This SDXL-Turbo model has been optimized to work with WebNN. This model is licensed under the STABILITY AI COMMUNITY LICENSE AGREEMENT. For terms of use, please visit the Acceptable Use Policy. If you comply with the license and terms of use, you have the rights described therin. By using this Model, you accept the terms.
SDXL-Turbo-WebNN is meant to be used with the corresponding sample here for educational or testing purposes only.
WebNN changes
The base model is SDXL-Turbo. SDXL-Turbo-WebNN is an ONNX adaptation of the SDXL-Turbo model, specifically optimized for WebNN execution.
We utilize the OLive SDXL-Turbo pipeline as the core infrastructure for ONNX model generation and optimization. Building upon this, we have implemented several enhancements in a fork to further optimize the models for WebNN (see repo):
Int4 Quantization Support: Enables QDQ or MatMulNBits quantization, reducing the total model size from 13GB to approximately 3.5GB.
FP16 Quantization Support: Further reduces the total model size to approximately 2.5GB.
FP16 I/O Enforcement: Forces model inputs and outputs to use the FP16 data type for improved performance.
UNET Quality Preservation: Addresses image quality issues in the UNET model by introducing the --keep_unet_conv_fp32 flag, which maintains IO Conv Nodes in FP32 precision.
--use_qdq_encoding: Enables QDQ quantization (defaults to MatMulNBits if omitted).
--keep_unet_conv_fp32: Retains IO Conv Nodes in FP32 precision to resolve specific precision-related quality issues.
--force_fp16_inputs: Enforces FP16 for input/output operations.
SDXL-Turbo Model Card
SDXL-Turbo is a fast generative text-to-image model that can synthesize photorealistic images from a text prompt in a single network evaluation. A real-time demo is available here: http://clipdrop.co/stable-diffusion-turbo
SDXL-Turbo is a distilled version of SDXL 1.0, trained for real-time synthesis.
SDXL-Turbo is based on a novel training method called Adversarial Diffusion Distillation (ADD) (see the technical report), which allows sampling large-scale foundational
image diffusion models in 1 to 4 steps at high image quality.
This approach uses score distillation to leverage large-scale off-the-shelf image diffusion models as a teacher signal and combines this with an
adversarial loss to ensure high image fidelity even in the low-step regime of one or two sampling steps.
For research purposes, we recommend our generative-models Github repository (https://github.com/Stability-AI/generative-models),
which implements the most popular diffusion frameworks (both training and inference).
The charts above evaluate user preference for SDXL-Turbo over other single- and multi-step models.
SDXL-Turbo evaluated at a single step is preferred by human voters in terms of image quality and prompt following over LCM-XL evaluated at four (or fewer) steps.
In addition, we see that using four steps for SDXL-Turbo further improves performance.
For details on the user study, we refer to the research paper.
Uses
Direct Use
The model is intended for both non-commercial and commercial usage. You can use this model for non-commercial or research purposes under this license. Possible research areas and tasks include
Research on generative models.
Research on real-time applications of generative models.
Research on the impact of real-time generative models.
Safe deployment of models which have the potential to generate harmful content.
Probing and understanding the limitations and biases of generative models.
Generation of artworks and use in design and other artistic processes.
SDXL-Turbo does not make use of guidance_scale or negative_prompt, we disable it with guidance_scale=0.0.
Preferably, the model generates images of size 512x512 but higher image sizes work as well.
A single step is enough to generate high quality images.
py
1from diffusers import AutoPipelineForText2Image
2import torch
34pipe = AutoPipelineForText2Image.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16")5pipe.to("cuda")67prompt ="A cinematic shot of a baby racoon wearing an intricate italian priest robe."89image = pipe(prompt=prompt, num_inference_steps=1, guidance_scale=0.0).images[0]
Image-to-image:
When using SDXL-Turbo for image-to-image generation, make sure that num_inference_steps * strength is larger or equal
to 1. The image-to-image pipeline will run for int(num_inference_steps * strength) steps, e.g. 0.5 * 2.0 = 1 step in our example
below.
The model was not trained to be factual or true representations of people or events,
and therefore using the model to generate such content is out-of-scope for the abilities of this model.
The model should not be used in any way that violates Stability AI's Acceptable Use Policy.
Limitations and Bias
Limitations
The generated images are of a fixed resolution (512x512 pix), and the model does not achieve perfect photorealism.
The model cannot render legible text.
Faces and people in general may not be generated properly.
The autoencoding part of the model is lossy.
Recommendations
The model is intended for both non-commercial and commercial usage.
