Model · Stability AI · Image-Enhance
Stable Diffusion x4 Upscaler
Stability AI's diffusion-based 4× upscaler. Trades speed for quality — invents plausible high-frequency detail rather than just sharpening, which suits AI-generated images especially well.
- Modality
- Image-Enhance
- License
- CreativeML Open RAIL++-M (Open weights)
- Released
- December 7, 2022
- Last verified
- September 8, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes, but slow. Any Apple Silicon Mac (M1 or newer) with at least 16 GB (32 GB recommended) of memory. It runs, but much slower than on an NVIDIA card.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 8 GB (16 GB recommended) of video memory.
- No graphics card
- Not practical. Too slow without a graphics card or Apple Silicon.
Works on a Mac through PyTorch MPS but slowly, and large inputs can run out of memory; upscale in tiles.
Numbers are for: official fp16 weights, small (e.g. 128-px) input tiles. Our estimate from the model’s size — the maker publishes no figure.
Strengths
- Diffusion-based — generates plausible high-frequency detail
- Pairs naturally with Stable Diffusion image-gen pipelines
- Open-weights, runs in ComfyUI / Automatic1111 with one node
Weaknesses
- Slower than GAN/transformer upscalers — minutes per image on consumer GPUs
- Can hallucinate detail that wasn't in the original — wrong choice for archival or evidentiary work
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | RAIL++ license |
| ComfyUI | local | Native upscaler nodes |
| Replicate | hosted | API key required |
Official sources
- Model card model
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (HF stabilityai/stable-diffusion-x4-upscaler card (attention slicing for low VRAM)).
- — Verified current — no newer open-weights Stability upscaler exists in the full org listing.
- — Initial entry.
Esc