Transforms image into a selfie taken from a standout hand.
Trained for image editing (image2image) but can also work with text2image.
Trigger words: Turn this into a selfie. View from above/below.
Lora can work with 3DCG from visual novels improving realism.
There were some nsfw pics in a dataset.
Description
FAQ
Comments (5)
Using this lora gives me "lora key not loaded" error.
without the lora the base model itself generates much better images .... so this lora is pretty much point less and wasted hdd space
I noticed that too. The lower the LoRA strength the better the likeness of the character, to the point where I have to lower strength to about 0.1 to 0.05.
I guess this LoRA does "above" / "below" a bit better than base.
@aungkhant0911 I can concur the findings of the OP. I tried img2img while trying to preserve the face, but it wasn't quite right. Removing the lora got me way better results in preserving the face along with maintaining the exact pose.
IMO this is really good, but also a little finicky.
1. Requires "view from above" or "view from below". It would be really nice if we could get a straight on shot.
2. On FluxKlein9b-Base, doesn't really preserve character consistency.
3. On FluxKlein9b (turbo), it's really sensitive to step count. Maybe this is just a Klein editing thing in general, but if I don't use 4-5 steps exactly the effect from the LoRA starts falling apart.
On the plus side, character consistency was pretty good in my tests, even with the wildly different views. Not perfect (at one point their eyes changed color) but generally pretty good, even with multiple characters. It also did a good job of preserving the style and scenery.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.