Inspiration
I wanted to create a sort of "flipbook" with AI-generated photos and, an extension of that, an AI gif generator. I felt it would look really cool to generate dynamic images with flow to capture ideas, ensuring the smoothness of all the images. I wanted to ensure the high quality of each image and the transition at a fraction of the cost of other generators.
What it does
Type in an idea, and the AI automatically generates multiple sequential prompts for the image and accompanying changes. Then, we create the images with those prompts, stitch them together, and display them!
See below!

How we built it
The project was developed in Python and primarily leverages the OpenAI API for generating creative prompts. It integrates Stable Diffusion techniques for image creation and employs a text-to-image and text+image-to-image conversion process. This is all served to a frontend with vanilla HTML/CSS/JS and Google Cloud to host gifs. The backend was created with fast API.
I generate an image from the first idea (text->image). Then, on each successive iteration, I combine the previous image with a text modification to produce the following image (text+image->image). I ensure smoothness with postprocessing and high-quality prompting to ensure each text modification is accurate and high quality.
Challenges we ran into
- Crafting high-quality text prompts that ensure a smooth flow from one image to the next without too much overlap
- Postprocessing and tuning to ensure smoothness and aesthetics of images
- Integration of the front and backend with fastAPI and google cloud
Accomplishments that we're proud of
- Built an AiGif generator!
- The amazing quality of all the gifs generated!
- High-quality performance against other products using novel methods
- Harmonizing AI and Art to create something beautiful!
What we learned
Alot!
- Prompting methods for LLMs and improving prompt engineering skills
- Intricacies of image models and how to best harness their powers
- The fundamentals of using basic HTML/CSS/JS without unnecessary modules
What's next for Ai Gif Generator
More user feedback definitely helps. Probably try fine-tuning a custom model to generate gifs as a sequential output of text->image models. Also, there are some cool new features to add, like manually changing the direction of the gif on a granular level and only animating certain parts of the gif.
Built With
- css3
- fastapi
- html5
- javascript
- openai
- python
- replicate
- stablediffusion
Log in or sign up for Devpost to join the conversation.