stabilityai/stable-diffusion-xl-Turbo-1.0-torrent
💡
Original seeder @aitracker
Overview
The SD-XL 1.0 base model is a diffusion-based text-to-image generative model developed by Stability AI. It can be used as a standalone module or as part of a two-stage pipeline with a refinement model.
Model Description
- Type: Latent Diffusion Model
- License: CreativeML Open RAIL++-M License
- Resources: GitHub Repository, SDXL Report on arXiv
Pipeline
The model can be used in two ways:
- Base Model Only: Generate latents using the base model, which can be further processed with a refinement model.
- Two-Stage Pipeline: Use the base model to generate latents, then apply SDEdit (img2img) to the latents using a high-resolution model.
Evaluation
The SDXL base model performs significantly better than previous variants, and the model combined with the refinement module achieves the best overall performance.
Usage
- Diffusers: Upgrade to >= 0.19.0 and install required packages (transformers, safetensors, accelerate, invisible_watermark).
- Code Examples: Provided for using the base model and the whole base + refiner pipeline.
- Optimum: Compatible with OpenVINO and ONNX Runtime.
Uses and Limitations
- Direct Use: Research purposes only, including generation of artworks, educational tools, and probing limitations and biases.
- Out-of-Scope Use: Not intended for generating factual or true representations of people or events.
- Limitations: Does not achieve perfect photorealism, struggles with compositionality, and has issues with faces and people.
- Bias: Can reinforce or exacerbate social biases.