Skip to content
Plinken Help

ComfyUI

ComfyUI is a node-based visual programming interface for AI image and video generation. Plinken integrates with ComfyUI running on a RunPod GPU container, providing a workflow editor for creating complex generation pipelines.

ComfyUI Module

Section titled “ComfyUI Module”

The ComfyUI module (comfyui) manages the ComfyUI Manager panel and connection to the active pod.

Module Parameters

Section titled “Module Parameters”
ParameterTypeDescription
visibleboolComfyUI Manager panel visibility

Commands

Section titled “Commands”
Command IDDescription
comfyui:toggle-visibleToggle the ComfyUI Manager panel
pl.cmd("comfyui:toggle-visible")

Setup

Section titled “Setup”

Prerequisites

Section titled “Prerequisites”
  1. A running ComfyUI container on RunPod (see Container Manager)
  2. RunPod API key configured in Settings

Connection

Section titled “Connection”

When a ComfyUI pod is running, Plinken automatically:

  1. Detects the active pod
  2. Establishes an SSH tunnel
  3. Routes the ComfyUI API endpoint (comfy_url) to the local client
  4. Enables the ComfyUI panel

Workflows

Section titled “Workflows”

ComfyUI uses a node graph to define generation pipelines. Each node performs one operation:

Common Node Types

Section titled “Common Node Types”
NodeDescription
Load CheckpointLoad a Stable Diffusion model
CLIP Text EncodeConvert text prompt to embeddings
KSamplerRun the diffusion sampling process
VAE DecodeConvert latents to pixel images
Save ImageOutput the generated image
Load ImageInput an image for img2img or ControlNet
ControlNet ApplyApply ControlNet conditioning

Workflow Types

Section titled “Workflow Types”
WorkflowDescription
Text-to-ImageGenerate images from text prompts
Image-to-ImageTransform existing images with AI
InpaintingFill in masked regions of images
ControlNetGuide generation with structural inputs (pose, depth, edges)
Video GenerationGenerate video frames with AnimateDiff or similar
UpscalingAI upscaling with ESRGAN or similar models

Integration with Plinken

Section titled “Integration with Plinken”

Importing Results

Section titled “Importing Results”

Generated images and videos can be imported directly into your Plinken project:

  • Images → Compositor image layers or timeline assets
  • Videos → Timeline video tracks or compositor video layers

LoRA Support

Section titled “LoRA Support”

LoRA (Low-Rank Adaptation) weights can be loaded in ComfyUI workflows for fine-tuned generation. Manage LoRA files through the LoRA Manager.

FeatureDescription
Upload LoRAUpload .safetensors files to the pod
Browse LoRAList available LoRA models
Apply LoRAAdd LoRA nodes to workflows

Workflow Editor

Section titled “Workflow Editor”

The workflow editor provides a visual canvas for building ComfyUI node graphs:

FeatureDescription
Drag & dropAdd nodes from the node library
Wire connectionsConnect node outputs to inputs
PreviewSee intermediate results at each node
QueueSubmit workflows for execution
HistoryView past generation results

API

Section titled “API”

ComfyUI exposes a WebSocket + HTTP API for workflow execution:

EndpointMethodDescription
/promptPOSTSubmit a workflow for execution
/historyGETRetrieve generation history
/viewGETView generated images
/wsWSReal-time progress updates

Performance Tips

Section titled “Performance Tips”
  • Batch generation — Queue multiple prompts for efficient GPU utilization
  • Model caching — Keep frequently used models loaded to avoid reload time
  • Resolution — Start with 512×512 or 1024×1024, upscale after
  • Steps — 20–30 steps is usually sufficient for SDXL; Flux needs fewer (4–8 for Schnell)

Troubleshooting

Section titled “Troubleshooting”
IssueSolution
ComfyUI not connectingEnsure pod is running and SSH tunnel is active
Model not foundUpload models to the pod via the LoRA Manager or SSH
Out of memoryUse a larger GPU or reduce resolution/batch size
Slow generationCheck GPU utilization; consider a faster GPU tier
Section titled “Related”