README.md

---
license: apache-2.0
pipeline_tag: text-to-image
---

# Transition Models: Rethinking the Generative Learning Objective

This repository contains the Transition Models (TiM) presented in the paper [Transition Models: Rethinking the Generative Learning Objective](https://arxiv.org/abs/2509.04394).

TiM is a novel generative model designed for flexible photorealistic text-to-image generation. It achieves state-of-the-art performance with high efficiency by learning arbitrary state-to-state transitions, unifying few-step and many-step generation within a single model.

*   **Paper**: [https://arxiv.org/abs/2509.04394](https://arxiv.org/abs/2509.04394)
*   **Code**: [https://github.com/WZDTHU/TiM](https://github.com/WZDTHU/TiM)

## Highlights

*   Our Transition Models (TiM) are trained to master arbitrary state-to-state transitions. This approach allows TiM to learn the entire solution manifold of the generative process, unifying the few-step and many-step regimes within a single, powerful model.
    ![Figure](https://github.com/WZDTHU/TiM/raw/main/assets/illustration.png)
*   Despite having only 865M parameters, TiM achieves state-to-art performance, surpassing leading models such as SD3.5 (8B parameters) and FLUX.1 (12B parameters) across all evaluated step counts on GenEval benchmark. Importantly, unlike previous few-step generators, TiM demonstrates monotonic quality improvement as the sampling budget increases.
    ![Figure](https://github.com/WZDTHU/TiM/raw/main/assets/nfe_demo.png)
*   Additionally, when employing our native-resolution strategy, TiM delivers exceptional fidelity at resolutions up to `4096x4096`.
    ![Figure](https://github.com/WZDTHU/TiM/raw/main/assets/tim_demo.png)

## Quickstart

### 1. Setup

First, clone the repo:
```bash
git clone https://github.com/WZDTHU/TiM.git && cd TiM
```

#### 1.1 Environment Setup

```bash
conda create -n tim_env python=3.10
pip install torch==2.5.1 torchvision==0.20.1 --index-url https://download.pytorch.org/whl/cu118
pip install flash-attn
pip install -r requirements.txt
pip install -e .
```

#### 1.2 Model Download

Download the Text-to-Image model:
```bash
mkdir checkpoints
wget -c "https://huggingface.co/GoodEnough/TiM-T2I/resolve/main/t2i_model.bin" -O checkpoints/t2i_model.bin
```

### 2. Sampling (Text-to-Image Generation)

We provide the sampling scripts on three benchmarks. You can specify the sampling steps, resolutions, and CFG scale in the corresponding scripts.

Sampling with TiM-T2I model on GenEval benchmark:
```bash
bash scripts/sample/t2i/sample_t2i_geneval.sh
```