Reference Sheet for Head | Concept LoRA
Big thanks to @YuruSama for assisting me with the dataset. Much appreciated!
Refer to the version details for Changelog. Keep in mind, that despite this LoRA, that some Views are very difficult to generate. This only helps, but dont solve the issue.
Weight can go from 0.5 up to 1. I recommend 0.7. Works with almost every Character. All previews are done without any additional help, except the tags and helping tags. Hairstyle can be messy sometimes. Use of 3D Checkpoint recommended.
ControlNet can also be helpful in this case, and I have included the depth map for it in the Training Images tab.
How to use it? Install ControlNet in your WebUI and include your image with the following settings:
KEEP IN MIND TO REMOVE ALL FACIAL FEATURES, IF YOU ARE DOING A BACKVIEW.
Trigger Words:
FIFront
FIFrontBelow
FIFrontLittleAbove
FIFrontLittleLeft
FIFrontLittleRight
FIFrontSideLeft
FIFrontSideRight
FISideLeft
FISideRight
FIBackAboveRight
FIBackAboveLeft
FIBackRight
FIBackLeft
FIFrontBelowLeft
FIFrontBelowRight
FISideAboveLeft
FISideAboveRight
FISideBelowLeft
FISideBelowRight
FISideLeftBentBackwards
FISideLeftBentForwards
FISideRightBentBackwards
FISideRightBentForwards
FIFrontBelow
FIAbove
FIBackLittleAbove
FIBack
FIBackBelowRight
FIBackBelowLeft
FIBackBelow
Currently little Unstable or bleeding:
FIFrontLittleLeft (Bleed)
FIBackBelow
FIAbove (Unstable)
FIFrontBelow (Unstable)
Use of additional helping Prompt recommended:
from above
from side
from below
from behind
How to use LoRA's in auto1111:
Download LoRA
Copy the file to
stable-diffusion-webui/models/loraSelect your LoRA and use the triggerword
Make sure to change the weight if needed (by default it's
:1)Have Fun!
Description
Added more Angles:
FIFrontBelowLeft
FIFrontBelowRight
FISideAboveLeft
FISideAboveRight
FISideBelowLeft
FISideBelowRight
FISideLeftBentBackwards
FISideLeftBentForwards
FISideRightBentBackwards
FISideRightBentForwards
FIFrontBelow
FIAbove
FIBackLittleAbove
FIBack
FIBackBelowRight
FIBackBelowLeft
FIBackBelow
FAQ
Comments (8)
we did a great job 👌👍
can this be used to make videos of people spinning more realistic with it not turning the back of the head into another face?
Kind of. My partner tried that and it worked, but keep in mind to not use facial prompts on backview
couldn't make it work, the hair's shape it's limited to follow the shape of the head
Might sound like a no brainer, where to put the training images?
Seems to control the hairstyle stronger than the face angle. Is there anything I should be doing to avoid that?
hello sir, i would like to ask
what should i do if i wanted to make a consistent face,
should i use ip-adapter? or is there any other way? what if i want to make a non human face, and make it consistent
Not him, but — I generally wouldn't recommend any technique based on that 128px thing, including ip-adapter, ReActor, and so on.
For humans, there's "nobody" embeddings, which are good at this! I've had success with inpainting by adding a reference image on the side, prompting that they're twins or clones or a character reference sheet, and using a "cloning machine" and/or character reference sheet LoRA.
For nonhumans, it's harder, and 128px-based methods won't work — I'd recommend my technique as somewhat effective, especially if you include multiple LoRAs for it.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.