ComfyUI
Node-based image generation UI for Stable Diffusion and Flux models.
Overview
| Property | Value |
|---|---|
| CDK8s file | platform/cdk8s/cots/ai/comfyui.go |
| Namespace | comfyui |
| Image | yanwk/comfyui-boot:cu128-megapak-20260223 |
| HTTPRoute | comfyui.madhan.app → comfyui:8188 |
| UI | Yes |
| Storage | 100 Gi Longhorn PVC (models, outputs, custom nodes) |
| Node | k8s-worker4 (GPU) |
Purpose
ComfyUI is a node-based GUI for Stable Diffusion and Flux image generation models. It downloads model checkpoints to a persistent Longhorn volume.
GPU Configuration
runtimeClassName: nvidia
resources:
requests:
memory: "1Gi"
limits:
memory: "8Gi"
nvidia.com/gpu: "1"
env:
- name: NVIDIA_VISIBLE_DEVICES
value: all
Deployment Strategy
Uses strategy: Recreate to avoid multi-attach conflicts on the Longhorn RWO PVC.
Storage
A 100 Gi Longhorn PVC is mounted at /home/runner (or the image's working directory). This stores:
- Model checkpoints (
.safetensors,.ckpt) - LoRA files
- Custom nodes
- Generated output images
Image Note
The image tag cu128-megapak-20260223 (CUDA 12.8, megapak variant) includes PyTorch, xformers, and common custom nodes pre-installed. The tag latest-cu128 does not exist on Docker Hub — use a dated tag.
Accessing the UI
Navigate to http://comfyui.madhan.app in a browser. The ComfyUI node graph interface loads directly.