Stable Diffusion, the open-source alternative to AI image generators such as MidJourney and DALL-E, has been updated to version 3.5. The new model seeks to fix some of the mistakes (which may have been underrated) of the widely criticised Stable Diffusion 3 Medium.
Stability AI says the 3.5 model follows prompts better than other image generators and competes with the much larger model in output quality. In addition, it has been tuned for a greater variety of styles, skin tones and features, without needing to be explicitly prompted.
The new model comes in three flavours. Stable Diffusion 3.5 Large is the most powerful of the three, with the highest quality, while also leading the industry in prompt follow-up. Stability AI says the model is suitable for professional use at 1 MP resolution.
Meanwhile, Stable Diffusion 3.5 Large Turbo is a “distilled” version of the larger model, focusing more on efficiency than maximum quality. Stability AI says the Turbo variant still produces “high-quality images with exceptional prompt adherence” in four stages.
Finally, Stable Diffusion 3.5 Medium (2.5 billion parameters) is designed to run on consumer hardware, balancing quality with simplicity. With its greater ease of customization, the model can create images between 0.25 and 2 megapixels in resolution. However, unlike the first two models, which are available now, Stable Diffusion 3.5 Medium won’t arrive until October 29.
The new trio follows the poorly received Stable Diffusion 3 Medium in June. The company admitted the release “did not fully meet our standards or our communities’ expectations,” as it produced some ridiculously grotesque body horror in response to a prompt that asked for no such thing. The repeated mention of exceptional prompt adherence by Stability AI in today’s announcement is likely not a coincidence.
Although Stability AI only briefly mentioned it in its announcement blog post, the 3.5 series has new filters to better reflect human diversity. The company describes the new model’s human output as “representative of the whole world, not just one kind of person, with a diversity of skin tones and features that doesn’t require a lot of prompting.”
We hope it’s sophisticated enough to take nuance and historical sensitivity into account, as was the case with Google’s debacle earlier this year. Without any prompting to do so, Gemini produced a collection of wildly inaccurate historical “photos” such as ethnically diverse Nazis and America’s founding fathers. The backlash was so intense that Google didn’t add human generations back until six months later.
If you’re a Prime member, you can save $50 on the recently released Colorsoft Kindle, also known as the first Kindle with color. It’s the first discount we’ve seen since this model debuted. We encountered a “yellow band” issue with our review unit, but we later received an updated reader for which Amazon made “appropriate adjustments” to fix the problem. The company The software and display adjustments applied worked – and we actually liked the overall effect better. Check out our review for the full story.