/augmented_reality_SD

Combining augmented reality and stability diffusion through api calls.

Primary LanguageJavaScriptMIT LicenseMIT

Augmented Reality and Stable Diffusion

A webapp that combines augmented reality and stability diffusion through api calls. Requires running stable diffusion on a seperate machine.

Examples

video

Installation

local backend setup

do on local computer

git clone https://github.com/graemeniedermayer/augmented_reality_SD.git

launch automatic1111 with api option (currently it's only setup for the 512 models)

python launch.py --xformers --api

launch proxy pass server from the project root directory.

python proxy_pass_server.py

do on android phone

launch webapp in portrait mode (I haven't done all the corrections for landscape mode.)

first go to https://192.168.0.1:8443 (using correct private ip address).

web browser should complain about using an inapproriate certificate. Allow the certificate. This should temporarily allow the self-signed certificate for all browser requests.

auto1111.html will contain the default setting. There are going to be a bunch of variants and we'll figure out what works best!

do on iOS phone

Todo. mozilla webxr mobile viewer might work for some setups (sort of doubt that depth api or camera access will work)

Cloud Backend

google colab

Todo.

Using extensions

Many extensions can be very helpful while using augmented reality.

Use full name that would appear in the script area.

Updates

v 0.0.1 boilerplate

Setup

So this setup currently requires a front-end, proxy-pass, and back-end.

Frontend

The current frontend is webxr. Frontend must be hosted with certificate SSL (last time I check self-signed was not sufficient).

Because of the limitations of webxr and iOS I'd like to explore creating a iOS app in the future.

Proxy pass

Using a fastapi proxy pass for SSL and cross-site origin resources (CORS) headers. If backend has SSL and CORS enabled this is unnecessary. Automatic1111 supports both but I was having trouble with self-signed certificates. I haven't looked into google Colab yet (it might not be necessary).

Backend

So far automatic1111 is the only backend. I'd like to look into invokeAI and other possible backends.

Known bugs

  • horizontal images does not work. Resolution changes in general aren't well supported.
  • must double click capture depth (this is a weird error).
  • there is no failure capturing/error capture.

Useful extensions

depthmaps

background removal

Acknowledgements

So many. To do