Pruna open-sources few-step LoRA adapters making Qwen-Image-2.1 up to 6.3x faster
huggingface · x · 2026-09-25
Pruna AI released Pruna-Qwen-Image-2.1, DMD-based few-step LoRA adapters that let Alibaba's Qwen-Image-2.1 generate and edit images in 5 or 8 steps instead of 40 — up to 6.3x faster.
- The 8-step adapter is the recommended default, supporting text-to-image plus single/multi-image editing (up to 3 reference images) at 1K resolution; the 5-step variant trades quality for speed
- Adapters load on top of the base model with no pipeline changes, each using its own sigma schedule and running without CFG
- v0.1 doesn't yet match base-model quality; distillation improvements are ongoing, with a fast Diffusers integration available
Related event: Pruna Open-Sources Few-Step LoRAs Speeding Up Qwen-Image-2.1 Up to 6.3x(4 posts)→
More from Multimodal
- Ant Group's open-source Ming-3.0-tts-0.5B tops Open TTS Leaderboard with 0.88 WER — NielsRogge · 2026-11-12
- Pika Integrates Seedream 5.5: Split Images Into 100+ Layers for Precise Multi-Turn Edits — aziz4ai · 2026-11-12
- Llama app adds simultaneous voice, image and video input, running locally on LFM3-Omni — mervenoyann · 2026-11-12
- Seedance 2.5 can now generate 4K video directly, no upscaling needed — charis_ai · 2026-11-12
- codec.audio benchmarks neural audio codecs on noisy in-the-wild speech; MOSS tops — kastnerkyle · 2026-11-11
- State of AI video: Isaiah Ross's hyper-realistic showcase hailed as current peak — LinusEkenstam · 2026-11-11