Qwen-Image 2.1 prompt enhancer hits 4.4x speedup in ComfyUI, now runs on 8GB VRAM

mozophe · reddit · 2026-10-01

A dev updated their ComfyUI custom node wrapping Qwen-Image 2.1's official prompt enhancer, adding a llama.cpp backend with MTP support:

• On an RTX 4090 Laptop (16GB), text-to-image throughput rose from 23.5 to 63.2 tok/s (2.69x), and two-image editing from 19.4 to 85.2 tok/s (4.39x)
• Q4KM quant runs on as little as 8GB VRAM; Q80 shows no quality loss, and maxlength barely affects speed under llama.cpp
• llama-server and GGUF models auto-download on first use; ComfyUI backend retained for AMD/Intel/Mac
• Same-seed comparisons show the enhancer helps most with in-image text, busy scenes, and multi-step edits
• Uncensored (Heretic) GGUFs also available

Project: github.com/mozophe/ComfyUI-Qwen-Image-2.1-PromptEnhancer-MTP

Original post →

More from Infra

• A goldmine resource for learning GPU programming internals — goyal__pramod · 2026-10-02
• Gradium claims fastest TTS yet with ~50ms time-to-first-audio, tops sub-100ms naturalness — mattturck · 2026-10-02
• Engineer: Gemini 4 Argon drives fleet-wide data center optimization gains — rakyll · 2026-10-02
• Aleph Alpha Details MoE Pre-training Scaling: 30B-A3B on 16-512 B200s at 35% MFU — CatAstro_Piyush · 2026-10-02
• HBM to eat 22% of world's DRAM wafers as memory prices triple in six months — aronchick · 2026-10-02
• Can Google's AI chip beat Nvidia? A video analysis — Ill_Vegetable169 · 2026-10-02

Infra channel →