MooMooE-comerceSDXL ACG Playground Hyper

MooMooE-comerceSDXL ACG Playground Hyper

Hyper accelerated version now available! The accelerated version offers both higher image quality and faster processing speed.

Built-in Accelerator: Hyper-SDXL-8steps-lora.safetensors

Enhanced NIJI-style anime weights, better aligning with the model's ACG (Anime, Comics, and Games) positioning. However, compared to anime-specific base models, MooMooE-comerceSDXL ACG Playground Hyper still excels at producing high-quality 2K images with intricate details and complex textures.

This branch model has been trained to improve 3D scene and character rendering effects. It also increases the weight of bright colors while reducing noise and decreasing over-fitting.

Recommended image generation settings:

Clip Skip: 1

CFG: 1

Steps: 10

Sampler: Euler a

VAE: Auto-select

Image size: 640 x 1280, with 1.5x high-resolution upscaling, 0.5 redraw strength, and Latent redraw method. Due to the sufficient final composition size, ADetailer restoration is not necessary.

Other recommended output resolutions:

Size: 1280x720 / 1536x768 / 960x1920 / 1920x960 / 2048x1024 / 1024x2048

MooMoo Model Usage Statement:

You may not host this model or its derivatives (such as merged model versions) on websites or applications intended to generate income or donations.

You may not directly sell this model or its derivatives (such as merged model versions) unless you have made substantial manual modifications to the extent that it can be legally considered your original work. If you violate this condition, you will bear all legal consequences, and the author will not be held responsible.

This model must not be used to intentionally create or share illegal or harmful content. Please adhere to public order and morality, using this model for positive purposes.

For output sizes exceeding 1536, it's recommended to use the Kohya HRFix plugin to correct structural errors and enhance upscaling. You can also enable the DynamicThresholding (CFG-Fix) plugin, although the difference may be minimal. This model has also been optimized for better compatibility with LoRA and ControlNet.