Hey everyone, a new model just dropped that I think is worth exploring: ERNIE-Image. It is a surprisingly capable text-to-image model from Baidu, and it is already showing strong results compared to some of the bigger open models.
What stood out to me right away is how well it handles text and structured layouts. Things like posters, thumbnails, comics, and infographic-style images come out much more accurate than we are used to seeing from most open models.
OSIRIS TOOLKIT has also added Day 1 support for ERNIE-Image, which means you can get started training models right away if that is part of your workflow or hobby.
Below is the Get Going Fast installer, which will download the models for you and provide the workflow. The workflow is powerful, but it can be a little confusing at first since you need to switch a few nodes on or off depending on which model version you are using.
That said, the model itself is very well put together and easy to use once you get going. You should be in good shape.
FREE for all FARM HAND Members.