Akash Karanam (ak111), Ayush Sachdeva (as216), Benjamin Thomas (bwt2), Nathan Powell (ndp3), Rambod Azarbad (ra53)
Over the last few years image generation deep learning models have become more accurate and accessible. There is a unique opportunity to generate digital art on-demand to match any query a user inputs. We propose a website where a user can enter a prompt for a sticker design they would want, and then we display a page of image results generated by an AI system like DALL-E. Furthermore, we want to enable that user to interact with the image generation AI, which could include a more hands-on approach to prompt generation, and image editing tools. Furthermore, DALL-E is trained to return images, giving us the opportunity to apply our own AI model (trained through transfer learning) to convert the images DALL-E outputs to images that are suitable for stickers.
Full Proposal link: https://drive.google.com/file/d/1GZxnnzWDUphWh31GJVrygf5UEgj5yMF7/view?usp=sharing
Model Training: https://arxiv.org/abs/2102.12092 https://www.reddit.com/r/StableDiffusion/comments/x41n87/how_to_get_images_that_dont_suck_a/
Blog on what differentiates an effective image prompt: https://www.creativebloq.com/news/how-to-use-dall-e
Style inspiration: https://www.instagram.com/openaidalle/
After you have cloned our repo into your own local Stickr folder change directories into that folder.
Before running our application, run the build file as follows from the top level of the project:
./build
From the top level of the project you can then run the web application by running
poetry run python3 application.py
After you have locally started our rigged demo, you must then open the link that it returns (which will look something like "http://127.0.0.1:5000").
Furthermore, the unit tests can be run by running
poetry run pytest