Stable Diffusion 3

Stable Diffusion 3 (SD3) was proposed in Scaling Rectified Flow Transformers for High-Resolution Image Synthesis by Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Muller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, Dustin Podell, Tim Dockhorn, Zion English, Kyle Lacey, Alex Goodwin, Yannik Marek, and Robin Rombach.

The abstract from the paper is:

Diffusion models create data from noise by inverting the forward paths of data towards noise and have emerged as a powerful generative modeling technique for high-dimensional, perceptual data such as images and videos. Rectified flow is a recent generative model formulation that connects data and noise in a straight line. Despite its better theoretical properties and conceptual simplicity, it is not yet decisively established as standard practice. In this work, we improve existing noise sampling techniques for training rectified flow models by biasing them towards perceptually relevant scales. Through a large-scale study, we demonstrate the superior performance of this approach compared to established diffusion formulations for high-resolution text-to-image synthesis. Additionally, we present a novel transformer-based architecture for text-to-image generation that uses separate weights for the two modalities and enables a bidirectional flow of information between image and text tokens, improving text comprehension typography, and human preference ratings. We demonstrate that this architecture follows predictable scaling trends and correlates lower validation loss to improved text-to-image synthesis as measured by various metrics and human evaluations.

Usage Example

As the model is gated, before using it with diffusers you first need to go to the Stable Diffusion 3 Medium Hugging Face page, fill in the form and accept the gate. Once you are in, you need to login so that your system knows you’ve accepted the gate.

Use the command below to log in:

huggingface-cli login

The SD3 pipeline uses three text encoders to generate an image. Model offloading is necessary in order for it to run on most commodity hardware. Please use the torch.float16 data type for additional memory savings.

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
pipe.to("cuda")

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
).images[0]

image.save("sd3_hello_world.png")

Note: Stable Diffusion 3.5 can also be run using the SD3 pipeline, and all mentioned optimizations and techniques apply to it as well. In total there are three official models in the SD3 family:

stabilityai/stable-diffusion-3-medium-diffusers
stabilityai/stable-diffusion-3.5-large
stabilityai/stable-diffusion-3.5-large-turbo

Image Prompting with IP-Adapters

An IP-Adapter lets you prompt SD3 with images, in addition to the text prompt. This is especially useful when describing complex concepts that are difficult to articulate through text alone and you have reference images. To load and use an IP-Adapter, you need:

image_encoder: Pre-trained vision model used to obtain image features, usually a CLIP image encoder.
feature_extractor: Image processor that prepares the input image for the chosen image_encoder.
ip_adapter_id: Checkpoint containing parameters of image cross attention layers and image projection.

IP-Adapters are trained for a specific model architecture, so they also work in finetuned variations of the base model. You can use the [~SD3IPAdapterMixin.set_ip_adapter_scale] function to adjust how strongly the output aligns with the image prompt. The higher the value, the more closely the model follows the image prompt. A default value of 0.5 is typically a good balance, ensuring the model considers both the text and image prompts equally.

import torch
from PIL import Image

from diffusers import StableDiffusion3Pipeline
from transformers import SiglipVisionModel, SiglipImageProcessor

image_encoder_id = "google/siglip-so400m-patch14-384"
ip_adapter_id = "InstantX/SD3.5-Large-IP-Adapter"

feature_extractor = SiglipImageProcessor.from_pretrained(
    image_encoder_id,
    torch_dtype=torch.float16
)
image_encoder = SiglipVisionModel.from_pretrained(
    image_encoder_id,
    torch_dtype=torch.float16
).to("cuda")

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    torch_dtype=torch.float16,
    feature_extractor=feature_extractor,
    image_encoder=image_encoder,
).to("cuda")

pipe.load_ip_adapter(ip_adapter_id, weight_name="ip-adapter.bin")
pipe.set_ip_adapter_scale(0.6)

ref_img = Image.open("image.jpg").convert('RGB')

image = pipe(
    width=1024,
    height=1024,
    prompt="a cat",
    negative_prompt="lowres, low quality, worst quality",
    num_inference_steps=24,
    guidance_scale=5.0,
    ip_adapter_image=ref_img
).images[0]

image.save("result.jpg")

IP-Adapter examples with prompt "a cat"

Check out IP-Adapter to learn more about how IP-Adapters work.

Memory Optimisations for SD3

SD3 uses three text encoders, one of which is the very large T5-XXL model. This makes it challenging to run the model on GPUs with less than 24GB of VRAM, even when using fp16 precision. The following section outlines a few memory optimizations in Diffusers that make it easier to run SD3 on low resource hardware.

Running Inference with Model Offloading

The most basic memory optimization available in Diffusers allows you to offload the components of the model to CPU during inference in order to save memory, while seeing a slight increase in inference latency. Model offloading will only move a model component onto the GPU when it needs to be executed, while keeping the remaining components on the CPU.

