AI yourself
Dreambooth model training + Deforum animation [Stable Diffusion]
A complete workflow of how to train your own text-to-image and text-to-video AI generation model
Dreambooth model training + Deforum animation [Stable Diffusion]
A complete workflow of how to train your own text-to-image and text-to-video AI generation model
This project was a result of my passion for generative art and the desire to explore the possibilities offered by Stable Diffusion.
Not being a coding expert, I faced initial challenges until I teamed up with the experienced developer Jaime Domènech, whose expertise brought this project to life, resulting in something truly extraordinary.
Not being a coding expert, I faced initial challenges until I teamed up with the experienced developer Jaime Domènech, whose expertise brought this project to life, resulting in something truly extraordinary.
1. Stable Diffusion Installation:
2. Training with DreamBooth
Next, we took about 20 pictures of ourselves, focusing on facial expressions and angles. Using DreamBooth and Google Colab, we fine-tuned the Stable Diffusion model with our faces. DreamBooth, a magical photo booth-like program, learns from real pictures to create new images in various scenarios. Find out more about DreamBooth here and Google Colab here. The training process proved to be the most challenging yet rewarding phase.
3. Loading the DreamBooth model into Automatic1111
With the DreamBooth model trained, we seamlessly integrated it with Stable Diffusion. This allowed us to generate entirely new pictures of ourselves based on written descriptions. The possibilities were limitless.
The creative journey was not just a project; it was a celebration!
The creative journey was not just a project; it was a celebration!
4. Prompting
This installation empowered us to create personalized AI art, bringing our imagination to life in ways we never thought possible. We enjoyed the creative process and let the artistry flow!
We experimented with various textual descriptions, each serving as a unique catalyst for image generation. Whether it was altering visual styles, changing placements, defining characters, or experimenting with clothing, the model responded dynamically to the nuances of our prompts.
There were absolutely no limitations to the transformations we could achieve.
Just like people, the model can be trained to generate not only faces but also objects, animals, or virtually anything one can imagine.
This opens up a world of creative possibilities, allowing us to push the boundaries of conventional AI art.
This opens up a world of creative possibilities, allowing us to push the boundaries of conventional AI art.
5. Deforum for Stable Diffusion
Ultimately we integrated Deforum into the workflow.
It is a user-friendly tool for crafting animated videos using Stable Diffusion.
Deforum facilitates the creation of animated videos by allowing the adjustment of camera movements and prompts without the need for programming skills. It seamlessly fited into this creative process, enabling us to enhance this project with visually engaging animation.
June 2023
