We’re pleased to announce that our version 5.4 is here! And now you can access the checkpoints through this link. For last couple of days, we’ve been through a lot of technical difficulties, but we could survive. In this article, we’ll explain what did we bring to the table with our new checkpoints.
A new dataset
For pretty much every new version, we find and create a better dataset. In this particular version, pretty much all images have been made using midjourney and then formed into a dataset.
Our datasets are usually a set of images with their relative text file which includes the prompt. In this case, prompts were human-made (yes, because the human mind is always superior to ChatGPT!) and for some tests, we also got help from image-labeling models such as BLIP in order to find more stable diffusion compatible prompts.
A new license
For our future releases, we’ve considered a new licensing system. This decision was because we needed a recovery from what happened last week and we were on the edge of failure.
So as soon as we reach version 6 of Mann-E models, we will switch back to our original MIT license.
The version 5.4 actually has a nice support of LoRa models and basically every “style” on our platform is linked to a LoRa model. We’ve trained and tested different LoRa’s and cherry picked the best ones for you.
Pretty much everything we’ve implemented in the model layer has been integrated to the platform. A few minor changes are also made which are as following:
- Added more details handle which controls “more details” LoRa model and can have a good effect on the resulting image.
- Error handling is done for authentication.
So you can now easily go ahead and boost your creativity using Mann-E AI platform. We hope you enjoy it!