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained("stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16)
pipe.enable_model_cpu_offload()

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
).images[0]

image.save("sd3_hello_world.png")

Dropping the T5 Text Encoder during Inference

Removing the memory-intensive 4.7B parameter T5-XXL text encoder during inference can significantly decrease the memory requirements for SD3 with only a slight loss in performance.

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    text_encoder_3=None,
    tokenizer_3=None,
    torch_dtype=torch.float16
)
pipe.to("cuda")

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
).images[0]

image.save("sd3_hello_world-no-T5.png")

Using a Quantized Version of the T5 Text Encoder

We can leverage the bitsandbytes library to load and quantize the T5-XXL text encoder to 8-bit precision. This allows you to keep using all three text encoders while only slightly impacting performance.

First install the bitsandbytes library.

pip install bitsandbytes

Then load the T5-XXL model using the BitsAndBytesConfig.

import torch
from diffusers import StableDiffusion3Pipeline
from transformers import T5EncoderModel, BitsAndBytesConfig

quantization_config = BitsAndBytesConfig(load_in_8bit=True)

model_id = "stabilityai/stable-diffusion-3-medium-diffusers"
text_encoder = T5EncoderModel.from_pretrained(
    model_id,
    subfolder="text_encoder_3",
    quantization_config=quantization_config,
)
pipe = StableDiffusion3Pipeline.from_pretrained(
    model_id,
    text_encoder_3=text_encoder,
    device_map="balanced",
    torch_dtype=torch.float16
)

image = pipe(
    prompt="a photo of a cat holding a sign that says hello world",
    negative_prompt="",
    num_inference_steps=28,
    height=1024,
    width=1024,
    guidance_scale=7.0,
).images[0]

image.save("sd3_hello_world-8bit-T5.png")

You can find the end-to-end script here.

Performance Optimizations for SD3

Using Torch Compile to Speed Up Inference

Using compiled components in the SD3 pipeline can speed up inference by as much as 4X. The following code snippet demonstrates how to compile the Transformer and VAE components of the SD3 pipeline.

import torch
from diffusers import StableDiffusion3Pipeline

torch.set_float32_matmul_precision("high")

torch._inductor.config.conv_1x1_as_mm = True
torch._inductor.config.coordinate_descent_tuning = True
torch._inductor.config.epilogue_fusion = False
torch._inductor.config.coordinate_descent_check_all_directions = True

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers",
    torch_dtype=torch.float16
).to("cuda")
pipe.set_progress_bar_config(disable=True)

pipe.transformer.to(memory_format=torch.channels_last)
pipe.vae.to(memory_format=torch.channels_last)

pipe.transformer = torch.compile(pipe.transformer, mode="max-autotune", fullgraph=True)
pipe.vae.decode = torch.compile(pipe.vae.decode, mode="max-autotune", fullgraph=True)

# Warm Up
prompt = "a photo of a cat holding a sign that says hello world"
for _ in range(3):
    _ = pipe(prompt=prompt, generator=torch.manual_seed(1))

# Run Inference
image = pipe(prompt=prompt, generator=torch.manual_seed(1)).images[0]
image.save("sd3_hello_world.png")

Check out the full script here.

Quantization

Quantization helps reduce the memory requirements of very large models by storing model weights in a lower precision data type. However, quantization may have varying impact on video quality depending on the video model.

Refer to the Quantization overview to learn more about supported quantization backends and selecting a quantization backend that supports your use case. The example below demonstrates how to load a quantized [StableDiffusion3Pipeline] for inference with bitsandbytes.

import torch
from diffusers import BitsAndBytesConfig as DiffusersBitsAndBytesConfig, SD3Transformer2DModel, StableDiffusion3Pipeline
from transformers import BitsAndBytesConfig as BitsAndBytesConfig, T5EncoderModel

quant_config = BitsAndBytesConfig(load_in_8bit=True)
text_encoder_8bit = T5EncoderModel.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    subfolder="text_encoder_3",
    quantization_config=quant_config,
    torch_dtype=torch.float16,
)

quant_config = DiffusersBitsAndBytesConfig(load_in_8bit=True)
transformer_8bit = SD3Transformer2DModel.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    subfolder="transformer",
    quantization_config=quant_config,
    torch_dtype=torch.float16,
)

pipeline = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    text_encoder=text_encoder_8bit,
    transformer=transformer_8bit,
    torch_dtype=torch.float16,
    device_map="balanced",
)

prompt = "a tiny astronaut hatching from an egg on the moon"
image = pipeline(prompt, num_inference_steps=28, guidance_scale=7.0).images[0]
image.save("sd3.png")

Using Long Prompts with the T5 Text Encoder

By default, the T5 Text Encoder prompt uses a maximum sequence length of 256. This can be adjusted by setting the max_sequence_length to accept fewer or more tokens. Keep in mind that longer sequences require additional resources and result in longer generation times, such as during batch inference.

