Browse Models
The simplest way to self-host epiCRealism. Launch a dedicated cloud GPU server running Lab Station OS to download and serve the model using any compatible app or framework.
Download model weights for local inference. Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on your system resources, particularly GPU(s) and available VRAM.
epiCRealism is a Stable Diffusion 1.5-based model with a custom VAE that enhances color accuracy and natural lighting. It excels at human features and textures, performing best with minimal prompting and specific parameters (Clip Skip 1, CFG 5). Works optimally at 512x768 resolution using SDE/DPM samplers.
The epiCRealism model represents a significant advancement in photorealistic image generation using Stable Diffusion technology. Built on the SD 1.5 base model, this final iteration of the epiCRealism family (as of October 2024) focuses on enhanced prompt understanding, improved hand rendering, and heightened overall realism compared to its predecessors.
The "Natural Sin RC1 VAE" variant introduces several notable improvements, particularly in the realm of photorealistic image generation. The model incorporates a specialized VAE (Variational Autoencoder) that enhances color vibrancy, though its use remains optional. This iteration demonstrates particular strength in rendering realistic human features, as evidenced by its output quality.
The model shows remarkable versatility in handling various scenarios and lighting conditions, from indoor portraits to environmental shots. However, users should note its tendency toward NSFW content generation and relatively lower sensitivity to prompt guidance compared to other models.
For best results, the creator recommends several specific parameters:
The model performs best with simple prompts, avoiding excessive detail keywords such as "masterpiece" or specific facial/hand/hair details. For achieving realistic outputs, simple negative prompts like "cartoon, painting, illustration" are recommended.
For enhanced results, the model can be used with several recommended external tools:
For high-resolution outputs, the creator specifically recommends the 4x_NMKD-Superscale-SP_178000_G upscaler with a denoising setting of 0.35 and 2x upscale factor.
The model is released under the CreativeML Open RAIL-M license with a specific addendum, ensuring proper usage rights while maintaining certain restrictions. More details can be found in the license addendum.