3d-to-photo/README.md

3D to Photo

3D to Photo is an open-source package by Dabble, that combines threeJS and Stable diffusion to build a virtual photo studio for product photography. Load a 3D model into the browser and virtual shoot it in any kind of scene you can imagine. The app currently uses Stable Diffusion 1.5-inpainting, hosted on Replicate.

Demo Video

How it Works

• Upload a 3D model of any object in .glb format.
• Adjust the position and orientation of the model on the canvas
• Describe the scene you want to create in the text box and click generate image

Use Cases

• Product Photography: Create product lifestyle photos in any backdrop you can imagine, without a physical photoshoot.
• Synthetic Data Generation: Generate synthetic images of an item in a variety of scenes. Useful when training object detection models.
• Previsualize Game Assets: Upload your game assets and generate level art around it to previsualize scenes.

Tech Stack

• ThreeJS to handle loading and viewing 3D models
• Stable Diffusion 1.5 (inpainting) by stability.ai
• Replicate to run Stable Diffusion
• NextJS by Vercel for the front-end
• Python Flask Server for some backend image processing functions

Install and Run

1. Clone this repository

git clone git@github.com:Dabble-Studio/3d-to-photo.git

2. Install necessary packages for the front end

cd 3d-to-photo
npm install

3. Install necessary packages for the python backend

cd image_proc_server
pip install -r requirements.txt

4. Run the Flask Server

flask run

5. Set your Replicate API key. In another terminal, navigate to the root folder and create a file called .env. You can use the .env.example file as a template. Paste your Replicate API key in this line REPLICATE_API_TOKEN=YOUR_API_TOKEN, in place of YOUR_API_TOKEN

6. Run the NextJS app

npm run dev

7. Use one of the sample 3D models and drag it into the upload area. Enter a prompt in the text box and click "Generate Image"