prompt = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. It features the distinctive, bulky body shape of a hippo. However, instead of the usual grey skin, the creature’s body resembles a golden-brown, crispy waffle fresh off the griddle. The skin is textured with the familiar grid pattern of a waffle, each square filled with a glistening sheen of syrup. The environment combines the natural habitat of a hippo with elements of a breakfast table setting, a river of warm, melted butter, with oversized utensils or plates peeking out from the lush, pancake-like foliage in the background, a towering pepper mill standing in for a tree.  As the sun rises in this fantastical world, it casts a warm, buttery glow over the scene. The creature, content in its butter river, lets out a yawn. Nearby, a flock of birds take flight"

image = pipe(
    prompt=prompt,
    negative_prompt="",
    num_inference_steps=28,
    guidance_scale=4.5,
    max_sequence_length=512,
).images[0]

Sending a different prompt to the T5 Text Encoder

You can send a different prompt to the CLIP Text Encoders and the T5 Text Encoder to prevent the prompt from being truncated by the CLIP Text Encoders and to improve generation.

The prompt with the CLIP Text Encoders is still truncated to the 77 token limit.

prompt = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. A river of warm, melted butter, pancake-like foliage in the background, a towering pepper mill standing in for a tree."

prompt_3 = "A whimsical and creative image depicting a hybrid creature that is a mix of a waffle and a hippopotamus, basking in a river of melted butter amidst a breakfast-themed landscape. It features the distinctive, bulky body shape of a hippo. However, instead of the usual grey skin, the creature’s body resembles a golden-brown, crispy waffle fresh off the griddle. The skin is textured with the familiar grid pattern of a waffle, each square filled with a glistening sheen of syrup. The environment combines the natural habitat of a hippo with elements of a breakfast table setting, a river of warm, melted butter, with oversized utensils or plates peeking out from the lush, pancake-like foliage in the background, a towering pepper mill standing in for a tree.  As the sun rises in this fantastical world, it casts a warm, buttery glow over the scene. The creature, content in its butter river, lets out a yawn. Nearby, a flock of birds take flight"

image = pipe(
    prompt=prompt,
    prompt_3=prompt_3,
    negative_prompt="",
    num_inference_steps=28,
    guidance_scale=4.5,
    max_sequence_length=512,
).images[0]

Tiny AutoEncoder for Stable Diffusion 3

Tiny AutoEncoder for Stable Diffusion (TAESD3) is a tiny distilled version of Stable Diffusion 3's VAE by Ollin Boer Bohan that can decode [StableDiffusion3Pipeline] latents almost instantly.

To use with Stable Diffusion 3:

import torch
from diffusers import StableDiffusion3Pipeline, AutoencoderTiny

pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3-medium-diffusers", torch_dtype=torch.float16
)
pipe.vae = AutoencoderTiny.from_pretrained("madebyollin/taesd3", torch_dtype=torch.float16)
pipe = pipe.to("cuda")

prompt = "slice of delicious New York-style berry cheesecake"
image = pipe(prompt, num_inference_steps=25).images[0]
image.save("cheesecake.png")

Loading the original checkpoints via `from_single_file`

The SD3Transformer2DModel and StableDiffusion3Pipeline classes support loading the original checkpoints via the from_single_file method. This method allows you to load the original checkpoint files that were used to train the models.

Loading the original checkpoints for the `SD3Transformer2DModel`

from diffusers import SD3Transformer2DModel

model = SD3Transformer2DModel.from_single_file("https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium.safetensors")

Loading the single checkpoint for the `StableDiffusion3Pipeline`

Loading the single file checkpoint without T5

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium_incl_clips.safetensors",
    torch_dtype=torch.float16,
    text_encoder_3=None
)
pipe.enable_model_cpu_offload()

image = pipe("a picture of a cat holding a sign that says hello world").images[0]
image.save('sd3-single-file.png')

Loading the single file checkpoint with T5

Tip

The following example loads a checkpoint stored in a 8-bit floating point format which requires PyTorch 2.3 or later.

import torch
from diffusers import StableDiffusion3Pipeline

pipe = StableDiffusion3Pipeline.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3-medium/blob/main/sd3_medium_incl_clips_t5xxlfp8.safetensors",
    torch_dtype=torch.float16,
)
pipe.enable_model_cpu_offload()

image = pipe("a picture of a cat holding a sign that says hello world").images[0]
image.save('sd3-single-file-t5-fp8.png')

Loading the single file checkpoint for the Stable Diffusion 3.5 Transformer Model

import torch
from diffusers import SD3Transformer2DModel, StableDiffusion3Pipeline

transformer = SD3Transformer2DModel.from_single_file(
    "https://huggingface.co/stabilityai/stable-diffusion-3.5-large-turbo/blob/main/sd3.5_large.safetensors",
    torch_dtype=torch.bfloat16,
)
pipe = StableDiffusion3Pipeline.from_pretrained(
    "stabilityai/stable-diffusion-3.5-large",
    transformer=transformer,
    torch_dtype=torch.bfloat16,
)
pipe.enable_model_cpu_offload()
image = pipe("a cat holding a sign that says hello world").images[0]
image.save("sd35.png")