# RAE: Diffusion Transformers with Representation Autoencoders 

 

This repository contains the official PyTorch checkpoints for Representation Autoencoders. 

 

Representation Autoencoders (RAE) are a class of autoencoders that utilize pretrained, frozen representation encoders such as DINOv2 and SigLIP2 as encoders with trained ViT decoders. RAE can be used in a two-stage training pipeline for high-fidelity image synthesis, where a Stage 2 diffusion model is trained on the latent space of a pretrained RAE to generate images. 

# FreeFlow: Flow Map Distillation Without Data 

 

This repository contains the official PyTorch implementation for the paper: 

[**Flow Map Distillation Without Data**](https://huggingface.co/papers/2511.19428) 

 

# Vec2Face Model Card 

 

<div align="center"> 

 

[**Project Page**](https://haiyuwu.github.io/vec2face.github.io/) **|** [**Paper**](https://arxiv.org/abs/2409.02979) **|** [**Code**](https://github.com/HaiyuWu/vec2face) **|** [🤗 **Gradio demo**](https://huggingface.co/spaces/BooBooWu/Vec2Face) 

# HART: Efficient Visual Generation with Hybrid Autoregressive Transformer 

 

## Abstract 

We introduce Hybrid Autoregressive Transformer (HART), an autoregressive (AR) visual generation model capable of directly generating 1024x1024 images, rivaling diffusion models in image generation quality. Existing AR models face limitations due to the poor image reconstruction quality of their discrete tokenizers and the prohibitive training costs associated with generating 1024px images. To address these challenges, we present the hybrid tokenizer, which decomposes the continuous latents from the autoencoder into two components: discrete tokens representing the big picture and continuous tokens representing the residual components that cannot be represented by the discrete tokens. The discrete component is modeled by a scalable-resolution discrete AR model, while the continuous component is learned with a lightweight residual diffusion module with only 37M parameters. Compared with the discrete-only VAR tokenizer, our hybrid approach improves reconstruction FID from 2.11 to 0.30 on MJHQ-30K, leading to a 31% generation FID improvement from 7.85 to 5.38. HART also outperforms state-of-the-art diffusion models in both FID and CLIP score, with 4.5-7.7x higher throughput and 6.9-13.4x lower MACs. 

 

# Model Card for ImageNet 64x64 R3GAN Model 

 

This model card provides details about the R3GAN model trained on the ImageNet dataset found in the NeurIPS 2024 paper R3GAN: https://arxiv.org/abs/2501.05441 

 

## Model Details 

# Model Card for FFHQ 64x64 R3GAN Model 

 

This model card provides details about the R3GAN model trained on the CIFAR10 dataset found in the NeurIPS 2024 paper: https://arxiv.org/abs/2501.05441 

 

## Model Details 

# Model Card for FFHQ 256x256 R3GAN Model 

 

This model card provides details about the R3GAN model trained on the FFHQ-256 dataset found in the NeurIPS 2024 paper: https://arxiv.org/abs/2501.05441 

 

## Model Details 

# Model Card for FFHQ 64x64 R3GAN Model 

 

This model card provides details about the R3GAN model trained on the FFHQ-64 dataset found in the NeurIPS 2024 paper: https://arxiv.org/abs/2501.05441 

 

## Model Details 

# Model Card for ImageNet 32x32 R3GAN Model 

 

This model card provides details about the R3GAN model trained on the ImageNet dataset found in the NeurIPS 2024 paper: https://arxiv.org/abs/2501.05441 

 

## Model Details 

# Parallelized Autoregressive Visual Generation 

 

 

Autoregressive models have emerged as a powerful approach for visual generation but suffer from slow inference speed due to their sequential token-by-token prediction process. In this paper, we propose a simple yet effective approach for parallelized autoregressive visual generation that improves generation efficiency while preserving the advantages of autoregressive modeling. Our key insight is that parallel generation depends on visual token dependencies—tokens with weak dependencies can be generated in parallel, while strongly dependent adjacent tokens are difficult to generate together, as their independent sampling may lead to inconsistencies. Based on this observation, we develop a parallel generation strategy that generates distant tokens with weak dependencies in parallel while maintaining sequential generation for strongly dependent local tokens. Our approach can be seamlessly integrated into standard autoregressive models without modifying the architecture or tokenizer. Experiments on ImageNet and UCF-101 demonstrate that our method achieves a 3.6× speedup with comparable quality and up to 9.5× speedup with minimal quality degradation across both image and video generation tasks. 

 

This model is introduced in the paper [Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models](https://arxiv.org/abs/2501.01423). 

 

Code: https://github.com/hustvl/mar. 

# galaxy_gen 

 

`galaxy_gen` is a library to generate galaxy data/distributions. The models used are present in this page. 

 

## Installation 

# Inductive Moment Matching 

 

Model checkpoints for [Inductive Moment Matching](https://arxiv.org/abs/2503.07565). 

 

 

tag: unconditional-image-generation 

--- 

 

**Transformer AutoRegressive Flow Model** 

 

## 1. Overview 

 

**StyleGAN-VectorFind-v6,v7** models for [Oh-LoRA v2 Project](https://github.com/WannaBeSuperteur/AI_Projects/tree/main/2025_05_02_OhLoRA_v2). 

 

* These models are Fine-Tuned version of [GenForce StyleGAN celeba_partial-256x256](https://github.com/genforce/genforce/blob/master/MODEL_ZOO.md). 

## Boosting Generative Image Modeling via Joint Image-Feature Synthesis 

 

[Paper](https://arxiv.org/abs/2504.16064) | [Project Page](https://representationdiffusion.github.io/) | [Code](https://github.com/zelaki/ReDi) 

 

**ReDi** learns to generate coherent image-feature pairs from pure noise, significantly enhancing both generative quality and training efficiency. 

## 1. Overview 

 

**StyleGAN-VectorFind-v7,v8** models for [Oh-LoRA v3 Project](https://github.com/WannaBeSuperteur/AI_Projects/tree/main/2025_05_26_OhLoRA_v3). 

 

* These models are Fine-Tuned version of [GenForce StyleGAN celeba_partial-256x256](https://github.com/genforce/genforce/blob/master/MODEL_ZOO.md).