VAR: a new visual generation method elevates GPT-style models beyond diffusion🚀 & Scaling laws observed📈
Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction
NeurIPS 2024 Best Paper
- 2025-11: We Release our Text-to-Video generation model InfinityStar based on VAR & Infinity, please check Infinity⭐️.
- 2025-11: 🎉 InfinityStar is accepted as NeurIPS 2025 Oral.
- 2025-04: 🎉 Infinity is accepted as CVPR 2025 Oral.
- 2024-12: 🏆 VAR received NeurIPS 2024 Best Paper Award.
- 2024-12: 🔥 We Release our Text-to-Image research based on VAR, please check Infinity.
- 2024-09: VAR is accepted as NeurIPS 2024 Oral Presentation.
- 2024-04: Visual AutoRegressive modeling is released.
We provide a demo website for you to play with VAR models and generate images interactively. Enjoy the fun of visual autoregressive modeling!
We provide a demo website for you to play with VAR Text-to-Image and generate images interactively. Enjoy the fun of visual autoregressive modeling!
We also provide demo_sample.ipynb for you to see more technical details about VAR.
Visual Autoregressive Modeling (VAR) redefines the autoregressive learning on images as coarse-to-fine "next-scale prediction" or "next-resolution prediction", diverging from the standard raster-scan "next-token prediction".
For a deep dive into our analyses, discussions, and evaluations, check out our paper.
We provide VAR models for you to play with, which are on 
| model | reso. | FID | rel. cost | #params | HF weights🤗 |
|---|---|---|---|---|---|
| VAR-d16 | 256 | 3.55 | 0.4 | 310M | var_d16.pth |
| VAR-d20 | 256 | 2.95 | 0.5 | 600M | var_d20.pth |
| VAR-d24 | 256 | 2.33 | 0.6 | 1.0B | var_d24.pth |
| VAR-d30 | 256 | 1.97 | 1 | 2.0B | var_d30.pth |
| VAR-d30-re | 256 | 1.80 | 1 | 2.0B | var_d30.pth |
| VAR-d36 | 512 | 2.63 | - | 2.3B | var_d36.pth |
You can load these models to generate images via the codes in demo_sample.ipynb. Note: you need to download vae_ch160v4096z32.pth first.
-
Install
torch>=2.0.0. -
Install other pip packages via
pip3 install -r requirements.txt. -
Prepare the ImageNet dataset
assume the ImageNet is in `/path/to/imagenet`. It should be like this:
/path/to/imagenet/: train/: n01440764: many_images.JPEG ... n01443537: many_images.JPEG ... val/: n01440764: ILSVRC2012_val_00000293.JPEG ... n01443537: ILSVRC2012_val_00000236.JPEG ...NOTE: The arg
--data_path=/path/to/imagenetshould be passed to the training script. -
(Optional) install and compile
flash-attnandxformersfor faster attention computation. Our code will automatically use them if installed. See models/basic_var.py#L15-L30.
To train VAR-{d16, d20, d24, d30, d36-s} on ImageNet 256x256 or 512x512, you can run the following command:
# d16, 256x256
torchrun --nproc_per_node=8 --nnodes=... --node_rank=... --master_addr=... --master_port=... train.py \
--depth=16 --bs=768 --ep=200 --fp16=1 --alng=1e-3 --wpe=0.1
# d20, 256x256
torchrun --nproc_per_node=8 --nnodes=... --node_rank=... --master_addr=... --master_port=... train.py \
--depth=20 --bs=768 --ep=250 --fp16=1 --alng=1e-3 --wpe=0.1
# d24, 256x256
torchrun --nproc_per_node=8 --nnodes=... --node_rank=... --master_addr=... --master_port=... train.py \
--depth=24 --bs=768 --ep=350 --tblr=8e-5 --fp16=1 --alng=1e-4 --wpe=0.01
# d30, 256x256
torchrun --nproc_per_node=8 --nnodes=... --node_rank=... --master_addr=... --master_port=... train.py \
--depth=30 --bs=1024 --ep=350 --tblr=8e-5 --fp16=1 --alng=1e-5 --wpe=0.01 --twde=0.08
# d36-s, 512x512 (-s means saln=1, shared AdaLN)
torchrun --nproc_per_node=8 --nnodes=... --node_rank=... --master_addr=... --master_port=... train.py \
--depth=36 --saln=1 --pn=512 --bs=768 --ep=350 --tblr=8e-5 --fp16=1 --alng=5e-6 --wpe=0.01 --twde=0.08A folder named local_output will be created to save the checkpoints and logs.
