🔄 Auto-updated every 8 hours to always include the latest version.

💬 Feedback & Issues → GitHub Issues

🚀 This Docker image was originally built for running on RunPod, but it can also be used on your local machine. See the Local Setup Guide(WiP).

🔌 Exposed Ports

Port	Type	Service
22	TCP	SSH
3000	HTTP	ComfyUI
8080	HTTP	code-server
8888	HTTP	JupyterLab

---

🏷️ Tag Format

sombi/comfyui:(A)-torch2.8.0-(B)

• (A): slim or base
  • slim: ComfyUI + Manager only
  • base: slim + pre-installed custom nodes
• (B): CUDA version → cu124, cu126, cu128

---

🧱 Image Variants

Image Name	Custom Nodes	CUDA
sombi/comfyui:base-torch2.8.0-cu124	✅ Yes	12.4
sombi/comfyui:base-torch2.8.0-cu126	✅ Yes	12.6
sombi/comfyui:base-torch2.8.0-cu128	✅ Yes	12.8
sombi/comfyui:slim-torch2.8.0-cu124	❌ No	12.4
sombi/comfyui:slim-torch2.8.0-cu126	❌ No	12.6
sombi/comfyui:slim-torch2.8.0-cu128	❌ No	12.8

👉 To switch: Edit Pod/Template → set Container Image.

---

⚙️ Environment Variables

Variable	Description	Default
ACCESS_PASSWORD	Password for JupyterLab & code-server	(unset)
TIME_ZONE	Timezone (e.g., Asia/Seoul)	Etc/UTC
COMFYUI_EXTRA_ARGS	Extra ComfyUI options (e.g. --fast)	(unset)
INSTALL_SAGEATTENTION	Install SageAttention2 on start (True/False)	False
PRESET_DOWNLOAD	Download model presets at startup (comma-separated list). See below.	(unset)

👉 To set: Edit Pod/Template → Add Environment Variable (Key/Value).

⚠️ SageAttention2 requires Ampere+ GPUs and ~5 minutes to install.

---

🔧 Preset Downloads

The PRESET_DOWNLOAD environment variable accepts either a single preset or multiple presets separated by commas.
(e.g. WAINSFW_V140 or WAN22_I2V_A14B_GGUF_Q8_0,WAN22_LIGHTNING_LORA,WAN22_NSFW_LORA)
When set, the container will automatically download the corresponding models on startup.

You can also manually run the preset download script inside JupyterLab or code-server:

bash /download_presets.sh PRESET1,PRESET2,...

👉 To see which presets are available and view the download list for each, check the Wiki.

---

📁 Logs

App	Log Path
ComfyUI	/workspace/ComfyUI/user/comfyui_3000.log
code-server	/workspace/logs/code-server.log
JupyterLab	/workspace/logs/jupyterlab.log

---

🧩 Pre-installed Components

System

• OS: Ubuntu 24.04 (22.02 for CUDA 12.4)
• Python: 3.13
• Framework: ComfyUI + ComfyUI Manager + JupyterLab + code-server
• Libraries: PyTorch 2.8.0, CUDA (12.4–12.8), Triton, hf_hub, nvtop

Custom Nodes (only in base image)

• ComfyUI-KJNodes
• ComfyUI-WanVideoWrapper
• ComfyUI-GGUF
• ComfyUI-Easy-Use
• ComfyUI-Frame-Interpolation
• ComfyUI-mxToolkit
• ComfyUI-MultiGPU
• ComfyUI_TensorRT
• ComfyUI_UltimateSDUpscale
• comfyui-prompt-reader-node
• ComfyUI_essentials
• ComfyUI-Impact-Pack
• ComfyUI-Impact-Subpack
• efficiency-nodes-comfyui
• ComfyUI-Custom-Scripts
• ComfyUI_JPS-Nodes
• cg-use-everywhere
• ComfyUI-Crystools
• rgthree-comfy
• ComfyUI-Image-Saver
• comfy-ex-tagcomplete
• ComfyUI-VideoHelperSuite
• ComfyUI-wanBlockswap

👉 More details in the Wiki.