Download
1 variant available
It is a simple workflow that takes advantage of the Qwen-image-vae and the Text Encoder Qwen3-VL.
It works like Flux1.Redux but way much better. (I always hated Redux)
Note: it is not supposed to be a Krea-edit workflow or model, This does not exist. It doesn't keep the identity, just gives variations of the uploaded images. You can even change the style from realistic to anime, just by prompting for this. The prompts are not always respected without the proper LORAs (talking about NSFW).
Since the Text encoder qwen3-VL "sees" the image, it may not correctly see the whole image, or you want to change some details. Then you can just prompt with tags. For example if a man in the image is Caucasian just write in the prompt "African man"
Why you should use it:
1. You have old e.g. flux1.dev generation concepts and you want a better quality with Krea-2
2. You found a random image in the internet and you want to imitate the style.
3. You have an image and you want to change the lighting ( it will change everything but the idea will stay the same, but the lighting will change)
The idea is about using the qwen-image-vae. There are special nodes in more complicated workflows that you might keep some of the identity, but they are not perfect. Krea-2 is not an editing model.
This workflow is a simple one just to get you variations and trust me it can do a lot of things when you have the right LORAs and prompting.
I haven't added a prompt enhancer, cause I never use it.
Also, if you haven't SageAttention installed, just bypass the node.