You can monitor the training process by checking the logs in local_output/log.txt and local_output/stdout.txt, or using tensorboard --logdir=local_output/.
If your experiment is interrupted, just rerun the command, and the training will automatically resume from the last checkpoint in local_output/ckpt*.pth (see utils/misc.py#L344-L357).
For FID evaluation, use var.autoregressive_infer_cfg(..., cfg=1.5, top_p=0.96, top_k=900, more_smooth=False) to sample 50,000 images (50 per class) and save them as PNG (not JPEG) files in a folder. Pack them into a .npz file via create_npz_from_sample_folder(sample_folder) in utils/misc.py#L344.
Then use the OpenAI's FID evaluation toolkit and reference ground truth npz file of 256x256 or 512x512 to evaluate FID, IS, precision, and recall.
Note a relatively small cfg=1.5 is used for trade-off between image quality and diversity. You can adjust it to cfg=5.0, or sample with autoregressive_infer_cfg(..., more_smooth=True) for better visual quality.
We'll provide the sampling script later.
In this pargraph, we cross link third-party repositories or research which use VAR and report results. You can let us know by raising an issue
(Note please report accuracy numbers and provide trained models in your new repository to facilitate others to get sense of correctness and model behavior)
| Time | Research | Link |
|---|---|---|
| [5/12/2025] | [ICML 2025]Continuous Visual Autoregressive Generation via Score Maximization | https://github.com/shaochenze/EAR |
| [5/8/2025] | Generative Autoregressive Transformers for Model-Agnostic Federated MRI Reconstruction | https://github.com/icon-lab/FedGAT |
| [4/7/2025] | FastVAR: Linear Visual Autoregressive Modeling via Cached Token Pruning | https://github.com/csguoh/FastVAR |
| [4/3/2025] | VARGPT-v1.1: Improve Visual Autoregressive Large Unified Model via Iterative Instruction Tuning and Reinforcement Learning | https://github.com/VARGPT-family/VARGPT-v1.1 |
| [3/31/2025] | Training-Free Text-Guided Image Editing with Visual Autoregressive Model | https://github.com/wyf0912/AREdit |
| [3/17/2025] | Next-Scale Autoregressive Models are Zero-Shot Single-Image Object View Synthesizers | https://github.com/Shiran-Yuan/ArchonView |
| [3/14/2025] | Safe-VAR: Safe Visual Autoregressive Model for Text-to-Image Generative Watermarking | https://arxiv.org/abs/2503.11324 |
| [3/3/2025] | [ICML 2025]Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator | https://research.nvidia.com/labs/dir/ddo/ |
| [2/28/2025] | Autoregressive Medical Image Segmentation via Next-Scale Mask Prediction | https://arxiv.org/abs/2502.20784 |
| [2/27/2025] | FlexVAR: Flexible Visual Autoregressive Modeling without Residual Prediction | https://github.com/jiaosiyu1999/FlexVAR |
| [2/17/2025] | MARS: Mesh AutoRegressive Model for 3D Shape Detailization | https://arxiv.org/abs/2502.11390 |
| [1/31/2025] | [ICML 2025]Visual Autoregressive Modeling for Image Super-Resolution | https://github.com/quyp2000/VARSR |
| [1/21/2025] | VARGPT: Unified Understanding and Generation in a Visual Autoregressive Multimodal Large Language Model | https://github.com/VARGPT-family/VARGPT |
| [1/26/2025] | [ICML 2025]Visual Generation Without Guidance | https://github.com/thu-ml/GFT |
| [12/30/2024] | Next Token Prediction Towards Multimodal Intelligence | https://github.com/LMM101/Awesome-Multimodal-Next-Token-Prediction |
