AI is changing the world. Don't stay behind. Clear summaries, community insight, delivered without the noise. Subscribe to never miss a beat.

Automatic1111 for Apple metal, 40% speed up sd1.5

Teaching Automatic1111 to Speak Metal on an M1

therad.ninja

August 12, 2026

13 min read

🔥🔥🔥🔥🔥

45/100

Summary

Automatic1111 runs slower on Apple hardware compared to expectations, prompting a search for performance improvements. A five-step DPM++ SDE method on an M3 Pro reduces processing time from 8-10 seconds to 3-7 seconds.

Key Takeaways

Automatic1111's generation time on an M3 Pro improved from 8-10 seconds to 3-7 seconds after optimizations were implemented.
The implementation of Metal Flash Attention specifically for Stable Diffusion 1.x shapes resulted in performance gains without compromising output quality.
The optimizations focused on workload-specific improvements rather than broad benchmarks, ensuring that changes preserved the desired generation results.
The native extension's inefficiency in submitting MPS command buffers after every attention call was identified as a significant factor affecting runtime performance.

Read original article

Community Sentiment

Mixed

Positives

The 40% speed boost on SD 1.5 is a game changer for those still using it, making high-quality image generation feel effortless on localized hardware.
Invoke AI offers a seamless canvas-style experience for outpainting and inpainting, clearly standing out in terms of workflow efficiency.
Despite being labeled old, SD 1.5 remains relevant for production systems, enabling users to extract high-quality images quickly without hassle.

Concerns

Automatic1111's lack of updates has pushed users towards alternatives like Forge Neo and ComfyUI, signaling a decline in its popularity.
Many see SD 1.5 as outdated, especially with newer models like Mini Max H3 available, questioning the value of continuing to use it.
ComfyUI's steep learning curve is a turn-off for some, leading to frustration when transitioning from Automatic1111.