Image Gen · Model Guide · 5 min read

Stable Diffusion 3.5

The most extensible image model family — a huge fine-tune and LoRA ecosystem running on modest hardware.

Basic Information

Stable Diffusion 3.5 is Stability AI's open image generation family, spanning multiple sizes from a fast large-tier model down to a medium model light enough for 10 GB cards. Its biggest asset is the ecosystem: years of tooling (ComfyUI, Automatic1111, Diffusers), thousands of community fine-tunes, ControlNets for pose/composition control, and mature workflows for every creative pipeline imaginable.

The 3.5 generation brought much better prompt adherence and typography than the SD1.5/SDXL era, while keeping the open-weights, run-anywhere character that made Stable Diffusion famous.

Best use cases

• Local image generation on affordable GPUs (the medium tier fits ~10 GB)
• Highly controlled pipelines: composition control, inpainting, batch pipelines in ComfyUI
• Community fine-tunes and LoRAs for very specific styles and subjects
• Commercial product imagery under the community license terms

Computational Requirements

Runs where you are: the medium tier works on ~10 GB cards, the large tier prefers 16–24 GB, and both quantized community builds go lower.

• Generation speed depends more on your sampler and resolution than raw VRAM; a mid-range card produces a 1024px image in seconds.

Commercial license

Stability AI Community License: free for individuals, researchers, and businesses under $1M annual revenue, with a prohibited-use list; larger organizations need a paid enterprise license. Outputs you generate are yours to use commercially under the community license.