Skip to content

ComfyUI Setup

Orb renders images with a ComfyUI server. Orb does not install ComfyUI or image models — you run ComfyUI yourself and point Orb at it.

This page is a quick, out-of-the-box setup that gets you to a running ComfyUI server Orb can reach. For hardware requirements, GPU drivers, and advanced configuration, see the official ComfyUI documentation.

Before you start

  • A GPU is strongly recommended. Check the system requirements.
  • Enough disk space for a checkpoint (typically 2–7 GB each).

Install and launch ComfyUI

It's recommended to run with ComfyUI Manager. We'll install custom nodes later more easily with it.

Easiest — ComfyUI Desktop

  1. Download the installer from comfy.org/download.
  2. Run the installer and launch ComfyUI. ComfyUI-Manager is included and enabled by default.
  3. Desktop listens on port 8000, not 8188. Open Settings → Server Config → Port, set it to 8188, and restart ComfyUI. The server is then at http://127.0.0.1:8188.

Portable build (advanced)

  1. Download the portable .7z from the ComfyUI releases.

  2. Extract it. The .bat launchers ignore any arguments you append, so to run with the Manager enabled, open a terminal in the extracted folder and run:

    .\python_embeded\python.exe -m pip install -r ComfyUI\manager_requirements.txt
.\python_embeded\python.exe -s ComfyUI\main.py --windows-standalone-build --enable-manager

    (Add --cpu to the second command if you don't have an NVIDIA GPU.) If you don't want the Manager, run_nvidia_gpu.bat / run_cpu.bat work as-is.

ComfyUI Desktop

  1. Download the macOS build from comfy.org/download.
  2. Open the .dmg and drag ComfyUI to Applications.
  3. Launch it. ComfyUI-Manager is included and enabled by default.
  4. Desktop listens on port 8000, not 8188. Open Settings → Server Config → Port, set it to 8188, and restart ComfyUI. The server is then at http://127.0.0.1:8188.

Requires Apple Silicon (M1 or later) and macOS 13 Ventura or newer — Intel Macs are not supported. First launch may be slow while dependencies initialize.

comfy-cli (recommended)

pip install comfy-cli
comfy install
comfy launch

comfy install includes ComfyUI-Manager, and comfy launch enables it for you.

Manual install

git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu130
pip install -r requirements.txt
pip install -r manager_requirements.txt
python main.py --enable-manager

(The torch line is the NVIDIA/CUDA build; see the ComfyUI README for AMD, Intel, or CPU-only.)

The server starts at http://127.0.0.1:8188.

Make sure ComfyUI starts up without any problems.

Download checkpoints

ComfyUI needs at least one checkpoint (image model) to render anything.

  1. Go to https://civitai.com/models/2458426/anima?modelVersionId=2945208
  2. Download the Anima checkpoint (anima-base-v1.0.safetensors) as a .safetensors file and place it in ComfyUI/models/checkpoints/.
  3. Download the text encoder (qwen_3_06b_base.safetensors) and put it in ComfyUI/models/text_encoders/.
  4. Download the VAE (qwen_image_vae.safetensors) and put it in ComfyUI/models/vae/.
  5. Restart ComfyUI, or refresh the UI.

Do the same for the realistic model: https://civitai.red/models/153568/real-dream?modelVersionId=3098044

Simply download and put real-dream-v2-anima-bf16.safetensors in ComfyUI/models/checkpoints/.

Create your first ComfyUI gen

  1. Download the Anima_Default.png and drag it into ComfyUI. The embedded workflow loads automatically.
  2. If drag and drop doesn't work, go to ComfyUI -> File -> Open, then select the image.
  3. Click Run button (top right corner) and wait, your GPU will work, then an image will show up.
  4. Export/Save the output image as a PNG file. This file contains the whole workflow config which we'll import into Orb later.

Or you can also just import the above default PNG workflows straight into Orb, no need to even touch ComfyUI.

For the realistic model, do RealDream_Default.png

A great anime-only model in case you find base Anima lacking:

https://civitai.com/models/934764/miaomiao-harem?modelVersionId=3125933

Workflow: MiaoMiaoHarem_Default.png

Download https://huggingface.co/Kim2091/UltraSharpV2/resolve/main/4x-UltraSharpV2.safetensors and put it in ComfyUI/models/upscale_models/

If you're GPU-rich (24GB+ VRAM), try Krea 2:

https://civitai.red/models/2760803/dasiwa-krea2-or-turbo-or-raw?modelVersionId=3151280 (Download the int8-convrot version and put it in ComfyUI/models/checkpoints/)

Workflow: Krea2_Default.png

https://civitai.red/models/2731465/qwen3-vl-4b-abliterated-comfyui-krea-2-text-encoder-bf16-fp8?modelVersionId=3070870 (Download the fp8 version and put it in ComfyUI/models/text_encoders/)

https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI/blob/main/split_files/vae/qwen_image_vae.safetensors (Download the full weights and put it in ComfyUI/models/vae/)

There are other ways to run Krea 2 with lower VRAM but won't go into that here.

To edit an existing picture instead of drawing a new one — keeping a character's face across images — see Reference Image Setup.

Make ComfyUI reachable from Orb

Orb talks to ComfyUI from your browser, so the server must accept requests from Orb's origin.

  • Same machine, default port: the URL is http://127.0.0.1:8188. Nothing extra needed.

  • Different machine, or Orb served over HTTPS: launch ComfyUI so it listens on the network and allows cross-origin requests:

    python main.py --listen 0.0.0.0 --enable-cors-header --enable-manager

    Then use http://<server-ip>:8188 as the URL in Orb.

Warning

--listen 0.0.0.0 exposes ComfyUI on your network. Only do this on a trusted network, and consider a Bearer token / reverse proxy if it's reachable more broadly. Orb supports an API key for Bearer-token servers.

Next step

Your ComfyUI server is ready. Head to Image Generation to enter the URL, pick a checkpoint per style, and test the connection.