Youkengi Anime SDXL Base

Youkengi_anime_SDXL_base is a Checkpoint in the Youkengi series that features a stable and singular art style based on Stable Diffusion XL (SDXL). It carries forward the high quality of the Youkengi models, which were at the T1 level when released.

Version 2.0 has made significant improvements across the board compared to Version 1.0, forming a well-rounded model. In various tests, it performs strongly with no obvious weaknesses.

Its core capabilities include:

1. The ability to produce images in a very stable and exquisite anime style: the basic art style is clean-lined with moderate detail. Although tags like 'realistic' and '3D rendering' still work, they carry low weight and will still have a strong anime flavor;

2. Better hands and good body poses: especially for challenging hand gestures, the model handles them without breaking down; even if there's a slight breakdown, the basic shape remains intact;

3. Strong text understanding: it can draw according to the prompt description, maintaining most of the concepts of the official SDXL base model;

4. High accuracy in drawing: the likelihood of making mistakes while drawing is greatly reduced, such as being able to get example images right in one go;

5. Ability to draw men (not feminine): this point doesn't need much explanation since most collections feature more cute girls, and male characters are relatively rare;

6. Easy to use: it uses a mixed tagging method, supporting both long sentences and short phrases well;

7. Not face-locked and strong LoRA combination capability: LoRAs trained under this and the original official model have better compatibility. Due to the severe divergence of various branches of the XL model, the effect of LoRAs trained under other branches cannot be guaranteed.

8. Details are not easily messy, hair does not tangle easily: this mainly makes the viewing experience more pleasant, reducing the AI feel to some extent.

Regarding usage:

1. VAE has been merged: the current VAE has slightly higher saturation and contrast than the default SDXL VAE. If you choose another VAE, it can override the built-in VAE effect of the model;

2. Sampling methods are not picky; using common ones works fine. Testing was done in the DPM++3M SDE Karras environment, where this sampler performed best.

3. It is recommended to use high-resolution fix (hires.fix) for upscaling, with a denoising strength of about 0.4. The only scaling algorithm not recommended is latent, as it can blur the image. A CFG of 7 is recommended.

4. The recommended base steps are 25-50, and the high-resolution fix upscaling steps are 10-20;

5. Recommended resolutions: 864*1152*2, 1152*864*2, 1024*1024*2, 1152*1152*2. Minor changes are acceptable, but do not use SD1.5 resolutions, as all training images were high-definition, and a too small base resolution can lead to various strange limb errors.

6. General negative prompts: worstquality, lowquality. If you want to call upon certain styles inherent to the official model, it is suggested to use an empty negative prompt.

My other models:

Youke youkengi_realistic(fantasy)SDXL https://www.liblib.art/modelinfo/12fb2246263d46f880829fe4320e89f2

Youke youkengi_anime_SDXL_base https://www.liblib.art/uploadmodel/4a340e48e0d34d698d0d54d57715b631

Youke youkengi cute:https://www.liblib.ai/modelinfo/ac8fb054451245778b5ea3452104c731

Youke youkengi chibi:https://www.liblib.art/modelinfo/33253de83fcb4465a63cd3c421723cf4?mine=1

Youke youkengi 3D:https://www.liblib.ai/modelinfo/ad6942b5acbc4301a2a2b30a2b4591ae?mine=1

Youke youkengi delicate:https://www.liblib.ai/modelinfo/cdb5d856b52d49329a481f552d48c4e6?mine=1

Youke youkengi anything:https://www.liblib.ai/modelinfo/b0466fd9edd34ea7a67dab412db47fb3

Youke youkengi clear:https://www.liblib.ai/modelinfo/e5be47b63e5b4408be94942374ebd57c?mine=1

Youke youkengi delicate if: https://www.liblib.ai/modelinfo/19c85fcbdf424520b0fb70613da88ff6?mine=1