Stable Diffusion V1 2

🚀 Stable Diffusion v1-2 Model Card

Stable Diffusion is a latent text-to-image diffusion model that can generate photo-realistic images from any text input. For more details on how it works, check out 🤗's Stable Diffusion with D🧨iffusers blog.

The Stable-Diffusion-v1-2 checkpoint was initialized with the weights of Stable-Diffusion-v1-1 and then fine-tuned for 515,000 steps at a resolution of 512x512 on "laion-improved-aesthetics" (a subset of laion2B-en, filtered to images with an original size >= 512x512, an estimated aesthetics score > 5.0, and an estimated watermark probability < 0.5). For more info, refer to Training.

These weights are meant to be used with the D🧨iffusers library. If you're looking for weights to load into the CompVis Stable Diffusion codebase, click here.

✨ Features

• Generate photo-realistic images from text input.
• Can be fine-tuned on specific datasets.
• Compatible with the D🧨iffusers library.

📦 Installation

We recommend using 🤗's Diffusers library to run Stable Diffusion. Install the necessary packages with the following command:

pip install --upgrade diffusers transformers scipy

💻 Usage Examples

Basic Usage

Running the pipeline with the default PNDM scheduler:

import torch
from torch import autocast
from diffusers import StableDiffusionPipeline

model_id = "CompVis/stable-diffusion-v1-2"
device = "cuda"


pipe = StableDiffusionPipeline.from_pretrained(model_id)
pipe = pipe.to(device)

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt)["sample"][0]  
    
image.save("astronaut_rides_horse.png")

Advanced Usage

Limited GPU Memory

If you have less than 10GB of GPU RAM, load the StableDiffusionPipeline in float16 precision:

import torch

pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to(device)

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["sample"][0]  
    
image.save("astronaut_rides_horse.png")

Swap Noise Scheduler

To swap out the noise scheduler, pass it to from_pretrained:

from diffusers import StableDiffusionPipeline, LMSDiscreteScheduler

model_id = "CompVis/stable-diffusion-v1-2"
# Use the K-LMS scheduler here instead
scheduler = LMSDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
pipe = StableDiffusionPipeline.from_pretrained(model_id, scheduler=scheduler, use_auth_token=True)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
with autocast("cuda"):
    image = pipe(prompt, guidance_scale=7.5)["sample"][0]  
    
image.save("astronaut_rides_horse.png")

📚 Documentation

Model Details

Uses

Direct Use

The model is for research purposes only. Possible research areas and tasks include:

• Safe deployment of models with 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.
• Applications in educational or creative tools.
• Research on generative models.

Misuse, Malicious Use, and Out-of-Scope Use

⚠️ Important Note

This section is taken from the DALLE-MINI model card and applies to Stable Diffusion v1 as well.

The model should not be used to intentionally create or spread images that create hostile or alienating environments for people. This includes generating disturbing, distressing, or offensive images, or content that propagates historical or current stereotypes.

Out-of-Scope Use

The model was not trained to provide factual or true representations of people or events, so using it to generate such content is beyond its capabilities.

Misuse and Malicious Use

Using the model to generate cruel content towards individuals is a misuse. This includes, but is not limited to:

• Generating demeaning, dehumanizing, or harmful representations of people, their environments, cultures, religions, etc.
• Intentionally promoting or propagating discriminatory content or harmful stereotypes.
• Impersonating individuals without consent.
• Sexual content without the consent of those who might see it.
• Mis- and disinformation.
• Representations of egregious violence and gore.
• Sharing copyrighted or licensed material in violation of its terms of use.
• Sharing altered copyrighted or licensed material in violation of its terms of use.

