Reduce Time and Cost with Generative AI-based Character Image Creator

The client, the operator of a live chat-streaming application, wanted their avatar characters to have daily outfit changes – something human artists cannot do. That’s why they looked to Rikkeisoft for a solution.

About the client

Artificial Intelligence

Entertainment

This article describes a real-life project. However, we cannot disclose our client’s name for privacy purposes.

The client is a Japanese company focusing on the operation of live chat applications, the development of information services, and the operation and management of information network-related communication systems. The client’s newest addition to their portfolio is a live chat-streaming application that allows users to talk to streamers with Anime-styled avatars.

Project Overview

Industry

Entertainment

Technology

PyTorch, Stable Diffusion, Lambda, Gradio

Country

Japan

Duration

3 months

Challenges

As mentioned above, the client company’s latest service embraces the use of anime-styled avatars for its streamers. Currently, these avatars are hand-drawn by external artists hired by the client.

Art takes time

As the application develops, the client wants to have these avatars change their outfits daily for a touch of realism. Unfortunately, this is too much to ask from human artists, as they cannot finish drawings of the avatars’ new outfits at the speed required by the client company. Art takes time to be perfected, and the daily requirements put a huge strain on the artists.

Artists are expensive and inconsistent

The client company also cannot hire new artists due to the high costs attached to the process. Good artists come with a high price, and looking for good artists is a labor-extensive task in itself. Furthermore, artists have different styles, and having multiple artists working on the same character’s outfits will inevitably result in stylistic inconsistencies in the final products.

Solution

Stable Diffusion-based Character Image Creator

To address the client company’s challenges, Rikkei AI developed an AI model based on Stable Diffusion to help them quickly and easily create new outfits for their virtual streamer avatars. To begin with, the client would need to train the model with preexisting images of the avatars, with each avatar requiring 10 images for the best final results. The model would then pre-process these input images to determine the avatar’s gender before adding them to its database. It then accesses Stable Diffusion’s existing database of gender-corresponding images to train and ready itself for the user’s input prompts. After training, the user can now input a prompt of an outfit as well as the final image’s aspect ratio into the AI model. After a short time, an image of the avatar in the desired outfit would be created, which human artists can further refine for use in the live chat-streaming service.

0 %

reduction in the number of artists needed

0 mins

required to train the AI model

0 s

required for image creation

Result

The client is highly satisfied with the AI solution, as it has helped them reduce the time and cost it would take to create avatars and outfits traditionally. It shows that while human artists are still needed, Artificial Intelligence can be a great assistant to them if correctly used in the future. Rikkei AI’s capabilities in generative AI development are also on display, as we have managed to deliver the ideal solution quickly with high user satisfaction.