After integrating several major video models such as Luma, Keling, HaiLuo, and Runway, Krea AI has now introduced the Lora training feature.
Regardless of your computer's configuration, as long as you upload more than three images, you can easily train your own characters, styles, products, etc. (ps: Although it says more than three images are sufficient, the more images you have, the better the training results will be.)
The website interface is also very simple, with no excessive buttons. After uploading the images, you will enter the basic parameter settings for image training.
If you are mainly training painting styles, choose the first style. If it's for character roles, choose the third one.
I directly downloaded a few images from Little Red Book for testing. For example, the unified feature of this style is that the 3D girls all have flowers on their heads, so I set the trigger word to "flower girl".
Other configurations can be left as default. For Batch Size, if your computer performance is poor, choose 1; if it's good, choose 4; for medium performance, choose 2. It's designed to cater to everyone. Of course, the larger the number, the faster the training, but less powerful computers will have to take it slow.
Simply set it up and click the training button. Although the official provides an invitation code for direct access to the experience, there is no free trial opportunity (so I didn't succeed). If you want to save time and train a model, just go for the premium option (those who have trained, please share your results in the comments, I'd love to see them).
Looking at the training effects of many bloggers on X, the overall results are quite good, being very user-friendly for beginners.
Currently, Krea AI's membership costs $10 per month and includes the ability to generate 720 Flux images and 36,000 real-time images, with commercial use support. If you want an official invitation code, feel free to comment below!
Product entry: https://www.krea.ai/train
