Character Canvas

CreatorCanvas Logo

Inspiration

AI is widely criticized for cheapening art, but how can artists use this tool to enrich their art? The answer? Using artificial intelligence to generate reference photos, serving as the basis for human art. Reference photos are a limited resource, and obtaining reference photos can be an obstacle to artists. Further, publicly available reference photos are dominated by certain demographics. How can this be rectified? By generating reference photos in the instructed pose and with the instructed features.

What it does

The user takes a photo of a 3D articulated model, such as mannequin.js, in their desired reference pose, and submits it to CharacterCanvas. The user then inputs their desired features as a prompt for the image generator. ChatGPT is then instructed to recognize the posture of the image, and describe the precise locations of each body part, which is then fed into DALL-E along with the prompt.

How we built it

Front end: Flask, Bootstrap Back end: DALL-E API

Challenges we ran into

The majority of our team was made up of first-time hackers, so isolating roles and developing a workflow took some extra time.
Utilizing Flask for the project posed a challenge, as none of our team members had any experience with it. Certain code segments did not behave as expected within the Flask environment.
Utilizing the OpenAI image recognition was difficult since we couldn't find much documentation, and there was some trouble with the API key.
We spent some time getting the virtual environment to work between the various systems we were running, Linux, Windows, Mac