Limitations and Bias

Limitations

• The model doesn't achieve perfect photorealism.
• It can't render legible text.
• It performs poorly on more complex tasks involving compositionality, like rendering an image for “A red cube on top of a blue sphere”.
• Faces and people may not be generated correctly.
• It was mainly trained with English captions and works less well with other languages.
• The autoencoding part of the model is lossy.
• It was trained on the large-scale dataset LAION-5B, which contains adult material and isn't suitable for product use without additional safety measures.
• No additional measures were used to deduplicate the dataset, resulting in some memorization of duplicated training images. You can search the training data at https://rom1504.github.io/clip-retrieval/ to detect memorized images.

Bias

Although image generation models are impressive, they can reinforce or worsen social biases. Stable Diffusion v1 was trained on subsets of LAION-2B(en), which mainly consists of images with English descriptions. Texts and images from non-English-speaking communities and cultures are likely underrepresented, affecting the model's overall output as white and western cultures are often the default. Also, the model's ability to generate content with non-English prompts is significantly worse than with English prompts.

Training

Training Data

The model was trained on the following dataset:

• LAION-2B (en) and its subsets (see next section).

Training Procedure

Stable Diffusion v1-4 is a latent diffusion model that combines an autoencoder with a diffusion model trained in the autoencoder's latent space. During training:

• Images are encoded by an encoder into latent representations. The autoencoder uses a relative downsampling factor of 8 and maps images of shape H x W x 3 to latents of shape H/f x W/f x 4.
• Text prompts are encoded by a ViT-L/14 text-encoder.
• The non-pooled output of the text encoder is fed into the UNet backbone of the latent diffusion model via cross-attention.
• The loss is a reconstruction objective between the noise added to the latent and the prediction made by the UNet.

We currently offer four checkpoints, trained as follows:

• stable-diffusion-v1-1: 237,000 steps at a resolution of 256x256 on laion2B-en, then 194,000 steps at a resolution of 512x512 on laion-high-resolution (170M examples from LAION-5B with a resolution >= 1024x1024).
• stable-diffusion-v1-2: Resumed from stable-diffusion-v1-1, 515,000 steps at a resolution of 512x512 on "laion-improved-aesthetics" (a subset of laion2B-en, filtered as described above).
• stable-diffusion-v1-3: Resumed from stable-diffusion-v1-2, 195,000 steps at a resolution of 512x512 on "laion-improved-aesthetics" with a 10% drop in text-conditioning to improve classifier-free guidance sampling.
• stable-diffusion-v1-4: Resumed from stable-diffusion-v1-2, 225,000 steps at a resolution of 512x512 on "laion-aesthetics v2 5+" with a 10% drop in text-conditioning to improve classifier-free guidance sampling.

Training details

• Hardware: 32 x 8 x A100 GPUs
• Optimizer: AdamW
• Gradient Accumulations: 2
• Batch: 32 x 8 x 2 x 4 = 2048
• Learning rate: Warmed up to 0.0001 for 10,000 steps and then kept constant.

Evaluation Results

Evaluations with different classifier-free guidance scales (1.5, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0) and 50 PLMS sampling steps show the relative improvements of the checkpoints:

Evaluated using 50 PLMS steps and 10000 random prompts from the COCO2017 validation set at a resolution of 512x512. Not optimized for FID scores.

Environmental Impact

Stable Diffusion v1 Estimated Emissions Based on this information, we estimate the following CO2 emissions using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

🔧 Technical Details

• The model combines an autoencoder with a diffusion model trained in the latent space of the autoencoder.
• It uses a ViT-L/14 text-encoder to encode text prompts.
• Cross-attention is used to feed the text encoder's output into the UNet backbone.
• The loss is a reconstruction objective between the added noise and the UNet's prediction.

📄 License

This model is open access and available to all under the CreativeML OpenRAIL-M license. The license specifies:

1. You can't use the model to deliberately produce or share illegal or harmful outputs or content.
2. The authors claim no rights on the outputs you generate. You're free to use them but accountable for their use, which must not violate the license provisions.
3. You may redistribute the weights and use the model commercially or as a service. If you do, include the same use restrictions as in the license and share a copy of the CreativeML OpenRAIL-M with all your users (read the license carefully).

Please read the full license here.