Stable Diffusion

Running on modest hardware, broad community support, and access to thousands of community-made fine-tunes and LoRAs

The old workhorse. Not the sharpest output today, but it has the largest ecosystem of extensions, models, and tutorials. Pick this if you are learning or need maximum compatibility with community tools.

Slightly cleaner outputs than SD 1.5 with a similar hardware footprint

An incremental upgrade over 1.5, but it never gained as much community traction. The fine-tune and LoRA library is smaller. Most people skip straight from 1.5 to SDXL now.

High-resolution, photorealistic or highly detailed images; serious hobbyist and professional use

A major quality jump. Runs natively at 1024×1024. Has a two-stage pipeline (base + refiner model) for sharper detail. Pick this if your GPU can handle it and you want near-commercial image quality.

Real-time or near-instant image generation, rapid prototyping, live prompt experimentation

A distilled version of SDXL that can generate a usable image in 1–4 steps instead of 20–50. Quality is slightly lower than full SDXL but the speed is dramatic. Good for interactive tools or fast iteration.

Accurate text rendering inside images, complex multi-subject compositions, and prompt coherence

Stability AI's most capable model as of 2024. Uses a diffusion transformer architecture. Noticeably better at putting legible text in images and following complicated prompts. Requires a license agreement for commercial self-hosting.

Generating short video clips (2–4 seconds) from a single still image

Extends Stable Diffusion into video. Output is short and can be inconsistent, but it is one of the few open-weight video models available. Still experimental for most practical uses.

Users who want Stable Diffusion access without installing anything locally

Stability AI's own hosted front-end. Convenient but more expensive per image than self-hosting. Reasonable starting point if you just want to try the technology before committing to a local setup.