# FoundationVision/VAR

[NeurIPS 2024 Best Paper Award][GPT beats diffusion🔥] [scaling laws in visual generation📈] Official impl. of "Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction". An *ultra-simple, user-friendly yet state-of-the-art* codebase for autoregressive image generation!

Repository: https://github.com/FoundationVision/VAR
Canonical: https://ross.abutalabs.com/products/var
Language: Jupyter Notebook
License: MIT
License Family: permissive
Topics: auto-regressive-model, diffusion-models, image-generation, transformers, autoregressive-models, generative-ai, generative-model, gpt, gpt-2, large-language-models, vision-transformer, neurips
Last push: 2025-11-10T21:42:29+00:00

## Health v2 (maintenance only)
Score: 48/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 51, release rhythm 35, longevity 63
- inputs: {"age_days": 884, "days_push": 296, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 8729, forks 572 (observed 2026-08-28T04:10:24.462664+00:00)

## What it is
Official PyTorch implementation of Visual Autoregressive Modeling (VAR), a NeurIPS 2024 Best Paper-winning method for scalable image generation via next-scale prediction. It demonstrates GPT-style autoregressive models beating diffusion models on ImageNet generation with observed scaling laws.

## Use cases
- generate images with autoregressive transformers
- train a GPT-style image generation model
- reproduce NeurIPS 2024 best paper results
- study scaling laws in visual generation
- compare autoregressive vs diffusion image generation
- fine-tune pretrained VAR checkpoints on custom datasets

## When to choose
- you want state-of-the-art autoregressive image generation
- you're researching scaling laws for visual generation
- you need a simple, well-documented generative model codebase
- you want an alternative to diffusion models

## When to avoid
- you need production text-to-image with complex prompt following out of the box
- you lack GPU resources for large transformer training or inference
- you need video or audio generation (see the authors' InfinityStar instead)

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, image-processing, transformers
- domain: deep-learning, computer-vision, image-processing, artificial-intelligence, large-language-models
- platform: python, cross-platform
- tags: autoregressive-model, image-generation, generative-models, next-scale-prediction, vision-transformer, text-to-image, scaling-laws, neurips-2024, gpt-style, research-code, gpu, linux

## Member repositories
- FoundationVision/VAR (main) score 48

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:10:24.462664+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:25:19.520791+00:00, confidence not recorded.
  - readme: https://github.com/FoundationVision/VAR (fetched 2026-08-28T04:10:24.462664+00:00, sha 6051df7a724a)
- Data as of 2026-08-30T08:39:29.467469+00:00.