MistoLine: hand-drawn sketches directly into high-quality images that fit the outline of the sketch
MistoLine is an SDXL-ControlNet-based model focused on enabling flexible adaptation to different types of line drawings and high precision image generation. It is able to utilize a variety of user-supplied line drawings as input, including hand-drawn sketches, line drawings generated by different preprocessors, and contour lines automatically generated by the model.
Main Features
- Diversity line drawing adaptation:
- Hand-drawn sketches: MistoLine can directly use the user’s hand-drawn sketches as input to generate high-quality images that conform to the outline of the sketch.
- Preprocessor-generated walked: Adapts to the walked generated by different ControlNet preprocessors, such as Canny, Hough, Scribble, M-LSD, etc. The walked styles generated by the different preprocessors can be directly used as MistoLine’s input.
- Automatically generate outlines from models: Supports automatically generating lines or outlines from images, and line drawings generated from models can also be used as input.
- High quality image generation:
- Resolution: the short side of the generated image can exceed 1024 pixels and the image remains high resolution and rich in detail.
- Detail Reduction: With advanced Anyline line preprocessing algorithms and re-trained ControlNet models, MistoLine is able to preserve the intricate details of the linework and render them fully in the final output.
- Stability and consistency:
- Consistency: MistoLine is highly versatile, adapting to a wide range of line art inputs and producing image output that is consistent with the original line art.
- High Stability: The model remains highly stable in complex line drawing scenarios, avoiding the limitations of traditional ControlNet models that switch between different preprocessors.
- Innovative Architecture Design:
- Mixed Response: By combining capacitive and resistive response sensors
Hybrid Response: By combining capacitive and resistive response sensors, the model ensures the accuracy of the output image while maintaining the style of the line art. - Large-scale model training: MistoLine uses an innovative engineering approach to large-scale model training that ensures models remain consistent across different input conditions.
- Wide range of application scenarios:
- Hand-drawn sketch rendering: Users can hand-drawn sketches for high-quality image rendering, keeping the flexibility of hand-drawn lines.
- Image Stylized Translation: With line art input, you can transform the original image into different styles, suitable for art style creation and comic generation.
This model is compatible with most SDXL models, except PlaygroundV2.5 and CosXL. it works with LCM and other ControlNet models.
ComfyUI recommended parameters
sampler steps:30
CFG:7.0
sampler_name:dpmpp_2m_sde
scheduler:karras< br>denoise:0.93
controlnet_strength:1.0
stargt_percent:0.0
end_percent:0.9[img]
Chinese (Mainland China) Convenient Download Address:
Link: https://pan.baidu.com/s/ 1DbZWmGJ40Uzr3Iz9RNBG_w?pwd=8mzs
Extract code: 8mzs
Model address: https://huggingface .co/TheMistoAI/MistoLine
