Janus-Series provides unified multimodal understanding and generation models (Janus, JanusFlow, Janus-Pro) with public model downloads on HuggingFace and examples for multimodal understanding and text-to-image generation. Installation is via pip install -e . and usage includes sample code for both Janus-Pro and Janus.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Janus-Series comprises unified multimodal understanding and generation models, including Janus, JanusFlow, and Janus-Pro. The project offers model downloads (Janus-1.3B, JanusFlow-1.3B, Janus-Pro-1B, Janus-Pro-7B) and online demos. The README describes capabilities across multimodal understanding and image generation, with references to associated papers (Janus, JanusFlow, Janus-Pro).
How it works
The models use a unified transformer architecture with decoupled visual encoding paths to support multimodal understanding and generation. The README presents a workflow where an image encoder produces embeddings consumed by a language model to generate text, and also demonstrates a text-to-image generation path that builds image tokens and decodes them via a vision model.
Getting started
Installation
pip install -e .
Quick Start (Janus-Pro)
pip install -e .
Simple Inference Example (Janus-Pro)
import torch
from transformers import AutoModelForCausalLM
from janus.models import MultiModalityCausalLM, VLChatProcessor
from janus.utils.io import load_pil_images
model_path = "deepseek-ai/Janus-Pro-7B"
vl_chat_processor: VLChatProcessor = VLChatProcessor.from_pretrained(model_path)
tokenizer = vl_chat_processor.tokenizer
vl_gpt: MultiModalityCausalLM = AutoModelForCausalLM.from_pretrained(
model_path, trust_remote_code=True
)
vl_gpt = vl_gpt.to(torch.bfloat16).cuda().eval()
# ... additional setup as shown in the README ...
Gradio Demo
pip install -e .[gradio]
python demo/app_januspro.py
Recent releases
- News entries include 2025.01.27 (Janus-Pro release) and 2024.11.13 (JanusFlow release), with related papers and demos linked in the README. No formal releases are listed under the RELEASES section (latest 0: none).
Traction
Stars: 17756
Behind the repo
Hosted on HuggingFace spaces for online demos: Janus-Pro-7B, Janus, JanusFlow references to online demos are provided in the README.
Caveats
License: MIT (Code), Model License: Model_Agreement. The README includes model download table with sequence length and Hugging Face links. Creation date: 2024-10-18; Last push: 2025-02-01. Open issues: 181. Language: Python.