Introduction to Stable Diffusion 3.5
Stability AI has introduced Stable Diffusion 3.5, which comes with multiple model variants that cater to various users. These models are designed to run efficiently on consumer-grade hardware and are accessible for both commercial and non-commercial use under the flexible Stability AI Community License.
Features and Availability
Developers have the freedom to customize and integrate the models without the constraints of restrictive licensing, making them suitable for a wide range of applications. The Stable Diffusion 3.5 Large and Stable Diffusion 3.5 Large Turbo models can be downloaded from Hugging Face, and the inference is available on GitHub.
Stable Diffusion 3.5 offers a diverse range of models tailored to meet the needs of researchers, startups, and enterprises. The Large model, with 8 billion parameters, ensures superior image quality and quick adherence, making it ideal for professional use at a 1-megapixel resolution. On the other hand, the Large Turbo version provides a faster alternative, producing high-quality images in just 4 steps.
The models are optimized for efficient performance on standard consumer hardware, especially the Medium and Large Turbo versions. They generate inclusive and diverse images that accurately represent various skin tones and features without the need for extensive prompts.
Training and Licensing
These models are trained on a subset of the LAION-5b dataset, curated by the DeepFloyd team, and utilize the dataset's NSFW filter to further filter adult content. For non-commercial purposes, including academic research, the model is available at no cost. Startups, small to medium businesses, and creators can also leverage the model for commercial use for free, provided their annual revenue does not exceed $1M. Users retain full ownership of the generated content without any restrictive licensing.
Recent Developments in AI
On a related note, Google recently announced the suspension of its Gemini artificial intelligence image generation feature due to reported inaccuracies in historical images. The decision came after viral Gemini-generated pictures sparked controversy on social media, with users criticizing Google for prioritizing social awareness over accuracy.
Social media users raised concerns about the AI tool generating images of historical figures, such as the U.S. Founding Fathers, with inaccurate representations of people of color. In response, Google stated that they are pausing the image generation feature to enhance the accuracy of its responses.
Looking ahead, the technology landscape continues to evolve, with companies like Stability AI pushing boundaries with innovative AI models like Stable Diffusion 3.5.
