3X Your Interview Chances
AI Resume Builder
Import LinkedIn, get AI suggestions, land more interviews
4 min to read
The DeepSeek Janus Pro 1B model represents a significant advancement in the field of multimodal AI, capable of both image understanding and generation. With the integration of this model into ComfyUI, users can harness its capabilities on macOS systems, particularly those equipped with Apple Silicon (M1, M2, or M3 chips).
This comprehensive guide will walk you through the entire process of setting up and running DeepSeek Janus Pro 1B on macOS using ComfyUI, ensuring that you can leverage its powerful features effectively.
DeepSeek Janus Pro is a state-of-the-art multimodal model released by DeepSeek on January 27, 2025. It is designed to process and generate images based on textual prompts and vice versa.
The model comes in two versions: Janus Pro 1B and Janus Pro 7B, with the former being optimized for local deployment on consumer hardware.
DeepSeek Janus Pro 1B revolutionizes local AI with dual image-text capabilities on consumer hardware. Perfect for developers, artists, and researchers seeking:
Component | Minimum | Recommended |
---|---|---|
Mac Model | M1 (2020) | M3 Max (2023) |
RAM | 8GB | 16GB+ |
Storage | 10GB | 20GB SSD |
# Install Homebrew if missing
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
# Install Python 3.10 with Pyenv
brew install pyenv
pyenv install 3.10.13
pyenv global 3.10.13
# Verify installation
python --version # Should show 3.10.13
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
# Create isolated virtual environment
python -m venv venv
source venv/bin/activate
# Install with Metal acceleration
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/cpu
pip install -r requirements.txt
Method A - Plugin Manager (Recommended):
Manager
β Install Custom Nodes
Janus-Pro
β Click Install
Method B - Manual Installation:
cd ComfyUI/custom_nodes
git clone https://github.com/CY-CHENYUE/ComfyUI-Janus-Pro
pip install -r ComfyUI-Janus-Pro/requirements.txt
ComfyUI/
βββ models/
βββ Janus-Pro/
βββ Janus-Pro-1B/
βββ config.json
βββ pytorch_model.bin
βββ tokenizer.json
βββ # Other downloaded files
Checkpoint Loader
for "Janus-Pro-1B"Task | VRAM Usage | Optimization Strategy |
---|---|---|
Image Generation (512px) | 3.8GB | Use --medvram flag |
Batch Processing | 5.1GB | Limit batch size to 2 |
Text Analysis | 2.3GB | Enable CPU offload |
# In ComfyUI/config.yaml
janus_pro:
device: mps # Use Metal Performance Shaders
precision: fp16 # Half-precision mode
cache_dir: ~/janus_cache # Speed up subsequent loads
JanusProLoader
β Select "Janus-Pro-1B"CLIPTextEncode
β Enter prompt: "A cyberpunk cat wearing VR goggles"JanusImageGeneration
β Set resolution: 768x512PreviewImage
β Connect outputAdvanced Mode
β Increase guidance_scale
to 7.5Symptom | Fix |
---|---|
"Model not found" error | Verify folder structure capitalization |
Slow generation speeds | Enable --force-fp16 in launch arguments |
Black output images | Update torch to nightly build |
Japanese text garbled | Install extra fonts: brew install cask-fonts/font-ipafont |
# Check Metal support
python -c "import torch; print(torch.backends.mps.is_available())"
# Profile VRAM usage
system_profiler SPDisplaysDataType | grep VRAM
Task | Janus 1B | Janus 7B |
---|---|---|
Image β Text (1024px) | 1.8s | 4.9s |
Text β Image (512px) | 3.2s | 9.7s |
Bilingual Translation | 0.4s | 1.1s |
import requests
response = requests.post('http://localhost:8188/prompt', json={"prompt": workflow_json})
export PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.8
prevents VRAM over-allocationWhile setting up DeepSeek Janus Pro on macOS using ComfyUI may go smoothly for many users, some common issues may arise:
If you encounter issues during installation:
If models fail to load:
If experiencing slow performance:
Running DeepSeek Janus Pro 1B on macOS using ComfyUI opens up exciting possibilities in multimodal AI applications. By following this detailed guide, users can successfully install and utilize this powerful tool for both image understanding and generation tasks.