Skip to content

Latest commit

 

History

History
62 lines (42 loc) · 2.86 KB

README.md

File metadata and controls

62 lines (42 loc) · 2.86 KB
Raw

napkinmatic

Architecture Overview

Napkinmatic is a FOSS spatial computing framework for backend and traditional full-stack developers to easily interface their AI serverside apps with a mobile-native-Unity AR app.

The Spatial Computing App (SCA) refers to the mobile-native-Unity app, and Napkinmatic* refer to the serverside endpoints, with parameters defined below. Download the sample SCA here or visit Napkinmatic.com for the latest beta.

Note: you don't have to worry about the SCA or the frontend. Simply setup your serverside endpoints as defined in backend below.

The call architecture works like this:

sequenceDiagram
Spatial Computing App (SCA) ->> Napkinmatic Begin (NB): Posts image and prompt
Napkinmatic Begin (NB) ->> Spatial Computing App (SCA): Returns Callback URL (callbaack)
Spatial Computing App (SCA) --x Napkinmatic Callback (NC): polls with url=callback (status == success)
Napkinmatic Callback (NC) ->> Spatial Computing App (SCA): CallbackClass json
Loading

Frontend

Design your UI using React UIElements / Typescript.

This may make sense once you watch the Napkinmatic video

Included/Required UI Elements
  1. Create augmentable surface (currently, top down photo assuming parallel to plane)
  2. Capture image to be sent (in example, currently combined with above step)
  3. Enter prompt and submit
  4. Waiting/Progress (optional)
  5. Data returned UI set (optional)
Additional js functionality support

The image becomes a surface in 3-space, so you can assume the usual raycast data like point and normal hit, etc. For example: You can also write additional functionality using javascript, such as this mini hack made at a GDC 2023 AI Games hackathon giving you the ability to play paintball on your AI image!

Serverside

The serverside includes two general endpoint: the first to begin the call and the second is a polled callback that eventually returns the result when successfully processed. It is agnostic other than some standardized data structures described below.

Endpoint URLs

  • Napkinmatic Begin (NB) - accepts POST parameters image and prompt and returns a callback url
  • Napkinmatic Callback (NC) - this is polled with url= callback url returned above, as the results are being processed. When done, return status: success in the JSON format defined below for CallbackClass. output is an array of url to images. (The controlnet demo app uses output[1] by default)
{ 
  "id":"00000000-0000-0000-0000-000000000000",
  "status":"success",
  "output":["https://domain.com/image0.png","https://domain.com/image1.png"]
}

public class CallbackClass
{
    public string id;
    public string status;
    public string[] output;
}