Low-VRAM showdown: Qedit 2511 skips prompt enhancer but Qwen 2.1 wins on facial drift
Theonewhoknocks_001 · reddit · 2026-11-09
A ComfyUI beginner with an RTX 3050 4GB + 8GB RAM shares a hands-on comparison on low-end hardware:
- Qedit 2511 (fp8 + 4-step Lightning LoRA, default scheduler): 110s per image, works fine without any prompt enhancer, but suffers severe facial drift (attributed to the turbo LoRA and weak hardware).
- Qwen 2.1 Turbo (int8, 8 steps): 90s, much less facial drift, but poor prompt adherence — even well-written prompts and third-party enhancers (including qwen 3.5 2b abliterated) give inconsistent results, and the official enhancer is both restricted and low quality.
The author is torn on whether to just switch to Qedit, which is friendlier for small-VRAM setups by skipping the enhancer entirely.
More from Multimodal
- Slow Krea2 LoRA training on RTX 5080 with Fizgig: full config shared, asking for tuning help — BeautifulStation4 · 2026-11-09
- Dreamina Marketing Studio turns one product photo into a full ad campaign, now wired to Codex CLI — mhdfaran · 2026-11-09
- Krea 2 ships great quality but almost no character LoRAs — community puzzled — Shaz0r94 · 2026-11-09
- Addis Voices 2 TTS adds seven African languages across 56 voices — strickvl · 2026-11-09
- GPT Image 2.6 lands on Leonardo with support for up to 16 reference images — aziz4ai · 2026-11-09
- Midjourney sref 3911489532 yields moody blue-orange clock-tower film aesthetics — umesh_ai · 2026-11-09