Latent Web Browser
A browser for web content generated by GPT-3.
Check out the Discord for news, updates, tips, and feature requests! https://t.co/uySHZIr3cc
See the latest demo here! https://twitter.com/flngr/status/1609616068057698304
Presentation
this project is free but experimental, you will have to configure your own API credentials to access OpenAI (for GPT-3 / text-davinci-003, and dall-e 2).
Currently Replicate (stable diffusion) is disabled, as I had some latency issues with it (maybe I will put it back in the future!)
Quick start
Prerequisites
You will need NVM and Node 18.12.1
Initial setup
cd web4
nvm use
yarnDon't mind too much those errors (we don't use IPFS yet):
electron@npm:1.8.8 couldn't be built successfully
wrtc@npm:0.4.7 couldn't be built successfully
The app will still load just fine.
BEFORE YOU RUN THE APP /!\
This application is based on OpenAI for text and image generation.
*You need a valid (billable) OpenAI account before you can use the application.
Upon starting it, you will have to enter you OpenAI token in the configuration panel (the "cog" icon).
Running the app in development
You have 3 different options to run the app:
In a browser (recommended)
This solution is recommended during development or if you experience build issues.
yarn devThen go to http://localhost:1420
In a standalone browser (advanced)
This solution requires a working Rust environment, and is recommended if you need to develop things interacting with the OS (eg. custom windows, system toolbar, auto updater..)
yarn tauri:devGenerate a production build for yourselves (advanced)
This is not recommended for day-to-day development as it is slow, and currently there is an issue with images.
yarn tauri buildthen:
- copy the app to your Application dir
- start the app
Known important bugs
Images don't seem to work when runing a standalone built using yarn tauri build, it might be caused by a security setting.
Using the browser
Working (most of the time) examples
Here are some examples to get you started:
a back-office application to manage users. There is a table with editable cells, a button to add a new user, and a counter of users.a simulation of calculating PI by generating random dots inside a circle. The simulator should include a slider to adjust the speed, a reset button, and the current estimate of PI.a simple app to compute your BMI, using form inputs for age, height and weight (in kilos)a whack-a-mole game but with spiders, a css 3-per-3 grid, emojis, and JS codea clone of asteroid using <canvas>, the mouse should orient the spaceship, it should fire bullets when clicking, and bullets can destroy asteroids.website for a company selling time travel visit packages (great pyramids, Trojan wars..). The website features 3 polaroid pictures taken by tourists of those eras
Non-working examples
Those examples don't work yet.. maybe one day in text-davinci-004 or 005?
a 120 BPM drum machine made using tone.js, with a step sequencer made using html checkboxes, to indicate when to play. Each row should be a different instrument (kick, snare, hihat), 8 buttons per row. There is a button to start/stop.
FAQ
The UI/UX should be improved
I agree!
Uhh.. something went wrong
Try clicking again on generate 🎲
No I mean, something went REALLY wrong
Maybe you did too many requests to OpenAI?
I don't.. know?
Wait a bit then restart the application, eg. kill it from the terminal.
Developer Guide (work in progress)
HTTP Request Mocking
Today web4 can mock the following file formats when the generated code tries to access them:
.jpg,.jpeg,.webm,.tga,.gif,.bmp->mockImage(filename).json->mockJSON(filename).txt->mockText(filename).stl->mockSTL(filename).svg->mockSVG(filename).*->mockAny(filename, extension)
You can find the code doing the mocking in src/pages/api/mocker/[filename].ts.
You can find the mocking prompts in src/engine/prompts/mocker/.
Do not hesitate to propose improvements to those prompts!