ComfyUI Setup

Orb uses a ComfyUI server to generate images. You install and run ComfyUI and its models separately from Orb.

For hardware and driver requirements, see the official ComfyUI documentation.

Before you start

• A GPU is strongly recommended.
• Reserve several gigabytes for each checkpoint.
• Install ComfyUI Manager if possible. It makes custom-node installation easier.

Install and start ComfyUI

WindowsmacOSLinux

ComfyUI Desktop

1. Download it from comfy.org/download.
2. Install and start ComfyUI.
3. Desktop uses port 8000 by default. In Settings → Server Config → Port, change it to 8188 and restart ComfyUI.

The portable build is for advanced users. Download it from the ComfyUI releases, then follow ComfyUI's instructions for enabling Manager. Use --cpu when you do not have an NVIDIA GPU.

1. Download the macOS build from comfy.org/download.
2. Open the DMG and move ComfyUI to Applications.
3. Start it and change the port from 8000 to 8188 in Settings → Server Config → Port.

The desktop build requires Apple Silicon and macOS 13 Ventura or newer.

comfy-cli

pip install comfy-cli
comfy install
comfy launch

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 manual commands use the NVIDIA/CUDA build. See the ComfyUI README for AMD, Intel, or CPU-only installs. The default server URL is http://127.0.0.1:8188.

Install a model

ComfyUI needs a checkpoint, text encoder, and VAE. The following files match the included starter workflows:

1. Download Anima and put anima-base-v1.0.safetensors in ComfyUI/models/checkpoints/.
2. Put qwen_3_06b_base.safetensors in ComfyUI/models/text_encoders/.
3. Put qwen_image_vae.safetensors in ComfyUI/models/vae/.
4. For a realistic style, download Real Dream and put real-dream-v2-anima-bf16.safetensors in models/checkpoints/.
5. Restart or refresh ComfyUI.

Test a starter workflow

1. Open Anima_Default.png in ComfyUI by dragging it into the window or selecting File → Open.
2. Select Run and wait for the image.
3. Save or export the result as a PNG. The PNG includes the workflow metadata.

You can import the included PNG directly into Orb without running it first. For the realistic model, use RealDream_Default.png.

Optional models

• MiaoMiao Harem is an anime-focused option. Its workflow is MiaoMiaoHarem_Default.png. It also needs UltraSharpV2 in models/upscale_models/.
• Krea 2 needs at least 24 GB of VRAM for the setup described here. Use Krea2_Default.png, the int8 checkpoint, the fp8 text encoder, and the Qwen VAE.

For image editing with a reference image, see the optional Krea 2 Edit Workflow.

Connect Orb to ComfyUI

For Orb and ComfyUI on the same computer, use http://127.0.0.1:8188 in Orb.

For another computer, or when Orb uses HTTPS, start ComfyUI with network access and CORS enabled:

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

Then use http://<server-ip>:8188 in Orb. Add a Bearer-token API key if the server requires one.

Warning

--listen 0.0.0.0 exposes ComfyUI to your network. Use it only on a trusted network, or protect the server with a token or reverse proxy.

Continue with Image Generation.