Hey @fantaseed,

Thanks for reaching out! Just to clarify, V7’s dataset doesn’t include any non-realistic images. That said, all versions can create non-realistic images super easily with just prompts or LoRAs. For V5 and V6, I worked on cutting down accidental non-realistic outputs. Sorry if that caused any mix-up—happy to clear things up if you have more questions!

Best,

Thank you @Ant6431 I'm glad you liked.

Say thank you to VISA and MASTERCARD

Unfortunately this is not true. I tried to mute the faces from the model using several different techniques. One mute completely the layers 7,12,16 and 20 from the model. The faces became 100% unchanged. But, turns out that the anatomy parts became garbage. It uses the same layers to learn how to draw it.

But this is a very small problem, we can circumvent this on several ways, one is just to change the str, another is to use your lora with inpaint and mask after the creation, and if you want to be very hard you can diff your lora from mine or even fuse it putting the face layers over it.

If you want to experiment with this I recommend the LoRa Explorer https://github.com/shootthesound/lora-the-explorer

After several tests I got a face LoRa that really get very different, changing the values to xxx -> 0.6 and letting the face LoRa 1 made it perfect. It will change from each face LoRa, you will need to experiment a little, but it's doable.

civitaifreezable801607 There are several options one is the
Quote from the project readme:
LoRA Difference

Remove unwanted influences from LoRAs:

Perfect for removing face changes from style LoRAs and creating character-neutral styles.
or
Layer-Based Merging

Surgically combine layers from different LoRAs:

Face A + Style B: Facial layers from character LoRA, style from artistic LoRA

Facial Priority: All potential facial layers from one LoRA

Complement Split: Early layers from one LoRA, late layers from another

Fix Overtrained LoRAs: Replace problematic layers with clean ones

ToxicBot this tool allows you to remove all parts of a LoRa that interfere with another LoRa a "LoRA Difference", in other words, if you think a LoRa is making your Character LoRa less effective, you can guarantee that this will not happen. If you read the Readme on the project there are more detailed explanations there.

Thank you for your kind words. I'm glad you liked.

Thanks, I'm glad you liked. I was thinking about it, but I'm not sure if it's worth with wan 2.2 and Gwen here.

Narz OK, I can do it. But it will take a while, currently I'm working on a commission and every byte of my VRAM is taken by my client. After this I can do a Krea version, I'm a little afraid that my huge dataset may destroy krea style, I don't know. We will need to test it out.

It's done