Stable Diffusion is a tool that uses generative AI to create images from text prompts. It is based on latent diffusion models and CLIP text encoders, which allow it to produce high-quality and diverse images at various resolutions, but there are many versions of Stable Diffusion that comes out regularly. Below we will talk about the latest Stable Diffusion releases as they become available.

What is the Latest Stable Diffusion?

Stable Diffusion has gone through several versions, each with different features and improvements. The latest Stable Diffusion is Stable Diffusion 3, which was announced on Thursday, February 22, 2024. In this blog post, we will review the evolution of Stable Diffusion and includes the recently announced Stable Cascade, a similar tool that uses a novel three-stage approach to generate images.

---

Stable Diffusion Releases

Stability AI has developed several image generators that use generative AI to create images from text prompts. The most well-known of these is Stable Diffusion, which has gone through various versions and improvements. The latest one is Stable Diffusion 3, which was announced on February 22, 2024. Another image generator that Stability AI has released is Stable Cascade, which uses a novel three-stage approach to generate images more efficiently and cheaply than Stable Diffusion. Stable Cascade is also a text-to-image model, so I consider it as part of the Stable Diffusion family. Both Stable Diffusion and Stable Cascade can be used by popular web UIs such as Clipdrop and Diffusers.

Stable Video 3D

Announced March 18th, Stable Video 3D (SV3D) is a generative model developed by Stability AI that advances the field of 3D technology. It’s based on Stable Video Diffusion and offers significant improvements in quality and view consistency for 3D generation from single images. Here are some key features of SV3D:

• Novel View Synthesis: SV3D can take a single object image and generate novel multi-views of that object, which can then be used to create 3D meshes.
• Two Variants: The model comes in two variants, SV3D_u and SV3D_p. SV3D_u generates orbital videos from single images without camera conditioning, while SV3D_p can create 3D video along specified camera paths using both single images and orbital views.
• Commercial and Non-Commercial Use: SV3D is available for commercial use with a Stability AI Membership. For non-commercial purposes, the model weights can be downloaded from Hugging Face.
• 3D Neural Radiance Fields (NeRF): It leverages multi-view consistency to optimize 3D NeRF and mesh representations, improving the quality of 3D meshes generated directly from novel views.

For more technical details and to view the research paper, you can visit the official release announcement or access the model on Hugging Face

Learn more at Stability AI | Learn more on Huggingface | Read more about SD3

Stable Diffusion 3

Stable Diffusion 3 is the latest version of Stable Diffusion, which was announced on Thursday, February 22, 2024. According to its makers, Stability AI, the new model improves the quality and diversity of the generated images, especially in handling text with multiple subjects, spelling errors, and complex scenes¹²³. Stable Diffusion 3 is currently in early preview and you can sign up for the waitlist here. The Stable Diffusion 3 suite of models currently range from 800M to 8B parameters, offering users a variety of options for scalability and quality to best meet their creative needs.

Learn More

Stable Cascade

Stable Cascade was released in research preview on February 12, 2024. Stable Cascade is another tool that uses generative AI to create images from text prompts. It is developed by Stability AI, the same company behind Stable Diffusion. Stable Cascade is different from Stable Diffusion in that it uses a novel three-stage approach to generate images, hence the name “Stable Cascade”. The first stage is a VAE that compresses an image to a small latent space. The second stage is a diffusion model that refines the latent space and adds more details.

The third stage is another diffusion model that generates the final image from the latent space and the text prompt. Stable Cascade achieves a higher compression factor than Stable Diffusion, meaning that it can encode a 1024×1024 image to 24×24, while maintaining crisp reconstructions. This makes Stable Cascade more efficient and cheaper to train and run than Stable Diffusion. Stable Cascade also has impressive results, both visually and evaluation wise. According to Stability AI, Stable Cascade performs best in both prompt alignment and aesthetic quality in almost all comparisons. Stable Cascade is also available in early preview and you can sign up for the waitlist here.

Learn More

SDXL Turbo

SDXL Turbo is a new text-to-image model based on a novel distillation technique called Adversarial Diffusion Distillation (ADD), which enables the model to generate image outputs in a single step and generate real-time text-to-image outputs while maintaining high sampling fidelity. SDXL Turbo is a distilled version of SDXL 1.0, a large-scale text-to-image model that could generate images at 1024×1024 resolution. SDXL Turbo was released on November 28, 2023 and is available for free download and testing on Stability AI’s image editing platform ClipDrop.

