Run Flux 2 Dev (30B) locally with block offloading, mix models for T2I and editing
Altruistic_Heat_9531 · reddit · 2026-09-21
A Reddit guide to local image generation model choices: Qwen 2.1 (7B, solid I2I editing), Krea 2 (12.9B, tuned for T2I), and Flux 2 Dev (30B, T2I+I2I, 1000 Elo on AA Arena vs GPT Image 1 at 1007 and NBP 1100). Key tip: it runs as long as RAM+VRAM roughly equals model size—modern ComfyUI auto-detects VRAM and uses block offloading. Workflow idea: generate with Krea 2, then edit with Qwen 2.1 or Flux 2 Dev.
More from Infra
- ISTA-DASLab splits prefill/decode quantization, squeezes 27B-class model into 13.7GB GGUF — victormustar · 2026-09-21
- Usage limit killed scheduled agent jobs for 19 hours — team shares 4 fixes so agents monitor themselves — lilythemoon54 · 2026-09-21
- Devs say avoiding cache misses could boost Claude Code/Codex effective usage limits 10-20% — chaseleantj · 2026-09-21
- How do you test LLM provider failure in production? A Reddit discussion — Rama_Surasani_ · 2026-09-21
- Jensen Huang stands by $3-4T AI infrastructure market forecast — emmanuelvivier · 2026-09-21
- Free Zoom meetup: disaggregated speculative decoding on d-Matrix chips plus inference engine tuning — cfregly · 2026-09-21