Coloring Canvass Generator!

This is a submission for the Cloudflare AI Challenge.

What I Built

Generate a printable coloring canvass in PDF format based on

any themes, genres powered by Artificial Intelligence.

Demo

https://illah-creatives.pages.dev/

Step 1

Step 2

Step 3

Sample outputs

Promt of Cyberpunk Motorcycle hero will generated this

Prompt of Lion, Bear, Wolf and Owl will generate this.

My Code

Web App
https://github.com/MarkMatute/illah-creatives

Microservices
https://github.com/MarkMatute/illah-creatives-text-generator
https://github.com/MarkMatute/illah-creatives-image-generator
https://github.com/MarkMatute/illah-creatives-art-assistant

Journey

I am currenlty in the journey of learning and exploring Machine Learning and Artificial Intelligence when I came accross the post. With this momentum I decided to pick a simple project that me or my family can use.

With this project that can generate almost unlimited pages of coloring books with any themes you want or based on who you are, or whatever preferences you have.

With this project, I learned alot about different type of models and how to leverage them. It also exposed me from what I do not know as of the moment, like proper prompting etc. Also opened doors for me to use Cloudflare more than its normal use case. The platform itself trully is cool! I am just proud to come up with this idea that surely some people can enjoy.

Tasks types used:

1. Text-To-Image
   1. Theme to canvass cover image
   2. Each themes to coloring page image
2. Text-Generation
   1. Illah’s chat assistant that focuses on generating themes for user
   2. Cool canvass titles that matches context with generated themes
   3. Cool names for generated canvass pages

Models used:

1. @hf/mistral/mistral-7b-instruct-v0.2
2. @cf/meta/llama-2-7b-chat-fp16
3. @cf/stabilityai/stable-diffusion-xl-base-1.0

Improvements for this project:

1. Parallel generation of each canvass page for faster processing
2. More tailored chat assistant, not only for generating themes but also for more tasks.
3. To add an image to image generation of canvasses

Multiple Models and/or Triple Task Types

Models used:

1. @hf/mistral/mistral-7b-instruct-v0.2
2. @cf/meta/llama-2-7b-chat-fp16
3. @cf/stabilityai/stable-diffusion-xl-base-1.0