While image generation AI may not be in high demand at exhibitions, we are developing it with the future goal of creating a beauty app. This time, we’ve tackled the somewhat tricky LoRA (Low-Rank Adaptation). blueqat is experimentally offering image generation models. Though the quality is currently low, we are making progress. We have developed a system that allows users to easily create LoRAs by simply uploading images. This service, with limited features, is expected to be offered at around 1,000 yen per month. Users can upload images and input prompt keywords related to those images. Our models are not publicly available, so you cannot upload or download LoRA weights. However, you can create specific LoRAs for our models. The learning time is optimized and does not take long; for about 10 images, the learning process is completed in approximately 10 minutes. We aim to enhance and adjust our machines to achieve even faster learning times, improving convenience. Creating and using LoRAs can still be quite challenging, making a managed service like this relatively rare. If you want to easily upload images and enhance models, we encourage you to try our service. The beta version will be available soon. Initially, you will be able to use LoRAs created by our company as optional filters. We are currently expanding our machines for image generation, especially those needed for LoRA training due to the required specifications. We plan to gradually increase the number of users who can utilize the service. We are committed to making it easier for users to fine-tune models without needing programming skills. If you&#39;re interested in trying out our service, stay tuned for the beta release. We are enhancing our machines to provide a faster and more efficient learning experience. Thank you for your interest and support.

Development of a Fully Managed LoRA System for Easy Fine-Tuning without Programming

Yuichiro Minato