# CompVis/taming-transformers

Taming Transformers for High-Resolution Image Synthesis

Repository: https://github.com/CompVis/taming-transformers
Canonical: https://ross.abutalabs.com/products/taming-transformers
Homepage: https://arxiv.org/abs/2012.09841
Language: Jupyter Notebook
License: MIT
License Family: permissive
Last push: 2024-07-30T18:27:31+00:00

## Health v2 (maintenance only)
Score: 32/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2085, "days_push": 764, "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 6521, forks 1219 (observed 2026-08-28T04:09:44.888244+00:00)

## What it is
The official implementation of 'Taming Transformers for High-Resolution Image Synthesis' (CVPR 2021), combining a convolutional VQGAN codebook with an autoregressive transformer to synthesize high-resolution images. It ships pretrained models for class-conditional ImageNet, FFHQ, CelebA-HQ, COCO, and ADE20k, plus training scripts and Colab notebooks.

## Use cases
- generate high-resolution images with transformers
- train a VQGAN on my own dataset
- run class-conditional ImageNet image synthesis
- semantically-guided image synthesis from segmentation maps
- reconstruct images with pretrained VQGAN models
- compare VQGAN with DALL-E

## When to choose
- you need the original VQGAN/transformer image synthesis models or their pretrained checkpoints
- you want to train a VQGAN tokenizer on custom data
- you're reproducing the CVPR 2021 paper or building on its codebook approach

## When to avoid
- you want modern latent diffusion image generation - use CompVis/latent-diffusion or Stable Diffusion instead
- you need a maintained production library - this is research code with infrequent updates
- you need text-to-image generation out of the box

## Facets
- artifact type: library
- maturity: maintenance
- function: machine-learning, deep-learning, image-processing, transformers
- domain: computer-vision, deep-learning, image-processing, artificial-intelligence
- platform: python, cross-platform
- tags: vqgan, image-synthesis, autoregressive-transformer, generative-models, research-code, cvpr-2021, gpu, linux

## Member repositories
- CompVis/taming-transformers (main) score 32

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:44.888244+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 2026-08-29T17:44:25.133312+00:00, confidence not recorded.
  - readme: https://github.com/CompVis/taming-transformers (fetched 2026-08-28T04:09:44.888244+00:00, sha d8c9e0231ce2)
  - homepage: https://arxiv.org/abs/2012.09841 (fetched 2026-08-29T08:40:46.980461+00:00, sha 9e806b3cb35a)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T08:40:46.989575+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T08:40:46.993056+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T08:40:46.994959+00:00, sha a1f16f915a9a)
  - registry_pypi: https://pypi.org/pypi/taming-transformers/json (fetched 2026-08-29T08:40:46.996639+00:00, sha f6af0db5500b)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T08:40:46.991455+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.