I use Midjourney a lot in my own work, and I say that openly. And here is another honest confession before anything else: I am not in the best position to give a deep opinion about Stable Diffusion. I simply do not have hardware good enough to really play with local generation — to push it long enough to create something with the quality that would satisfy me. So take this article as a map of the territory in 2026, drawn by someone watching from a distance with genuine respect. Because I do believe in the power of open source, and I would love to have that opportunity in the future.
The short version for newcomers: Stable Diffusion is an AI image model with open weights. That means you can download it, run it on your own computer, modify it, train it on your own art, and never pay a subscription. When it appeared in 2022 it changed everything, because suddenly image generation was not locked inside anyone's product.
Here is the surprising part of the 2026 story: the newest official model is not new. Stable Diffusion 3.5 — released back in October 2024 in Large, Large Turbo and Medium versions, with a redesigned architecture and much better text rendering — is still the latest major release. There has been no SD4. The momentum did not die, though. It moved from the company to the community.
In practice, when people say 'Stable Diffusion' in 2026, they usually mean SDXL — the 2023 model that the community never stopped improving. Its ecosystem is enormous: thousands of fine-tuned checkpoints like Juggernaut XL, DreamShaper XL and RealVis XL, LoRAs for every imaginable style, and mature ControlNet support for precise composition control. SD 3.5 is technically stronger, but eighteen months after release its community ecosystem is still catching up to what SDXL already has.
Meanwhile, a newer family called FLUX became the open-weights model many enthusiasts reach for first. It is approaching SDXL's ecosystem in breadth — though as of this spring it still counts its community fine-tunes in the hundreds while SDXL counts thousands. The gap is closing fast, and for photorealism many consider FLUX the new reference point.
The real center of gravity in 2026 is ComfyUI, the node-based tool where all of these models plug into visual workflows. It has become serious infrastructure: over a hundred thousand stars on GitHub, a new release roughly every two weeks, and this year a thirty-million-dollar funding round. Professional pipelines — the kind studios use for consistent characters, upscaling chains and video pre-processing — are built there.
There is also a market that rarely appears in polite reviews but absolutely shapes this ecosystem: adult content. Most commercial services block it entirely, so for that audience, local generation is essentially the only road. Whatever your personal view on it, it is one of the quiet engines that keeps the open-source scene moving with its own momentum, independent of any company's terms of service.
As for me, I watch all of this from the outside with honest curiosity. I wonder how far these tools really are today — I am in the dark about the day-to-day experience of working with them. And I wonder what truly drives the community more: the freedom of owning the whole pipeline? The cheap, subscription-free economics? The endless customization? Probably a mix that looks different for every artist. If you live inside the Stable Diffusion or FLUX world, I would honestly love to hear your perspective. And one day, with better hardware on my desk, I want to find out for myself.