Stable Distribution 3.5: Stable AI redeems itself with new models and expanded features.

Serenity AI may be starting its own redemption arc. After the middling disappointment of the SD3, they're back swinging with the release of two new models that were promised in July: the Stable Transmission 3.5 Big and the Stable Transmission 3.5 Big Turbo.

In June we released Stable Diffusion 3 Medium, the first open source from the Stable Diffusion 3 series. This release does not fully meet our standards or the needs of our communities,” said Serenity in an official blog post. “After listening to valuable community feedback, rather than a quick fix, we took the time to develop a version that furthered our mission to transform visual media.”

Before we rushed to write this breaking news, we created a few images to test – and the results were pretty cool. Especially for the base model.

The SD 3.5 family is designed to run on consumer-grade systems—even lower-end ones in some cases—making advanced imaging more accessible than ever before. And yes, they've heard complaints about the previous version so this one promises to be much better: hence their featured image of a woman lying on the grass, the same thing that refers to the horror scene that happened earlier. Challenging.

Another important aspect of this release is the new licensing model. Stable Distribution 3.5 is available under a more permissive license that allows both commercial and non-commercial use. Small businesses and those making less than $1,000,000 in device revenue can use and build these models for free.

Large earners should contact Statility to negotiate payments. In comparison, Black Forest Labs' low-end Flux Schnell is free, the mid-range model Flux Dev is non-commercial, and the SOTA model Flux Pro is a closed-source model. (For reference, Flux is generally considered the best open source image generator available today—at least in the current post-SDXL era.)

What's on the table with Stable Diffusion 3.5?

Stability AI is releasing three versions of Stable Diffusion 3.5, all catering to different needs.

Stable Transmission 3.5 Large: This is the largest: 8 billion parameters are designed to provide high-level image quality and tight, fast tracking. It's designed for professional use, typically with 1-megapixel resolution, but can handle a variety of styles and viewing formats.

Stable Transmission 3.5 Big Turbo: For those looking to trade a little quality for speed, this version of the big model is your go-to. It produces high-quality images in just four steps—unlike the conventional SD3.5, which requires 30 steps to generate a good-quality image. It will be equal to Flux Schell.

Stable Distribution 3.5 Medium: In the near future, this model has 2.5 billion measurements and is optimized for user hardware. With a resolution between 0.25 and 2 megapixels, it's a middle ground for users who need solid performance without sacrificing ease of customization.

The models are more flexible, which allows users to adjust to specific creative needs. And if you're worried about whether your consumer-grade GPU can handle it, Stability AI has your back. Our own tests show the Big Turbo spitting out images in under 40 seconds on a modest RTX 2060 with 6GB of VRAM.

The full-fat version that hasn't been benchmarked requires more than 3 minutes on the same lower-end hardware, but that's the price of quality.

Improvements in the lower abdomen

Stable AI is playing with Flux, the go-to model for customization. To improve the user experience, Stability SD 3.5 has rethought how it works. “We prioritized customization to provide a flexible foundation to build the models on while developing. To achieve this, we integrated query key routines into Transformer Blocks, streamlining the model training process and simplifying further tuning and development,” said Serge.

In other words, whether you're an artist looking to create custom patterns or a developer looking to build an AI-powered app, you can edit and refine these models more easily than ever before. To help you get things started quickly, Serenity has shared a LoRA training guide.

LoRA (low-level adaptation) is a technique for adjusting models to specialize in a particular concept – style or subject – without having to retrain the entire large-based model.

Of course, with flexibility comes some trade-offs. The model is now so innovative that stability warns that “nonspecific conditions may cause uncertainty in the results and the degree of beauty may vary.”

If you're still on the fence about Stable Diffusion 3.5 and the “uncertainty” is driving you away, here's a little futureproofing for you—it supports “negative requests,” meaning your request can include instructions not to do things. This is a great advantage for those who want to refine text and images without jumping through hoops.

It's a great addition for those who want a little more control over their generation. Also, it looks great in keeping with the old SDXL's style of inspiration. In fact, in some ways, SD3.5's motivational approach is closer to Midjourney than to Flux, allowing users to innovate without needing a degree in linguistics.

Beyond customization, Stable Distribution 3.5 advances in other areas:

Fast Compliance: The larger model now rivals much larger models in terms of how well it follows user input, and leads the pack in the world of image generators. So the stability SD 3.5 big hits Flux.1 Dev proves in terms of speed—it's still not in aesthetic quality, though.

Image quality: We're talking about generating images that can stand up to some of the most resource-hungry models without burning through your GPU's memory. In stability metrics, Flux.1 Dev is slightly the king, however, SD 3.5 is much more efficient and less resource-intensive. The SD 3.5 Big Turbo is comparable to the Flux.1 Schnell in terms of attachment and quality.

Style Versatility: Whether you're aiming for 3D shots, photo-realistic images, line art or sketch style, Stable Distribution 3.5 can handle it. It handles a wider range of styles than Flux—at least in our quick tests.

And yes, it's worth mentioning – it's uncensored. SD3.5 is large, although not perfect, capable of producing certain types of content, including nudity, without issue. For better or worse, the model is not intentionally restrictive, so it gives users complete creative freedom (although fine-tuning and some specific questions may be asked for better results).

This was heavily criticized when SD3 was launched and pointed to as one of the main reasons why it failed so badly in anatomical understanding. We can vouch for its ability to generate NSFW images, but the model isn't on the same level as the best Flux scans, but it's comparable to the original Flux models.

But fair warning: while SD3.5 is solid, you NSFW furry artists shouldn't expect a Pony Diffusion model anytime soon, or ever. The creator of the most popular and powerful NSFW model has confirmed that they are not interested in finetuning the SD3.5. Instead, they chose to build their models based on the mainstream. Once you're done, you might want to consider Flux.

For the tinkerers out there, ComfyUI now supports Stable Diffusion 3.5, which allows local awareness of signature node-based workflows. There are plenty of workflow examples ready to go, and if you're struggling with low RAM but want to try the full SD3.5 experience, Comfy has released an fp8-level test model that minimizes memory usage.

What's next?

On October 29th, we'll get our hands on Stable Diffusion 3.5 Medium, and shortly after Stability Control Nets promises to release 3.5 for SD.

ControlNets promise to bring advanced control features tailored for professional use cases and could very well take the power of SD3.5 to the next level. If you want to know more about them, you can read our short guide summary for SD 1.5. However, using controls allows users to do things like choose their subject's position, play around in depth maps, reimagine a scribble-based scene, and more.

So is stable transmission 3.5 flux killer? Not exactly, but it's definitely starting to look competitive. Some users still nitpick, especially after the SD3 mid-flop drama. But with better body handling, clearer consent and significant improvements in fast tracking and output quality, it's hard to argue that this isn't a big step forward. Stability AI is learning from past mistakes and moving toward a future where advanced AI tools are more accessible to everyone.