Originally released by Qwen on Hugging Face. All credit goes to the Qwen team.
Qwen Image 2.1 combines text-to-image generation and image editing in a single model. Its visual generation component has 7 billion parameters across 32 single-stream DiT layers, balancing image quality, efficiency and versatility.
The release introduces four main improvements:
- Efficient architecture: Mixed-granularity attention and reuse of cached prefixes reduce computation while maintaining image quality.
- Generation and editing with transparency: Create regular or transparent RGBA images, edit transparent layers and isolate subjects from photographs.
- Flexible editing: Work with up to ten reference images, guide local changes using circles, painted annotations or separate masks, and preserve the identity of people and products.
- Improved visual detail: Refined typography, portrait lighting, realistic textures and finer details.
Released under the Qwen Research License Agreement. See the original model card for examples and usage instructions.
Description
FAQ
Comments (6)
very nice. no instructions and no workflows. great job.
Just use google, noob.
The Turbo version isn't bad; you can create some nice illustrations with it.👍
I am SO excited to try this !! Thank you SO much !!!!!!
I have a 4070 Ti Super; I was getting intermittent black images in ComfyUI using the standard loader and the int8_convrot model, until I switched the loader to "Load Diffusion Model INT8 (W8A8)" from "int8-fast"
Interpolation of sigmas for Turbo:
10 steps: 1, 0.9829, 0.9643, 0.9436, 0.9206, 0.8951, 0.8598, 0.7766, 0.6072, 0.3387, 0
12 steps: 1, 0.9858, 0.9707, 0.9542, 0.9362, 0.9165, 0.8951, 0.8678, 0.8122, 0.7045, 0.5283, 0.2854, 0
14 steps: 1, 0.9879, 0.9752, 0.9614, 0.9467, 0.9308, 0.9136, 0.8951, 0.8728, 0.8325, 0.7584, 0.6376, 0.4653, 0.2463, 0
16 steps: 1, 0.9894, 0.9785, 0.9667, 0.9542, 0.9409, 0.9266, 0.9113, 0.8951, 0.8762, 0.8451, 0.7910, 0.7045, 0.5789, 0.4146, 0.2164, 0