Alibaba's Qwen team released Qwen-Image-2.1-Turbo, an accelerated checkpoint of Qwen-Image-2.1 that generates and edits images in 8 denoising steps instead of 40, on the same 7B architecture with a Qwen3-VL 8B text encoder. It supports 2K output, multi-reference editing, and native transparency via a 64-channel RGBA VAE. Unsloth estimates the base model needs 11 GB VRAM with GGUF. A hosted API costs CNY 0.1 per image. Weights are research-licensed, requiring separate permission for commercial self-hosting.
No score is assigned. Sources and their independence are shown in the citation chain below.