|
|
--- |
|
|
license: creativeml-openrail-m |
|
|
language: |
|
|
- en |
|
|
base_model: [] |
|
|
pipeline_tag: other |
|
|
tags: |
|
|
- upscaler |
|
|
- denoiser |
|
|
- comfyui |
|
|
- automatic1111 |
|
|
datasets: [] |
|
|
metrics: [] |
|
|
--- |
|
|
|
|
|
# Model Card for MidnightRunner/ControlNet |
|
|
|
|
|
This repository provides a **ready-to-use collection of ControlNet models** for SDXL, ComfyUI, and Automatic1111. |
|
|
These models include edge detectors, pose estimators, depth mappers, lineart adapters, tilers, and experimental adapters for advanced conditioning and structure control in AI art generation. |
|
|
All models are tested, practical, and selected for reliable integration into custom creative workflows. |
|
|
|
|
|
## Model Details |
|
|
|
|
|
### Model Description |
|
|
|
|
|
A curated toolbox of ControlNet models for high-precision structure control, pose transfer, lineart extraction, depth estimation, segmentation, inpainting, recoloring, and more. |
|
|
This set enables rapid workflow iteration for generative AI artists, illustrators, and researchers seeking robust conditioning tools for SDXL-based systems. |
|
|
|
|
|
- **Developed by:** MidnightRunner and open-source contributors |
|
|
- **Model type:** ControlNet Adapters (edge, depth, pose, etc.) |
|
|
- **License:** creativeml-openrail-m |
|
|
- **Language(s) (NLP):** N/A (image processing only) |
|
|
- **Finetuned from model:** ControlNet base models, original authors noted per file |
|
|
|
|
|
### Model Sources |
|
|
|
|
|
- **Repository:** https://huggingface.co/MidnightRunner/ControlNet |
|
|
|
|
|
## Uses |
|
|
|
|
|
### Direct Use |
|
|
|
|
|
Integrate with ComfyUI, Automatic1111, SDXL workflows, and other diffusion UIs for: |
|
|
- pose-to-pose transformation |
|
|
- edge/lineart guidance |
|
|
- depth-aware rendering |
|
|
- mask-based editing, recoloring, and inpainting |
|
|
- seamless tiling and upscaling |
|
|
|
|
|
### Downstream Use |
|
|
|
|
|
May be included in chained pipelines for creative tools, batch image post-processing, or AI-driven illustration tools. |
|
|
|
|
|
### Out-of-Scope Use |
|
|
|
|
|
Not for medical imaging, biometric authentication, or other critical inference domains. |
|
|
|
|
|
## Bias, Risks, and Limitations |
|
|
|
|
|
- All models inherit the limitations and biases of their upstream datasets and architectures. |
|
|
- May produce artifacts or degrade image quality in edge cases. |
|
|
- Outputs should be reviewed in all sensitive, safety-critical, or NSFW scenarios. |
|
|
|
|
|
### Recommendations |
|
|
|
|
|
Outputs should be manually reviewed before deployment in professional or public-facing applications. |
|
|
|
|
|
## How to Get Started with the Model |
|
|
|
|
|
```bash |
|
|
git lfs install |
|
|
git clone https://huggingface.co/MidnightRunner/ControlNet |
|
|
``` |
|
|
|
|
|
# Download a single file |
|
|
huggingface-cli download MidnightRunner/ControlNet controlnetxlCNXL_xinsirOpenpose.safetensors |
|
|
|
|
|
# Python example |
|
|
```bash |
|
|
from huggingface_hub import hf_hub_download |
|
|
|
|
|
file = hf_hub_download( |
|
|
repo_id="MidnightRunner/ControlNet", |
|
|
filename="controlnetxlCNXL_xinsirOpenpose.safetensors" |
|
|
) |
|
|
``` |
|
|
# Results |
|
|
Models selected based on strongest visual fidelity and lowest artifact rate in practical SDXL workflows. |
|
|
|
|
|
# Summary |
|
|
This ControlNet toolbox provides high success rates and reliability for AI-driven image control and conditioning tasks, based on both quantitative metrics and extensive real-world user testing. |
|
|
|
|
|
# Environmental Impact |
|
|
Hardware Type: Consumer and research GPUs (NVIDIA A100, RTX 3090, Apple Silicon, etc.) |
|
|
|
|
|
Carbon Emitted: Minimal for inference; training costs depend on model size and upstream provider. |
|
|
|
|
|
# Technical Specifications |
|
|
Model Architecture and Objective |
|
|
All models follow the ControlNet architecture paradigm, adapted for specific guidance (edge, pose, depth, etc.) |
|
|
Objectives are structure preservation, fidelity, and seamless integration with diffusion image synthesis. |
|
|
|
|
|
# Compute Infrastructure |
|
|
Hardware: NVIDIA GPUs (A100, 3090, etc.), Apple M1/M2 |
|
|
|
|
|
Software: Python 3.10+, PyTorch 2.x, ComfyUI, Automatic1111, HuggingFace Hub tools |
|
|
|
|
|
# Citation |
|
|
If you use these models in your research or product, please cite the original ControlNet paper and any upstream sources referenced per file. |
|
|
|
|
|
## More Information |
|
|
For more details, licensing, or integration tips, visit https://huggingface.co/MidnightRunner/ControlNet or contact MidnightRunner via HuggingFace. |
