In the last few hours, the folks at Stability.ai announced the launch of the new “Stable Diffusion 3 Medium” model, which supposedly includes two billion parameters, more accurate prompt comprehension, and overall improvements in areas such as photorealism. This model also prioritizes running on environments with more limited VRAM, but if you want to see what it can do before starting any download, the truth is that there are already free spaces on HuggingFace…
I admit that I’m disconnected from AI image generation. After getting hundreds of thousands of results with Auto1111 and the models available on Civitai, one inevitably starts to recognize patterns, errors, limitations. Add to this the hardware requirements: more complex models demand more VRAM, and the only alternative for many users has been recycling old models.
So we come to Stable Diffusion 3 Medium, which in theory received several optimizations to ensure its execution on consumer hardware. In my mind, that means “any GPU with 8 GB of VRAM or less”, but instead of sharing links to installation and download guides, today I recommend something much simpler: Visit HuggingFace and try the free demos of Stable Diffusion 3 Medium online.
Testing Stable Diffusion 3 Medium Online for Free
Why? Well… because they are trashing it on Reddit. The model is supposed to be stronger at generating photorealistic images, and that’s true if we reproduce landscapes or structures, but the problem is when humans appear in the equation. The “old tricks” seem to work well (hands in pockets or behind the back, broad negative prompts, etc.), however, it’s very difficult to accept that these models keep making the same mistakes after so long, when their competitors have been generating hands correctly for at least a year.
Users blame the “excessive censorship” by Stability.ai (something that already happened with previous versions), but others associate the model’s quality with the financial difficulties the company is currently experiencing. Internal politics aside, if you want to test with your own prompts in Stable Diffusion 3 Medium, click on the link below, choose one of the available “Spaces”… and be patient. At times, it’s very difficult to get a free GPU…
Try Stable Diffusion 3 Medium: Click here
