Tomas-Lawton / Reframer

Research project based on vector drawing with a text-to-image generative AI model called CLIP. View a demo video here: https://vimeo.com/760319552

Home Page:https://tomas-lawton.github.io/Reframer/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Reframer - Co-creative drawing with CLIP!

Tomas Lawton, Francisco Ibarrola and Kazjon Grace

Maximal control, minimal obtrusiveness has long been the hallmark of creativity support (Shneiderman, 2000). Reframer (shown by Figure 4.1) is functioning platform disrupting the status-quo by combining sketching with state-of-the-art AI algorithms for agent-augmented conceptual design. Unlike the vast majority of text-to-image generative systems, users are directly involved in the process of creation, and the AI is capable of making suggestions based on the user prompt. Users can create a sketch and a enter a prompt and Reframer will add strokes and modify the sketch. As this happens, users maintain seamless editorial control as they accept, modify, or reject AI contributions in real-time. In the final version, users can even describe parts of the sketch to the AI, thereby creating a goal hierarchy composed of multiple overlapping prompts. As the user draws, sketch developments are used as model inputs and the AI agent develops the sketch through optimisation based on differentiable rendering and CLIP (Frans et al., 2021; Radford et al., 2021). The model makes changes by gradually moving the lines on the canvas to look like an image described by the prompt, according to CLIP encoding (Radford et al., 2021).

Image of the Reframer Interface

Step by Step Installation

In order to run Reframer application you will need to run both the client and the server. The server works on CPU although it is much better to use Reframer with a suitable graphics card (GPU). The following steps should automatically install the dependencies but you may need to tweak dependencies depending on your machine/GPU.

For setup in the dwail computers...

Prerequisites

gcc

Check version is 8 or lower

gcc --version

If not,

sudo apt install build-essential
sudo apt -y install gcc-8
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 8
sudo update-alternatives --config gcc

curl

sudo apt install curl

Anaconda

curl -O https://repo.anaconda.com/archive/Anaconda3-2023.03-Linux-x86_64.sh
bash Anaconda3-2023.03-Linux-x86_64.sh

Chrome

download Google Chrome from https://www.google.com/chrome/

cd <download foder path>
sudo apt install ./google-chrome-stable_current_amd64.deb

Clone this repo and go to this branch

git clone https://github.com/Tomas-Lawton/Reframer.git
cd Reframer/

Installation

starting fom Reframer folder

cd /server
source setup.sh

Usage

starting fom Reframer folder

conda activate aidraw
cd /server/src
python3 main.py

in another terminal

cd /web
google-chrome index.html

About

Research project based on vector drawing with a text-to-image generative AI model called CLIP. View a demo video here: https://vimeo.com/760319552

https://tomas-lawton.github.io/Reframer/


Languages

Language:Python 62.7%Language:JavaScript 22.4%Language:CSS 9.5%Language:HTML 5.1%Language:Shell 0.2%Language:Makefile 0.1%Language:Procfile 0.0%