Article posted on November 21, 2022, discussing the new DreamBooth API from Replicate that allows users to train and deploy Stable Diffusion models customized to specific objects or styles.
DreamBooth can train models with as little as three images in under 30 minutes and has been updated for better results with SDXL as of August 2023.
The API is cloud-based, costs around $2.50 per training, and provides a simple guide for training a personalized DreamBooth model by using a unique identifier and generative training.
Post-training, the model is kept private by default and users can run their models via the Replicate website or API, with the option to make them public.
DreamBooth initially supports Stable Diffusion 1.5 model and includes options for different versions like SD 2.1-base, with information on how to find other versions and access community help via Discord.

Replicate allows users to execute and refine open-source AI models, as well as deploy custom models, with a single code line. It supports a variety of generative AI models, including image, language, and music generation, with a significant emphasis on scalability and simple integration. Developers can personalize models with their data for specific use cases. Replicate facilitates deployment using Cog, their tool for containerizing ML models, automating API server creation, and handling cloud deployment. The platform scales with demand, charging only for the computational time used, and caters to businesses looking to integrate AI into their products rapidly.

Replicate

Run, manage, and scale machine learning models in the cloud.

Replicate Launches DreamBooth Model Training API

Recent Product News