Atryon

Atryon (pronounced "a-try-on", the name comes from "AI Try On") is a free Chrome extension to see yourself wearing the various clothes you see on e-commerce websites.

Usage recommendations

Recommendations for the model

The plugin works by identifying pictures of clothes in the page (the "garment" pictures and replace them with your own photo (the "model" picture).

Please only use a picture of yourself as model.

Recommendations for the model model

I recommend to use model pictures adapted to the kind of clothes you want to try.

By default you should have least have a t-short and trousers. But for instance if you want to try a dress, you should also have a dress picture ready.

Recommendations for the garment image

Currently, the model works best with a clean garment image, with no body parts. If your image contains body parts (such as hair), those may appear in the final image.

Instructions for developers

run

run docker

docker compose up -d --build

Turn on developer mode in the Chrome Extension page and load dist file

restart

docker restart extension

logs

docker logs extension -f

stop

docker stop extension

background(service-worker) development

Clicking on service worker link in chrome://extensions/ launches the developer tool

build

Access within docker environment

docker compose exec extension ash

Run build command

/usr/src/app # yarn build

Deploying the backend

Note about billing

Running the try-on API can be expensive if you do too many requests, or leave the server on for too long.

Regardless of the provider, please verify your payment capacity and only spend what you can afford.

Deploy to Hugging Face

First install grog, then follow grog's README.md instructions.

To deploy the backend you will need two servers:

Segmentation server

A segmentation server (only necessary whenever you want to add a new picture). You can stop this server from running whenever you are done with preparing your pictures.

python grog.py --replicate_model_id jbilcke/oot_diffusion_with_mask --run_type huggingface_spaces --huggingface_token YOUR_OWN_HUGGINGFACE_TOKEN --space_hardware a10g-small

Substitution server

Then you need the actual substitution server, which does the image substitution job. Be careful of costs! You are strongly recommended to stop the server once you do not use it anymore.

python grog.py --replicate_model_id viktorfa/oot_segmentation --run_type huggingface_spaces --huggingface_token YOUR_OWN_HUGGINGFACE_TOKEN --space_hardware a10g-small

Deploy to Replicate

There is already a Replicate model deployed at viktorfa/oot_diffusion. Follow Replicate's instructions if you want your own "always-on" server.

Why using Atryon is better than a built-in try on widget that some websites have

Compatible with all platforms

A lot of e-commerce websites are a bit late when it comes to using AI technologies, and only a few are willing to develop or purchase a virtual try-on solution for their websites.

With Atryon this becomes a non-issue: regardless of the technology used by each shopping website, images will be replaced by a picture of yourself, all automatically.

Configure once, run everywhere

With Atryon, you do not need to setup a virtual 3D avatar, send your picture to each of the e-commerce website your visit.

Instead you do the setup once, and it "just works".

Privacy first

It can be a bit nerve wrecking to send your picture to a dozen of shopping platforms, you never know what they will do with it.

With Atryon, you only need to send your picture to one provider. And the best thing? You get to choose which one, you can even run it locally if you want!

By default we provide an access to Hugging Face, which preserves your privacy (images are not stored or logged).

jbilcke-hf / atryon Goto Github PK

atryon's Introduction