Stable Diffusion v1-5

Stable Diffusion v1-5

Generates images from text prompts with a latent diffusion model of 860M parameters, usable in diffusers, ComfyUI and other tools. Suits researchers and developers who want the classic open Stable Diffusion checkpoint.

Made by stable-diffusion-v1-5 on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Stable Diffusion v1-5 turns text prompts into images, and it can modify images too. Robin Rombach and Patrick Esser developed it as a latent diffusion model. The checkpoint started from the v1-2 weights and was fine-tuned for 595 thousand steps at 512 by 512 pixels. This repository is a mirror of a deprecated one and has no tie to RunwayML.

Who it suits: Researchers, developers and artists who want the classic open Stable Diffusion checkpoint, under the CreativeML OpenRAIL M license. The card says the model is intended for research.

Industry | Who uses it | How they use it | Example
Research | ML researchers | Study diffusion models and their limits | Run controlled experiments on prompts
Software | Developers | Build image generation into an app with diffusers | Generate an image from a text field
Creative work | Artists | Make images in a local interface such as ComfyUI | Sketch ideas from a prompt
Education | Students | Learn how text-to-image models work | Compare outputs from different prompts

Skip it if you need the newest image quality, since this is an older model, or if your project cannot accept the license terms.

Where to use it: The card lists several tools that run this checkpoint.

• The Diffusers library in Python
• ComfyUI
• AUTOMATIC1111
• SD.Next
• InvokeAI

Use case | Who | What they get back
Text-to-image | Artists | An image from a written prompt
Research runs | Researchers | Samples for studying model behavior
Fine-tuning | Developers | A base for your own tuned checkpoint

How to set it up: You need Python and a GPU for comfortable speed.

• Python with pip
• The diffusers library and torch for the Python route
• About 3.4 GB of space for the weights in F32

1. Install the Hugging Face hub tools with pip.
2. Download the model files with the hf command.
3. Load them in diffusers, or point ComfyUI, AUTOMATIC1111, SD.Next or InvokeAI at the checkpoint.

pip install -U huggingface_hub
hf download stable-diffusion-v1-5/stable-diffusion-v1-5

How to use it: In Python, load the pipeline, move it to the GPU, and call it with a prompt. The card shows this.

from diffusers import StableDiffusionPipeline
import torch

model_id = "sd-legacy/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")

The card also offers two single-file checkpoints for the other tools. One holds only the EMA weights, uses less memory and suits inference. The other holds both kinds of weights and suits fine-tuning.

Tool | What it does
StableDiffusionPipeline | Loads the model and generates images in diffusers
pipe.to cuda | Moves the pipeline to an NVIDIA GPU
v1-5-pruned-emaonly.safetensors | EMA-only weights for inference with less VRAM
v1-5-pruned.safetensors | EMA and non-EMA weights for fine-tuning

Example: You pass the prompt "a watercolor painting of a lighthouse at dusk" to the pipeline. It returns an image object that you save to a png file and open.

For newcomers and experts: If you are new to image models, run the Python snippet once with a simple prompt and look at the result. Then change a few words in the prompt and see how the image shifts.

If you know the field, it is a latent diffusion model with a fixed, pretrained CLIP ViT-L/14 text encoder. The card says text conditioning was dropped for 10 percent of training examples to improve classifier-free guidance sampling. Training data was laion-aesthetics v2 5+.

Download the model files
Load in diffusers or a local interface
Write a text prompt
Run the diffusion steps
Save the generated image

Good to know: The model has 860 million parameters. It is a community mirror, so check which source you want to depend on.

• License: CreativeML OpenRAIL M
• Cost to run: none named, apart from your own hardware
• Data it reaches: the model files on your computer, and nothing else the card names
• Not for: shipping a product before you read the license and the research-use note

On Hugging Face, 13 November 2026

• Task: text to image
• Library: diffusers
• Parameters: 860M
• Weights: about 3.4 GB in F32; about 430 MB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 1,012,982
• Likes: 1,364
• License: creativeml-openrail-m
• Quantized versions: 17
• Fine-tunes: 421
• Tags: diffusers, safetensors, stable-diffusion, stable-diffusion-diffusers, text-to-image

Links

• Model: https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5
• Files: https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download stable-diffusion-v1-5/stable-diffusion-v1-5

Then follow the usage in its model card. Nothing in this kit runs by itself.