image model · stable-diffusion · iOS
SD Can I run Stable Diffusion XL 1.0 on iPad Pro M4 (16GB, 1TB/2TB config)?
Yes. Stable Diffusion XL 1.0 runs on iPad Pro M4 (16GB, 1TB/2TB config) at fp16 (~7.5 GB of ~12 GB usable).
Runs at fp16 using ~7.5 GB of ~12 GB usable.
- Peak VRAM
- ~7.5 GB
- Usable on device
- ~12 GB
- Device memory
- 16 GB
- Quant
- fp16
How to run it
Use ComfyUI or AUTOMATIC1111 / Forge at fp16. It conditions on an image, not a text prompt; the pipeline offloads each stage off the GPU between passes, keeping peak VRAM near the active stage.
- Type
- image (UNET)
- Parameters
- 2.6B
- Peak VRAM
- ~7.5 GB at fp16
- Resolution
- 1024×1024
- License
- CreativeML OpenRAIL++-M
- Memory
- 16 GB unified
- Usable for weights
- ~12 GB
- Best runtime
- MLX (via Python or Swift; mlx-lm package)
You could also run
Run Stable Diffusion XL 1.0 on other hardware
FAQ
Can iPad Pro M4 (16GB, 1TB/2TB config) run Stable Diffusion XL 1.0?
Yes. Stable Diffusion XL 1.0 runs on iPad Pro M4 (16GB, 1TB/2TB config) at fp16 (~7.5 GB of ~12 GB usable).
How much VRAM does Stable Diffusion XL 1.0 need?
iPad Pro M4 (16GB, 1TB/2TB config) has room to spare. At fp16 the realistic peak is ~7.5 GB of VRAM, versus ~8.5 GB with every component kept resident (no offload). With aggressive CPU offload it drops to ~4 GB, much slower.
What do I use to run Stable Diffusion XL 1.0 locally?
Stable Diffusion XL 1.0 runs in ComfyUI or AUTOMATIC1111 / Forge (among others). It loads as a diffusion checkpoint plus its image encoder and VAE, not a single chat command.
Sources
VRAM figures are sourced peak-usage anchors at the noted quant, validated 2026-06-15. See methodology.