Have you ever wondered what it would take to create manga-style art using AI? I recently stumbled upon a fascinating experiment that did just that, and I wanted to dive deeper into the process.
The experiment, conducted by Reddit user Nyao, involved training an AI model using a dataset of 44 images from renowned manga artist Naoki Urasawa’s works, including ‘Monster’, ’20th Century Boys’, and ‘Pluto’. The goal was to generate images that mimicked Urasawa’s unique style.
The Training Process
The model used was Qwen Lora, a type of text-to-image model that can generate images based on a given prompt. The training process involved 2750 steps with a learning rate of 0.0002, using the Ostris Toolkit. You can find more details about the process in Ostris’ tweet and YouTube video.
The Results
The generated images are impressive, to say the least. While the likeness of the style may not be perfect, it’s clear that the AI model has captured the essence of Urasawa’s work. What’s more impressive is the prompt adherence of Qwen, which suggests that the model is capable of understanding and recreating complex styles.
The Future of AI-Generated Art
Experiments like this one raise interesting questions about the future of art and creativity. Can AI-generated art ever truly replace human artists? Or will it simply become a tool to augment and enhance human creativity?
One thing is certain, though – the possibilities are endless, and it will be exciting to see where this technology takes us.
Further Reading
If you’re interested in learning more about AI-generated art and the Qwen Lora model, check out the links below:
* Qwen-Image-Lightning by lightx2v
* Civitai model by Nyao