Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Tiny AutoEncoder for Stable Diffusion (and other image models)
| Date | Stars |
|---|---|
| 2026-07-31 | 958 |
| 2026-08-04 | 958 |
| 2026-08-06 | 958 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# 🍰 Tiny AutoEncoder for Stable Diffusion
## What is TAESD?
TAESD is very tiny autoencoder which uses the same "latent API" as Stable Diffusion's VAE*. TAESD can decode Stable Diffusion's latents into full-size images at (nearly) zero cost. Here's a comparison on my laptop:
TAESD is compatible with SD1/2-based models (using the `taesd_*` weights).
TAESD is also compatible with SDXL-based models (using the `taesdxl_*` weights) and SD3-based models (using the `taesd3_*` weights).
[Several non-SD models](#how-do-i-use-taesd-with-non-sd-models) have support too.
## How do I use TAESD with non-SD models?
* For **FLUX.1**, use the `taef1_*` weights.
* For **HiDream** and **Z-Image** (which use the same VAE as FLUX.1), you also want the `taef1_*` weights.
* For **FLUX.2**, use the `taef2_*` weights.
* For **Qwen Image** (which uses the Wan 2.1 VAE), check [TAEW2.1](https://github.com/madebyollin/taehv#what-video-models-does-taehv-support).
* For **Sana** ([`dc-ae-f32c32-sana-1.0`](https://huggingface.co/mit-han-lab/dc-ae-f32c32-sana-1.0-diffusers)), use the `taesana_*` weights.
## Where can I get TAESD?
* TAESD is already available in
* [A1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui)
* As a previewer, thanks to [Sakura-Luna](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/10365) (enable it in [Settings > Live Previews](https://github.com/madebyollin/taesd/issues/7#issuecomment-1669809631))
* As a encoder / decoder, thanks to [Kohaku-Blueleaf](https://github.com/AUTOMATIC1111/stable-diffusion-webui/pull/12311) (try it in [Settings > VAE](https://github.com/AUTOMATIC1111/stable-diffusion-webui/blob/dev/modules/shared_options.py#L152-L165))
* [vladmandic](https://github.com/vladmandic/automatic) thanks to [vladmandic](https://github.com/vladmandic/automatic/discussions/99#discussioncomment-6041142)
* [ComfyUI](https://github.com/comfyanonymous/ComfyUI)
* As a previewer, thanks to [space-nuko](https://github.com/comfyanonymous/ComfyUI/pull/713) (follow the instructions under ["How to show high-quality previews"](https://github.com/comfyanonymous/ComfyUI#how-to-show-high-quality-previews), then launch ComfyUI with `--preview-method taesd`)
* As a standalone VAE (download **both** `taesd_encoder.pth` and `taesd_decoder.pth` into `models/vae_approx`, then add a `Load VAE` node and set `vae_name` to `taesd`)
* TAESD is also available for [🧨 Diffusers](https://huggingface.co/docs/diffusers/main/en/api/models/autoencoder_tiny) in `safetensors` format
* [taesd](https://huggingface.co/madebyollin/taesd)
* [taesdxl](https://huggingface.co/madebyollin/taesdxl)
* [taesd3](https://huggingface.co/madebyollin/taesd3)
* [taef1](https://huggingface.co/madebyollin/taef1)
* [taef2](https://huggingface.co/madebyollin/taef2)
* TAESD's original weights are in this repo
## What can I use TAESD for?
Since TAESD is very fast, you can use TAESD to watch Stable Diffusion's image generation progress in real time. Here's a minimal [example notebook](examples/Previewing_During_Image_Generation.ipynb) that adds TAESD previewing to the 🧨 Diffusers implementation of SD2.1.
Since TAESD includes a tiny latent encoder, you can use TAESD as a cheap standalone VAE whenever the official VAE is [inconvenient](https://search.arxiv.org/?in=&query=taesd), like when doing real-time interactive image generation or when applying image-space loss functions to latent-space models. You can even run TAESD on a dedicated [tiny NPU](https://ieeexplore.ieee.org/document/10904763) for maximum efficiency.
Note that TAESD uses different scaling conventions than the official VAE (TAESD expects image values to be in [0, 1] instead of [-1, 1], and TAESD's "scale_factor" for latents is 1 instead of some [long](https://github.com/CompVis/stable-diffusion/blob/main/configs/Excerpt of 8,773 characters
Read on GitHub
36
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:0e29446c5e685caf, desc:stable diffusion