Mingji Zhang

Compared with SD, the Flux model has greatly enhanced the ability of semantic understanding. In AI comics, it can accurately draw the content of the picture in the generation of multiple characters, complex actions, and complex scenes. And it greatly reduces the confusion, distortion, and multiple limbs. However, compared with SD, the Flux ecosystem is immature, the generation speed is slow, and there are fewer control plug-ins, but we believe that the future will be good.

The Stable Diffusion XL version is very cost-effective. Although it lacks semantic understanding and is not very stable, its excellent image performance and rich ecosystem of creators make it very suitable for content tools. Among AI comic creation tools, SDXL is a very suitable model. We also considered Midjourney, but from the perspective of cost-effectiveness, SDXL is more suitable for us.

Dify is a tool that our backend engineers like very much because it is very flexible. Compared with coze, dify has fewer restrictions and can build more complex workflows. We use dify to build complex AI comic backend processes, so that the story, script, storyboard, picture, NUI and GUI interaction become very orderly.

Coze is an online AI low-code tool that we like very much. We use it to build some small AI comic functions. Coze is very simple to use, and our product managers can use it directly.

We used OpenAI's ChatGPT service before. To be honest, ChatGPT's ability to express AI comic stories, scripts, and storyboards is very outstanding. Its context understanding is very accurate, and even if users interact multiple times, they will not forget the background of the story. It can also maintain accurate output in the description of complex characters, shots, actions, postures, expressions, lines, and narrations. We really like to use ChatGPT to build the story part. However, due to cost and response speed issues, and the lack of workflow building function in OpenAI's API, we also use other language models to work.