Practical overview
When it fits
Use TiM when you want to investigate or build with generative models that learn state-to-state transitions, or when you need its documented pretrained image-generation models and research workflows.
Who it is for
- Generative-model researchers studying alternative learning objectives.
- Developers exploring few-step and multi-step image generation with pretrained checkpoints.
- Researchers adapting text-to-image or class-guided training workflows.
Documented uses
- Generate images from text prompts using the TiM-T2I checkpoint and sampling scripts.
- Run class-guided image generation at 256×256 or 512×512 using the documented scripts.
- Train class-guided models on ImageNet or a text-to-image model on the provided toy dataset.
- Compare sampling behavior at different step counts and use the documented benchmark workflows.
Requirements & limits
The repository is public on GitHub and its README provides links to pretrained models and datasets on Hugging Face. The project is licensed under Apache-2.0.
- The central research idea is learning arbitrary state-to-state transitions across a generative trajectory.
- The README describes text-to-image and class-guided generation, with scripts for three text-to-image benchmarks and two ImageNet resolutions.
- The setup instructions specify Python 3.10, PyTorch 2.5.1, torchvision 0.20.1, CUDA 11.8 wheels, Ninja, and FlashAttention.
- The README marks the model zoo as a work in progress and does not state hardware requirements.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the TiM Python environment
Clone the repository and install the documented Python environment and project dependencies.
Before you start
- Conda
- Python 3.10 environment
- PyTorch 2.5.1 and torchvision 0.20.1 using the CUDA 11.8 wheel index
- Ninja, FlashAttention, and the repository requirements
Steps
Clone https://github.com/WZDTHU/TiM.git and change into the TiM directory.
Create and activate a Conda environment named tim_env with Python 3.10.
Install the specified PyTorch and torchvision versions using the CUDA 11.8 wheel index, then install Ninja and FlashAttention with --no-build-isolation.
Install the repository requirements with pip install -r requirements.txt, then install the project with pip install -e .
Expected outcome
The README's described TiM environment and editable project installation are set up.
Guide limitations
- The repository model zoo is marked as a work in progress.
Download a TiM-T2I checkpoint
Place the documented pretrained text-to-image checkpoint in the repository's checkpoints directory.
Before you start
- A local TiM repository checkout
- wget
Steps
From the repository root, create the checkpoints directory.
Download the TiM-T2I checkpoint from the documented Hugging Face URL and save it as checkpoints/t2i_model.bin.
Expected outcome
The TiM-T2I checkpoint file is placed at the path used in the README's download instructions.
Guide limitations
- This guide covers checkpoint retrieval only; it does not run image generation.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 18 October 2026
- Original website title
- WZDTHU/TiM · Official homepage title · retained in its source language