Learn More

Stable Diffusion XL 1.0

Stable Diffusion XL was a special version of Stable Diffusion, with a larger model size and a longer training time. It could generate images at 1024×1024 resolution, and had the highest quality and diversity among all the versions. It also had the best performance in terms of prompt alignment and aesthetic quality, according to human evaluations. Stable Diffusion XL was a state-of-the-art text-to-image model, that showcased the power and beauty of generative AI. Stable Diffusion XL was released on July 26, 2023.

Learn More

Stable Diffusion XL 0.9

Despite its ability to be run on a modern consumer GPU, SDXL 0.9 presents a leap in creative use cases for generative AI imagery. The ability to generate hyper-realistic creations for films, television, music, and instructional videos and offer advancements for design and industrial use places SDXL at the forefront of real-world applications for AI imagery. 

Learn More

Stable Diffusion 2.1

Stable Diffusion 2.1 was a fine-tuned version of Stable Diffusion 2.0, with a less restrictive NSFW filtering of the training dataset. It could generate images at 768×768 resolution, and had more realistic and diverse results than the previous version. It also reduced some of the artifacts and noise that were present in Stable Diffusion 2.0, and improved the alignment and coherence of the generated images. Stable Diffusion 2.1 was released on December 22, 2022.

Learn More

Stable Diffusion 2.0

Stable Diffusion 2.0 was a major update of Stable Diffusion, with a new architecture and text encoder based on OpenCLIP-ViT. It could generate images at 768×768 resolution, and introduced new features such as v-prediction, depth-guided synthesis, and text-guided inpainting. V-prediction allowed the model to generate multiple images for the same prompt, by varying the latent vector. Depth-guided synthesis enabled the model to generate images with realistic depth and perspective, by using a depth map as an intermediate representation.

Text-guided inpainting allowed the model to fill in missing parts of an image, by using the text prompt as a guide. Stable Diffusion 2.0 was a significant improvement over the previous versions, as it increased the fidelity and diversity of the generated images, and added more control and flexibility for the users. Stable Diffusion 2.0 was released on July 15, 2022.

Learn More

Stable Diffusion 1.5

Stable Diffusion 1.5 was an improved version of Stable Diffusion 1.4, with more parameters and a larger training dataset. It could generate images at 512×512 resolution, and had better quality and diversity than the previous version. It also introduced a NSFW filter, which prevented the model from generating inappropriate or offensive images. Stable Diffusion 1.5 was widely used and praised by the community, as it demonstrated the potential of generative AI for creative and artistic purposes. Stable Diffusion 1.5 was released on November, 2022.

Learn More

Stable Diffusion 1.4

Stable Diffusion 1.4 was the first public release of Stable Diffusion, which could generate images at 256×256 resolution. It had some limitations in handling complex scenes and diverse concepts, and sometimes produced blurry or distorted images. However, it was still a remarkable achievement for text-to-image generation, as it could handle a wide range of prompts and generate realistic and coherent images. Stable Diffusion 1.4 was released on May 1, 2022.

Learn More

Conclusion

Stable Diffusion and Stable Cascade are two amazing tools that use generative AI to create images from text prompts. They have both evolved and improved over time, and offer different features and advantages for the users. Whether you prefer Stable Diffusion’s high-quality and diverse images, or Stable Cascade’s efficient and novel approach, you can’t go wrong with either of them. They are both examples of how generative AI can unleash your creativity and imagination, and help you create stunning and original images.

---

Understanding Stable Diffusion:

• What is Stable Diffusion?
• Is Stable Diffusion real?
• What do steps do in Stable Diffusion?
• What is the latest Stable Diffusion?
• How many people use Stable Diffusion?
• Is Stable Diffusion easy to use?
• What is similar to Stable Diffusion?
• What is Stable Diffusion style?
• How does Stable Diffusion training work?
• Is Stable Diffusion Pretrained?
• Can Stable Diffusion generate 3D models?
• How do you use Stable Diffusion at home?
• What is the new version of Stable Diffusion?
• What are the applications of Stable Diffusion?