Qwen-Image 2.1 runs natively on M1 Max: ~24s at 512x512, FP16 no quantization
adel_b · reddit · 2026-09-27
A developer open-sourced qwen-image-cplus, a native Apple Silicon runtime for Qwen-Image 2.1 running FP16 weights without quantization. Benchmarks on an M1 Max 32GB:
- 512×512 text-to-image in 24s
- 1024×1024 four-step Viggle t2i in 38–39s
- 832×1248 six-step Viggle LoRA image edit in 102s
- 1024×1024 base model, 40 steps: 451s uncached vs 127s with Cache-DiT
- Motivation: stable-diffusion.cpp took 1006s on the same machine
Weights are on Hugging Face; the author is asking for feedback on edit quality.
More from Multimodal
- Adobe MAX: MCP across Creative Cloud, AI agents editing layers, Firefly Builder and Graph — umesh_ai · 2026-11-11
- Comment: China leads the AI video revolution while the West argued about AI art — TinfoilTricorn · 2026-11-11
- "QT your Geisha" meme takes over X with stunning porcelain-style AI portraits — LudovicCreator · 2026-11-11
- MiniMax H3-Image early access test: 4 images, one coherent cyberpunk world — Aiden_Tech_Ai · 2026-11-11
- MiniMax H3 Image generates a GTA6-style hurricane scene from a single prompt — socialwithaayan · 2026-11-11
- Dragon Ball reimagined as a 1971 movie: another hit AI retro trailer — techhalla · 2026-11-11