GitHub radar
Boogu-Image: An Open Image Generator trying to Eat the Closed Ones!
Boogu-Image is an open source(Apache-2.0), PyTorch based family of image generation and image editing models. It has high commercial grade quality, but requires much smaller amount of data to train. There are three versions available: Base, Turbo(4 step) and Edit.
Boogu-Image is an open source, Apache-2.0 family of image generation and image editing models written in PyTorch. It contains three modules: Base (focus on diversity and readability of generated text), Turbo (a 4-step process for generating high-quality photos) and Edit (image-to-image translation). It supports both English and Chinese languages. To demonstrate the ELO rating of Boogu-Image, we launched the ‘Boogu Arena’. Its ELO rating is comparable to those of commercial products. The biggest advantage of Boogu-Image is that it offers almost the same performance as closed-source models but requires less data to train. Within the first week, it attracted nearly 400 stars.
Why a vibe-coder should care
Most of the time, we can only get access to high-quality image generation models via paid APIs. We cannot afford it and we cannot use it freely. If there were some open source models that we could download and run ourselves, we wouldn’t have to pay per image and we could use our own data to train the model and customize it. In this way, we might be able to easily incorporate a decent image generation component into our workflow/application.
▌ More finds