Browse Models
Launch a dedicated cloud GPU server running Laboratory OS to download and run ControlNet SD 1.5 Line Art using any compatible app or framework.
Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on local system resources, particularly GPU(s) and available VRAM.
Generate images and videos using a powerful low-level workflow graph builder - the fastest, most flexible, and most advanced visual generation UI.
Forge is a platform built on top of Stable Diffusion WebUI to make development easier, optimize resource management, speed up inference, and study experimental features.
Automatic1111's legendary web UI for Stable Diffusion, the most comprehensive and full-featured AI image generation application in existence.
Train your own LoRAs and finetunes for Stable Diffusion and Flux using this popular GUI for the Kohya trainers.
ControlNet SD 1.5 Line Art is a deep learning model that conditions Stable Diffusion 1.5 image generation using line art inputs, enabling precise structural control over generated images. Developed by lllyasviel as part of the ControlNet 1.1 release, it accepts both extracted line art from photographs and manual sketches to guide image synthesis while maintaining semantic adherence to text prompts.
ControlNet SD 1.5 Line Art is a deep learning model designed to provide fine-grained control over Stable Diffusion 1.5 image generation via line art inputs. Developed as part of the ControlNet 1.1 release, it extends the ControlNet architecture to enable precise conditioning of generative outputs based on structural cues from line drawings. This approach enables detailed and reproducible modification of generated images, making it especially applicable to tasks requiring alignment with user-provided sketches or extracted outlines.
Diagram explaining the Standard ControlNet Naming Rules (SCNNRs) used for ControlNet 1.1 models, detailing each segment of the model filename.
ControlNet SD 1.5 Line Art is part of the ControlNet 1.1 release, which includes fourteen models designed for a variety of conditioning modalities such as depth, normals, canny edges, and more. All 1.1 models retain the neural network structure first established in ControlNet 1.0, adhering strictly to the architecture to maintain backward compatibility and reproducibility. The ControlNet 1.1 documentation provides a detailed breakdown of the naming conventions and architecture, exemplified by the model filename control_v11p_sd15_lineart.pth and its associated configuration file.
The Line Art variant is specially trained to accept both automatically extracted line art from existing images and user-created manual sketches. This allows for flexible guidance, using either precise contours or more expressive, rough outlines.
ControlNet SD 1.5 Line Art enhances image generation by enforcing structural fidelity to provided line art constraints. The model is capable of handling a range of input complexity, from sparse, manually sketched outlines to intricate line extractions. Using Stable Diffusion 1.5 as its generative backbone, the model preserves the compositional intent specified by line art while adhering to the semantic and aesthetic guidance described in natural language prompts.
Demonstration of image generation guided by extracted line art of a backpack, using the prompt 'bag'. The right panel shows five model outputs conditioned on the lineart.
Diverse input types are supported:
Preprocessor algorithms, such as those for line detection or edge mapping, can be toggled to match the user's needs, ensuring that structural elements from the input are accurately reflected in the output.
Example generation where a photo of a woman is transformed via coarse lineart extraction and the prompt 'Michael Jackson's concert', producing portraits aligned structurally to the original.
ControlNet SD 1.5 Line Art is trained using the awacke1/Image-to-Line-Drawings dataset, which includes a mixture of automatically extracted line drawings and hand-drawn samples. Data augmentation strategies were applied to increase robustness, permitting the model to generalize across both artificial and organic line art styles.
Improvements over earlier versions stem from refining training protocols, such as de-duplicating grayscale samples to reduce bias, filtering out low-quality or artifact-laden images, and correcting misalignment in text-image prompt pairings. These optimizations, outlined in the 1.1 release notes, have contributed to increased reliability and accuracy in conditioned image generation tasks.
A notable variant within the ControlNet line art family is the ControlNet 1.1 Anime Lineart model, optimized for generation from anime-style line drawings. This version employs longer prompt sequences and alternate tokenization strategies, enhancing its ability to synthesize images in stylized, illustrative domains. Training details for the anime variant are not publicly described but it is engineered to leverage the unique properties of anime line art inputs.
Within the broader ControlNet 1.1 ecosystem, a series of models support other conditioning signals, including depth maps, normal maps, canny edge detections, soft edges, segmentation masks, pose estimation, scribbles, and more, each with their dedicated pre- and post-processing pathways as described in the model documentation. For context, this demonstrates the modularity and extensibility of the ControlNet design.
The principal application of ControlNet SD 1.5 Line Art is the generation of images that precisely adhere to the compositional constraints set by line drawings. This is particularly suitable for image-based content creation pipelines where user-provided or algorithmically extracted line art serves as a blueprint or guide for generating illustrations, concept art, or visualizations.
Generation results from a manually drawn wolf head sketch and the prompt 'wolf', illustrating flexibility in responding to hand-drawn structural guides.
In creative settings, the model facilitates:
The anime-oriented variant expands these capabilities to stylized illustration workflows. Model controls, such as prompt length and application of LoRA-based fine-tuning, allow for further refinements suited to specific artistic goals.
Outputs generated from an anime character line art, prompted with '1girl, saber, at night, sword, green eyes, golden hair, stocking', demonstrating the specialization of the Anime Lineart model.
While ControlNet SD 1.5 Line Art and its anime-specific counterpart offer robust control, there are some constraints. For example, the anime lineart variant does not support Guess Mode and may require external files for demonstration or evaluation, as noted in the project documentation. Certain experimental models in the series, such as Instruct Pix2Pix, may require user moderation of outputs or parameter adjustments to obtain optimal results.
As of the latest available documentation, explicit license information for the SD 1.5 Line Art model and the broader ControlNet-v1-1-nightly repository is not provided. Users are advised to consult the official repository for updates regarding licensing and usage restrictions.
For further technical insights and up-to-date information, refer to the ControlNet-v1-1-nightly project page.