unsloth/Qwen-Image-2.1-GGUF — new model trending #30 on Hugging Face
Unsloth published GGUF quantized versions of Qwen-Image-2.1, Alibaba's 7B text-to-image and editing model, using Dynamic 2.0 per-layer precision.
Unsloth released a GGUF quantization of Qwen-Image-2.1, a unified text-to-image and image-editing model whose visual generation component has 7B parameters across 32 DiT layers. Unsloth's Dynamic 2.0 methodology upcasts sensitivity-critical layers per tensor to preserve quality, with the Q4_K_XL encoder measuring LPIPS 0.029, SSIM 0.959, 5.15 GB footprint, and 36.5 s render time versus Q4_K_M. The GGUF contains only the denoiser and requires the separate BF16 VAE plus the Qwen3-VL-8B text encoder, running in Unsloth Desktop and stable-diffusion.cpp. The model supports native RGBA transparency, up to 10 reference images, and identity-preserving local edits via circles, annotations, or masks.
- GGUF quantization of Qwen-Image-2.1's 7B-parameter DiT visual generation component
- Dynamic 2.0 upcasts sensitivity-critical layers per tensor for near-lossless quality
- Q4_K_XL encoder benchmark: LPIPS 0.029, SSIM 0.959, 5.15 GB, 36.5 s per render
- Requires separate VAE and Qwen3-VL-8B text encoder to run
- Organizations
- Unsloth
- AI models
- QwenQwen3-VL-8B
From the original article
3 min read at huggingface.cotext-to-image · 130,465 downloads · 214 likes · license other
By unsloth · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.