Model Overview

Note: You need to request the model checkpoint and license from Stability AI

Request the model checkpoint from Stability AI

Description:

SDXL-Turbo, a distilled variant of SDXL 1.0, empowers real-time image synthesis via Adversarial Diffusion Distillation (ADD). ADD enables high-fidelity sampling from large-scale image diffusion models in 1-4 steps. Employing score distillation for leveraging teacher signals from pre-trained models, ADD seamlessly integrates an adversarial loss, guaranteeing image fidelity even at minimal sampling steps (1-2).

Developed by: Stability AI
Funded by: Stability AI
Model type: Generative image-to-video model
Finetuned from model: SDXL 1.0 Base

Terms of use

By using this software or model, you are agreeing to the terms and conditions of the license, acceptable use policy and Stability’s privacy policy.

References(s):

• Introducing SDXL Turbo - Stability.ai
• SDXL-Turbo on Huggingface
• SDXL-Turbo paper

Model Architecture:

Architecture Type: Transformer and Convolutional Neural Network (CNN)
Network Architecture: UNet + attention blocks
Model Version: SDXL Turbo

Input:

Input Format: Text, Image (Optional)
Input Parameters: inference_steps, seed, prompt_strength (for an image input)

Output:

Output Format: Red, Green, Blue (RGB) Image
Output Parameters: seed

Software Integration:

Supported Hardware Platform(s): Hopper, Ampere/Turing
Supported Operating System(s): Linux

Inference:

Engine: Triton
Test Hardware: Other

Request Model Checkpoint:

You can request the model checkpoint from Stability AI