Reduce e-commerce costs: Traditional clothing promotions require hiring models, photographers, and renting studios, which can be expensive. With Magic Wardrobe, you can generate realistic model images at a fraction of the cost, saving up to 99.9% on traditional photography expenses.

Generate model diversity: Choose elements such as age, ethnicity, and background to create diverse, realistic models with just one click.

Increase conversion efficiency: Traditional photography requires significant time and a complex workflow to produce high-quality images. With Magic Wardrobe, you can generate various realistic models with just one click, saving 99% of the time and workflow costs for e-commerce.

This product can recognize the text in the image and output it through the image, realizing the automation process of image to text conversion, greatly reducing the time that enterprises use to manually enter text and improving work efficiency.

This product uses deep learning algorithms to train AI models, and the data uses 1,000,000+ Japanese text data, including Japanese supermarket product lists, invoices, tickets, and other data, as well as handwritten data, and self-developed synthetic data.

The inference speed in the test data is only 0.084 seconds, while the text recognition accuracy is as high as 96.15%, exceeding the recognition rate of any Japanese OCR system on the market.

In addition to standard Japanese print, it also recognizes Japanese handwriting, English, and other characters.

This product is being piloted in cooperation with Tsinghua Tongfang at a number of accounting firms in Japan.

This product is an online Japanese open-domain dialog bot.

This product collects and modifies billions of high-quality Japanese conversation corpus using Japanese TV subtitles, Japanese news, Japanese wiki and other Japanese text media.

Self-developed language model, using advanced pre-training, fine-tuning training and other techniques, using 8 A100 GPUs, trained for 2 months to generate a conversation model with high quality Japanese.

Train a context coherence scoring model for generating Japanese dialog, and for generating dialog coherence, the scores are sorted to be with optimal utterances to make the context more coherent.

This product has been well received and promoted by KDDI, a major Japanese communications company.

This product is an online Japanese knowledge-based content dialog bot.

This product collects and generates hundreds of thousands of high-quality Japanese knowledge conversation corpus by using Japanese text media such as Japanese wiki and manual production.

Based on the open-domain dialog model, a knowledge comprehension module has been added, which is a way for the model to fully understand the use and selection of knowledge during the training process.

Using 8 A100 GPUs, trained for 2 months, a dialog model with high-quality Japanese knowledge was generated.

This product has been well received and promoted by KDDI, a major Japanese communications company.