Real Vis_ Medium_1.0b

Introduction

RealVis Medium is a fine-tuned model that generates high-quality realistic and photorealistic images. It is based on SD3.5 Medium and is currently under ongoing development to enhance its capabilities.

Architecture

RealVis Medium employs a sophisticated architecture that utilizes various components for text-to-image generation. It is fine-tuned for generating detailed and realistic imagery.

Training

The model has been trained with a focus on achieving realistic outputs. It features a different training data distribution compared to larger models, which might result in varied responses to the same prompts. The model is optimized to handle long prompts but may exhibit artifacts when exceeding 256 T5 tokens.

Guide: Running Locally

1. Download Required Files:

   • From the text_encoders folder, place clip_g.safetensors, clip_l.safetensors, t5xxl_fp16.safetensors, or t5xxl_fp8_e4m3fn.safetensors in ComfyUI\models\clip.
   • From the transformer folder, place diffusion_pytorch_model.safetensors in ComfyUI\models\unet.
   • From the vae folder, place diffusion_pytorch_model.safetensors in ComfyUI\models\vae.

2. Use Provided Workflows:

   • Utilize the RealVis Medium 1.0b Basic Workflow or the Basic Workflow with Skip Layer Guidance available in the repository.

3. Hardware Recommendation:

   • For optimal performance, consider using cloud GPUs from providers like AWS or Google Cloud Platform.

License

The model is available under the stabilityai-ai-community license. For more details, refer to the license document.