High-Level Project Summary
We developed a web application where our users can generate images related to the Nasa and the space using AI . Our challenge was to create an app that generate images using ai and make sharing the result simplest therefore we did as we were told . We used ai to generate images and to keep it under the theme of la NASA we used a nasa dataset "NASA's Picture Dictionary" to control the user input . We had the idea of adding another AI using NLP that gives suggestion to our user if his input doesn't contain any related word to NASA .We think that this project was important because it allows the person to dream and witness it within an image and to share it with friends .
Link to Final Project
Link to Project "Demo"
Detailed Project Description
Solution :
Our app have an input where the user can write anything in it related to the nasa and space , the backend process the input and gives an image as an output .
in the process part , we have two cases : the input in related to nasa and space or it isn't related . If our input is related our backend will automaticly generate an image . If it isn't , we will use NLP to give a suggestion to the user (it will contain his input and some other related words in it ) .
For the NLP part :
LSTM text generation by words. Used to generate multiple sentence suggestions based on the input words or a sentence : the model will generate the probabilities for all the words which are present in our vocabulary and then we choose the highest probability and the index of that probability will give us the word and generate sentence as suggestion.
https://colab.research.google.com/drive/1LbSxcy1WC0I4soeESEi6CJ19cSjRMy6#scrollTo=UrLzCtvA-TGQ
For the image generation part : Diffusion model (stable diffusion)
General diffusion models are machine systems that are trained to denoise step by step, to get to a sample of interest, such as an image.
https://colab.research.google.com/drive/1R7WLgb7RAqGrfe_aKch01oWEn0Y3iG1b?usp=sharing
Stable diffusion is based on a particular type of diffusion model
called Latent Diffusion,in latent diffusion the model is trained to generate latent (compressed) representation of the images .
For the sharing part :
We added some functions as dowloading the image , and sharing it thought facebook and twitter .
Benefits:
Our app is user friendly , simple to use and make the sharing on social media and downloading easy . It also generate hight quality images .
Expectations :
We want to create a community of our website . This is why we are thinking of adding a blog where our users could easily share their creations with one another .
Tools :
- We used figma to design our website
- Visual studio code to code .
- html , css js for our front end .
- django (python) for the backend.
- Stable diffusion model for the image generation .
- Lstm for our nlp model .
- Replicate api to host our model and images .
- Facebook and twitter api for the sharing .
- for the hardware we used a cpu and nvidia geoforce .
Space Agency Data
We use : NASA's Picture Dictionary
We used it to filter our user input in order for it to be always related to the nasa or space ...
we also used it in our NLP model training .
Hackathon Journey
it was so nice doing the hackathon with Mc , we met many people here . The enivrement was motivating us to work harder . We assisted many workshops as problem solving , therefore we tried implementing those notions in our solution .
Tags
#art, #ai, #space, #drawing, #generating, #stable diffusion, #lstm, #nlp, #search engine