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 (19)
since krea 2 exists this is pretty much dead on arrival.
As the other user mentioned, pretty much useless, you can't even use it for upscaling as it will completly destroy the original textures of your prefered model, maybe, MAYBE the best use case would be using it for quick edits taking advatange of the model speed, THEN upscaling with a refiner of your choice, a example would add say Mickey Mouse to any of your renders then upscaling using Krea.
I used the workflow where Krea2 rendered the draft in Text--to-Image. SDXL model used as a textures refiner. Worked well for me some time until I've trained the Krea2 with good textures.
I think tandem of Qwen Image 2.1 with Krea2 or SDXL might be viable and usable for some cases.
skill issue, change CFG to 3.5
TS is so ass
A waste of space. A waste of compute.
lol there is some shilly shilling going on. All negative comments, in reality this is the best edit model out right now. K2 is awesome, but only sort of works as an editor, Q2.1 seems pretty great.
Explain how it's the best edit model? The Qwen VAE is bad comapred to the Flux 2 VAE
@elevendr lol, Qwen 2.1 doesn't use the shitty Qwen 1 VAE... But I'm sure you knew that, right?!
After spending a long time messing around with K2 edit, i can actually say it's pretty powerful. It just takes a really long time to figure it out. So ease of use .. no .. but it is very capable.
Performs better photorealism than Krea2's one out of box. Nothing special for other purposes. Would be interesting to see Qwen2.1 style lora for Krea2
>Nothing special for other purposes.
You mean like EDIT that it's supreme at?
Editing feels like SDXL-era inpainting, but edit zones are automated. It doesn't follow spatial logic, and even detailed LLM-assisted prompts with their own skill applied generate random results on different seeds, vaguely related to what you prompted, but spawning artifacts constantly
I like all the gremlins coming out of the woodworks to state their patently absurdist world views out loud.
Incredible model, with features that aren't even utilized by interfaces. Fantastic recursive editing and style transfers.
"It's trash"
sure bro, speak to the mirror, this model is lean and mean with incredible knowledge of existing visual topics.
I look forward to all the cool things people will unlock with it, and all the cool Edit things we can do with it :- )
It's shills and fanbots I think.
I'm not deleting Krea2, but this model is insane for edit and reference.
@wkwjhskssksbg465 It's good to have a variety of models for different tasks.
Anima for example, is real (fun:3) if you enjoy (the process of:2.5) really strange (prompting:-1.5) to get the desired wacky outcome. A little puzzle-solving exercise with keywords.
Interesting model, but Ming Image has been also released, I have generated few images with both and by now I prefer Ming Image
Here some samples of images generated with Ming Image
https://civitai.red/posts/31228588
Those are all SUPER AI and plastic. Can you point to ming doing photographic?
It works well; at first, I read the comments and passed it by, but I'm glad I checked it out myself.