NMKD Stable Diffusion GUI

NMKD Stable Diffusion GUI is a free Windows desktop application that provides a comprehensive graphical interface for running Stable Diffusion locally on personal hardware, enabling text-to-image generation, image-to-image editing, and instruction-based image modification without requiring cloud services or ongoing subscriptions. It is developed by N00MKRAD and distributed through itch.io with all required dependencies included for a simplified setup experience.

Key Features

• All software dependencies are bundled in the installer, eliminating manual configuration of Python environments or machine learning frameworks. The application supports standard text-to-image generation from user-provided text prompts and image-to-image generation that combines a base image with a text prompt for guided creative workflows.
• Instruction-based image editing through InstructPix2Pix enables modifying existing images using natural language instructions. Advanced prompting features include attention and emphasis controls for weighting specific terms, negative prompts to exclude unwanted elements, and wildcard support for introducing randomized words or phrases into prompts dynamically.
• Support for custom Stable Diffusion models and custom VAE models allows integration of community-trained checkpoints. Multiple prompts can be processed in batches using a queue system, prompt history saves previous generations for reuse, and seamless tileable image generation creates patterns suitable for game textures and repeating backgrounds.
• Built-in post-processing includes RealESRGAN upscaling for increasing image resolution and CodeFormer or GFPGAN face restoration for improving facial details. The integrated image viewer displays generation metadata, supports keyboard shortcuts for efficient navigation, and includes a pop-up viewer with zoom and fullscreen modes.
• Safety features scan downloaded models for malware before loading. The integrated updater keeps the application current, and the installation manager can validate, repair, or reinstall components including Python dependencies, upscalers, and model files.

Hardware Requirements

• The minimum system requires an Nvidia GPU with 4 GB VRAM based on the Maxwell architecture from 2014 or newer, 8 GB of system RAM, and 10 GB of available disk space. The recommended configuration is an Nvidia GPU with 8 GB VRAM based on the Pascal architecture from 2016 or newer, 16 GB of RAM, and a solid-state drive.
• Limited AMD GPU support is available through DirectML, requiring 8 GB VRAM, though advanced features such as InstructPix2Pix are exclusive to Nvidia hardware. Performance scales with GPU capability, achieving under one second per 512x512 image on an RTX 4090.

Training and Customization

• Textual Inversion embeddings allow the application to learn new visual concepts from a small set of example images. LoRA support enables lightweight fine-tuning for specific characters, art styles, or subjects, and the application includes a dedicated LoRA training interface for creating custom models directly.
• Generation settings provide control over step count, CFG guidance scale, seed values, image resolution in multiples of 64, and sampler selection with DPM++ 2M Karras as the default. A high-resolution fix prevents repeated patterns at resolutions above 768 pixels.
• Developer tools include a Stable Diffusion command-line interface, model merging with configurable blend weights, model pruning to reduce file sizes, and model conversion between PyTorch, Diffusers, Diffusers ONNX, and SafeTensors formats.

Privacy

• All image generation runs entirely offline on local hardware with no external data sharing. The developer confirms no data is collected by the application or transmitted to third parties, and the software operates without censorship filters.