Redditor shares full Draw Things LoRA training setup on M1 Pro, asks community to spot errors
Playswith_squirrel · reddit · 2026-10-25
A Reddit user posted their complete on-device SDXL LoRA training configuration using Draw Things on a Mac M1 Pro (32GB RAM), asking the community to spot any mistakes:
- Dataset: 15 high-quality photos (5 close-up, 5 mid-shot, 5 full-body) with varied outfits and backgrounds to prevent concept bleed, all manually cropped to 1:1 squares
- PEFT settings: 1024x1024, Network Dim 32, Scale 1.0, LR 0.0001→0.0, 2000 training steps, checkpoints every 250 steps, gradient accumulation 4, Noise Offset 0.05, Memory Saver Balanced
- Plan: no cloud services; they intend to test the 1000/1250-step checkpoints instead of the final 2000 to avoid overfitting. Their 2021 MacBook is underpowered so training is slow, but acceptable until an upgrade
Useful reference for anyone attempting local LoRA fine-tuning on consumer Mac hardware.
More from Infra
- Free iPhone idea: 42-month 0% Apple Lease paid off by renting idle compute to AI — SuB8u · 2026-10-25
- 4B model fine-tuned 27 min on one A100 lifts letter-exam accuracy to 74.5% — MaziyarPanahi · 2026-10-25
- Starlink in-flight WiFi hits 408Mbps, ~30% faster than median US home broadband — elonmusk · 2026-10-25
- Musk: Mine lunar silicon to build AI satellites — Earth launches can't scale to 500-1,000 TW a year — elonmusk · 2026-10-25
- Router dev crowdsources production model picks beyond the big four as small-model gap narrows — Remarkable-Guille · 2026-10-25
- Ultrapolis dev breaks down what it takes to render every frame at 60fps — mflux · 2026-10-25