HTTP and polls until complete
• Prompt & seed — built-in prompt box and seed control (with ComfyUI's native randomize/fixed/increment)
• Optional image input — connect an image for img2img workflows, or leave it disconnected for text-to-image
• Optional mask input — connect a mask for inpainting workflows
• Dynamic fields — any additional [ui]-tagged nodes in your workflow automatically appear as inputs on the node

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/YOUR_USERNAME/KargaRemoteWorkflow

Restart ComfyUI. The Karga Remote Workflow node will appear under the Karga category.

Setup

1. Export your remote workflow using Save (API Format) in ComfyUI
2. Place the exported JSON in custom_nodes/KargaRemoteWorkflow/workflows/
3. Make sure your remote ComfyUI is running with --listen

Usage

Add the Karga Remote Workflow node to your graph and configure:

| Input | Description | |---|---| | workflow | Select a workflow JSON from the workflows/ folder | | remote_address | Address of the remote ComfyUI instance, e.g. 192.168.1.50:8188 | | prompt | Text prompt — injected into the workflow's [ui] prompt node | | noise_seed | Seed — injected into the workflow's [ui] seed node | | image (optional) | Input image for img2img workflows | | mask (optional) | Mask for inpainting workflows | | poll_interval | How often to check for completion (seconds) | | timeout | Max wait time before raising an error (seconds) | | image_index | Which output image to return if the workflow produces multiple |

Tagging your remote workflow

Tag nodes in your remote workflow so this node knows where to inject values. Rename a node's title using either format:

Suffix style (recommended for standard nodes):

Load Image [ui]
CLIP Text Encode (Positive Prompt) [ui]
RandomNoise [ui]

Prefix style (for custom labels or non-standard nodes):

[ui] prompt:text
[ui] steps:steps
[ui] cfg:cfg

Supported suffix-style class types

| Class type | Injected value | Input key | |---|---|---| | LoadImage | Uploaded input image filename | image | | LoadImageMask | Uploaded mask filename | image | | CLIPTextEncode | Prompt text | text | | RandomNoise | Seed value | noise_seed |

Any other node type tagged with [ui] will fall back to using its first scalar input.

Prefix-style format

[ui] label_name:input_field

Examples:

[ui] steps:steps
[ui] cfg:cfg
[ui] input_image:image
[ui] input_mask:image

Any prefix-style [ui] fields that aren't prompt, noise_seed, input_image, or input_mask will appear as additional input widgets on the node automatically.

License

MIT