
Stability AI has just unveiled Stable Diffusion 3.5, marking what is appearing to be a significant turnaround following the underwhelming reception of SD3 Medium in June.
The latest release introduces multiple model variants, including the flagship SD3.5 Large and a performance-focused Large Turbo version, with a Medium variant scheduled to be released soon come October 29th.
In a refreshingly candid acknowledgment, Stability AI addressed the elephant in the room regarding their June release, stating that SD3 Medium “didn’t fully meet our standards or our communities’ expectations.”
Yeah.. that is putting it lightly..
Anyhow.. it seems that rather than rushing out fixes, the company took time to develop a more robust solution that better serves the community
The new SD3.5 Large model, boasting 8 billion parameters, represents the most powerful iteration in the Stable Diffusion family to date. It’s designed for professional use cases at 1 megapixel resolution and demonstrates superior prompt adherence compared to its predecessors.
For users seeking faster performance, the Large Turbo variant maintains high quality while requiring only 4 steps for generation.
Key improvements in the SD3.5 release include:
- Easier fine-tuning through Query-Key Normalization
- Efficient performance on consumer hardware
- Improved diversity in outputs without extensive prompting
- Versatile style generation capabilities, from photography to line art
- Market-leading prompt adherence
The model comes with the same “revised” license that SD3 Medium was given offering free usage for both non-commercial purposes and commercial applications (for businesses with annual revenue under $1M) under their Community License.
With this release, Stability AI appears to have learned valuable lessons from the SD3 Medium launch, delivering a more polished and capable suite of models.
That couldn’t come at a better time as competition in the space has become quite fierce as of late, with Flux becoming the go-to image diffusion model in the community, and Nvidia just announcing the release of their own flagship image diffusion model “Sana”.
The upcoming October 29th release of SD3.5 Medium will be particularly interesting to watch, as it promises to address the shortcomings of its predecessor while maintaining accessibility for consumer hardware.
Hi, I’m the guy behind Helpful Tiger. This website shares a little bit of everything related to generative AI and online marketing. Have questions? Reach out, and I’ll do my best to help!