Qwen-Image-2.1-Turbo An accelerated open-weights checkpoint of Qwen’s 7B image generation and editing model that creates and edits 2K images in 8 denoising steps while keeping quality close to the 40-step base model.
Visit productFeatured media
Loading media…
View on XCosigns Received
“You can run locally Google Nano Banana 2 level 7B editing model fit your home around 6 GB VRAM with offload.”
“Run Qwen-Image-2.1-Turbo GGUF locally 🖼️ We released quants on Hugging Face, from 2-bit (2.6 GB) up to lossless BF16 (14.2 GB) Our 4.2 GB AD-Q4_K has 21% lower LPIPS than a plain llama.cpp Q4_K, so its images stay closer to the original GGUFs: https://huggingface.co/AtomicChat/Qwen-Image-2.1-Turbo-GGUF”
“Qwen just dropped qwen-image-2.1-turbo. here are the zero-hour gguf quants. (config & repo below) download, convert, test, upload, all in ~45 minutes on my homeserver with a single RTX3060. q8_0, q6_k, q5_0, q4_k and q3_k are all up. 8 steps, about 35 seconds per image, and the text rendering is still clean.”
“Swapped Qwen-Image-2.1 for the new 2.1-Turbo on my home AI setup (Mac Studio M4 Max, running locally). Same prompts, same seeds: • 2.1 @ 25 steps: 213–251s per image • Turbo @ 8 steps: 87–104s per image ≈2.4× faster, same memory use Better text, too. Tested birthday flyer , 2.1 put gibberish lines under the date, and Turbo got all the text right on 5 out of 5 tests. Photos look about the same, with Turbo a bit more contrast. Old weights deleted. Turbo is the new default for my family's chat assistant. @Alibaba_Qwen nice drop 🔥”
“I updated all my wf to use this version, with loras included, face and body swap, image edit and generation, all great for now. Did not do anything to crazy so I don't know about refusals, I did not get any”
“i compared Qwen-Image-2.1 vs Qwen-Image-2.1-Turbo vs GPT Image 2.5 - tested on 1 DGX Spark - turbo is fast af with the same quality as 2.1”
“Local AI keeps getting crazier: Google's Nano Banana 2 now runs locally, and Qwen-Image-2.1-Turbo delivers impressive speed. My GPUs are working overtime today. Have fun experimenting! 🚀”
“This has been amazing. Got @FootballManager face generation down to 9 seconds per image (previously 40-60s) - it's so fast and quality is spot on - no artifacts so far. This is on 16GB 5060 Ti at 640 resolution. Super impressed with it”
“Congrats! We're excited to make Qwen-Image-2.1-Turbo quants and GGUFs for it so everyone can run it locally! 💜”