【Detective Conan】Conan's mid-term drawing style is retro.

Selected images from the middle period of Detective Conan as the training set, with a slightly retro style.

Based on 万象熔炉 | Anything V5 model trained, tested on 动漫AnyLoRA with high fidelity.

_________________________________________________________________________________________________________________

Some generation suggestions:

Sampling method: Euler A or DPM++ 2M KARRAS

Resolution: 512*768

Recommended LORa weight: 0.8-1.0

Recommended base model: Anything V5, Anime AnyLoRA

_________________________________________________________________________________________________________________

To prevent everyone from not being able to generate desired images, here are some suggestions:

How to generate a good image at the initial stage of creating:

(1) Firstly, you need to identify an excellent seed. Under conditions of not changing generation information, iteration steps, generation size, and prompt word guiding coefficient, quickly generate images to find a seed with good results for easy future generation.

(2) Have you turned on high-resolution enhancement (Hires. fix) and selected the correct upscaling algorithm? For 2D, the R-ESRGAN 4x + Anime6B effect is good. Also, high-resolution enhancement is not suitable for too high amplitudes, recommended 0.2-0.6.

(3) Have you enabled the versatile After Detailer (ADetailer)? After generating the image, give the face and hands an advanced spa treatment.

In addition, a suitable base model is the foundation for a perfect LORa performance.

Additional:

Testing images on some 2D models.