Stable Diffusion 3.5 is an artificial intelligence

Our strongest models to date, Stable Diffusion 3.5, are being released today. Numerous versions of this open release are available for use under the permissive Stability AI Community License, are configurable, and run on consumer hardware. Now available on GitHub are Hugging Face’s Stable Diffusion 3.5 Large and Stable Diffusion 3.5 Large Turbo models as well as the inference code.

We launched the first open release of the Stable Diffusion 3 series, Stable Diffusion 3 Medium, in June. Neither our standards nor the expectations of our communities were entirely met by this release. We took the time to further develop a version that furthers our mission to transform visual media after hearing the insightful community feedback.

Stable Diffusion 3.5 is a reflection of our dedication to providing builders and creators with state-of-the-art, widely available, and, for the majority of use cases, free tools. Regardless of whether the work is fine-tuning, LoRA, optimizations, applications, or artwork, we support its distribution and monetization throughout the pipeline.

what’s going to be issued.

A range of models created to satisfy the requirements of scientific researchers, enthusiasts, startups, and businesses are available from Stable Diffusion 3.5.

The most potent base model in the Stable Diffusion family, Stable Diffusion 3.5 Large, has 8 billion parameters, excellent quality, and fast adherence. With a 1 megapixel resolution, this model is perfect for professional use cases.

In just 4 steps, Stable Diffusion 3.5 Large Turbo—a condensed version of Stable Diffusion 3.5 Large—produces excellent photos with remarkable prompt adherence, significantly outperforming Stable Diffusion 3.5 Large.

Stable Diffusion 3.5 Medium (to be released on October 29th): This model is made to function “out of the box” on consumer hardware, balancing quality and ease of customization at 2.5 billion parameters with enhanced MMDiT-X architecture and training techniques. It can produce pictures with resolutions ranging from 0 to 25 megapixels.

Making the models.

We gave customizability top priority when creating the models in order to provide a versatile foundation on which to build. In order to do this, we incorporated Query-Key Normalization into the transformer blocks, which stabilized the model training procedure and made additional development and fine-tuning easier.

This degree of downstream flexibility required some compromises on our part. It is possible for the same prompt to produce more varied outputs from different seeds, which is deliberate because it keeps the base models’ diverse styles and larger knowledge base intact. But as a result, prompts that aren’t specific enough could cause more ambiguity in the final product, and the aesthetic quality could change.

We specifically modified the architecture and training protocols for the Medium model in order to improve its quality, coherence, and multi-resolution generation capabilities.

where the models perform exceptionally well.

While maintaining excellent performance in terms of prompt adherence and image quality, the Stable Diffusion 3.5 version is one of the most adaptable and easily accessible image models available.

Customizability: You can easily adapt the model to your unique creative requirements or create applications using workflows that you’ve customized.

Effective Performance: Designed to operate smoothly on common consumer hardware, particularly the Stable Diffusion 3.5 Medium and Stable Diffusion 3.5 Large Turbo models.

Diverse Outputs: Without requiring a lot of prompting, it produces images that are representative of the world, not just one particular kind of person, with a range of skin tones and facial features.

A quick look at the Stability AI Community license.

We are happy to make this model available under our community license, which is permissive. These are the main elements of the license.

For non-commercial purposes, including scientific research, the model is freely available to individuals and organizations.

Free for commercial use (up to $1M in annual revenue): As long as their combined yearly revenue is less than $1M, startups, small to medium-sized enterprises, and creators are permitted to use the model for commercial purposes without paying anything.

Ownership of outputs: Maintain ownership of the produced media without any restrictions on licensing.

To learn more about an Enterprise License, please get in touch with us here if your company makes over $1 million a year.

The models are available in more ways.

While Hugging Face currently offers the model weights for self-hosting, the following platforms also offer access to the model.

AI stability API.

Do the same.

Comforty.

DeepInfra.

our dedication to security.

We firmly believe in responsible, safe AI practices, and we take proactive steps to guarantee Integrity from the very beginning. That means we have taken and will continue to take appropriate measures to stop bad actors from abusing Stable Diffusion 3.5. Please visit our Stable Safety page for additional details about our safety philosophy.

Soon.

We will make Stable Diffusion 3.5 Medium available to the public on October 29. ControlNets will also debut soon, offering sophisticated control capabilities for a broad range of professional use cases.

Your thoughts on Stable Diffusion 3.5 and your creations using the models are eagerly anticipated. This form allows you to communicate with us directly.

Join our Discord Community and follow us on Instagram, LinkedIn, X, and X to keep track of our progress.