| [12/30/2024] | Varformer: Adapting VAR’s Generative Prior for Image Restoration | https://arxiv.org/abs/2412.21063 |
| [12/22/2024] | [ICLR 2025]Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching | https://github.com/imagination-research/distilled-decoding |
| [12/19/2024] | FlowAR: Scale-wise Autoregressive Image Generation Meets Flow Matching | https://github.com/OliverRensu/FlowAR |
| [12/13/2024] | 3D representation in 512-Byte: Variational tokenizer is the key for autoregressive 3D generation | https://github.com/sparse-mvs-2/VAT |
| [12/9/2024] | CARP: Visuomotor Policy Learning via Coarse-to-Fine Autoregressive Prediction | https://carp-robot.github.io/ |
| [12/5/2024] | [CVPR 2025]Infinity ∞: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis | https://github.com/FoundationVision/Infinity |
| [12/5/2024] | [CVPR 2025]Switti: Designing Scale-Wise Transformers for Text-to-Image Synthesis | https://github.com/yandex-research/switti |
| [12/4/2024] | [CVPR 2025]TokenFlow🚀: Unified Image Tokenizer for Multimodal Understanding and Generation | https://github.com/ByteFlow-AI/TokenFlow |
| [12/3/2024] | XQ-GAN🚀: An Open-source Image Tokenization Framework for Autoregressive Generation | https://github.com/lxa9867/ImageFolder |
| [11/28/2024] | [CVPR 2025]CoDe: Collaborative Decoding Makes Visual Auto-Regressive Modeling Efficient | https://github.com/czg1225/CoDe |
| [11/28/2024] | [CVPR 2025]Scalable Autoregressive Monocular Depth Estimation | https://arxiv.org/abs/2411.11361 |
| [11/27/2024] | [CVPR 2025]SAR3D: Autoregressive 3D Object Generation and Understanding via Multi-scale 3D VQVAE | https://github.com/cyw-3d/SAR3D |
| [11/26/2024] | LiteVAR: Compressing Visual Autoregressive Modelling with Efficient Attention and Quantization | https://arxiv.org/abs/2411.17178 |
| [11/15/2024] | M-VAR: Decoupled Scale-wise Autoregressive Modeling for High-Quality Image Generation | https://github.com/OliverRensu/MVAR |
| [10/14/2024] | [ICLR 2025]HART: Efficient Visual Generation with Hybrid Autoregressive Transformer | https://github.com/mit-han-lab/hart |
| [10/12/2024] | [ICLR 2025 Oral]Toward Guidance-Free AR Visual Generation via Condition Contrastive Alignment | https://github.com/thu-ml/CCA |
| [10/3/2024] | [ICLR 2025]ImageFolder🚀: Autoregressive Image Generation with Folded Tokens | https://github.com/lxa9867/ImageFolder |
| [07/25/2024] | ControlVAR: Exploring Controllable Visual Autoregressive Modeling | https://github.com/lxa9867/ControlVAR |
| [07/3/2024] | VAR-CLIP: Text-to-Image Generator with Visual Auto-Regressive Modeling | https://github.com/daixiangzi/VAR-CLIP |
| [06/16/2024] | STAR: Scale-wise Text-to-image generation via Auto-Regressive representations | https://arxiv.org/abs/2406.10797 |
This project is licensed under the MIT License - see the LICENSE file for details.
If our work assists your research, feel free to give us a star ⭐ or cite us using:
@Article{VAR,
title={Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction},
author={Keyu Tian and Yi Jiang and Zehuan Yuan and Bingyue Peng and Liwei Wang},
year={2024},
eprint={2404.02905},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@misc{Infinity,
title={Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis},
author={Jian Han and Jinlai Liu and Yi Jiang and Bin Yan and Yuqi Zhang and Zehuan Yuan and Bingyue Peng and Xiaobing Liu},
year={2024},
eprint={2412.04431},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2412.04431},
}