Please save your buzz! I generate locally and don't really need it :D. If you'd like to support my work consider https://buymeacoffee.com/middlingmaker
Check the version info for specific usage on the different LORAs. Some LORAs may have a single HIGH and LOW file meant to be used in both high and low slots for WAN 2.2. Some may have separate high and low files.
For testing and to create the showcase videos, I start with an image of an empty doorway or background and call the character with the lora eg. "a well lit clothing store is out of focus in the background, then tifa enters the scene from the right, tifa strikes a pose with her hand on her hip and smiles at the viewer, focus tifa" and then use that frame to start the video.
High Strength - 0.7-1
Low Strength - 0.7-1
If starting from a frame with the character already visible lowering lora strength to 0.4-0.6 can help preserve character context (like what their back looks like) and will improve motion. Use with caution as the smaller details (eye color, arm bands, footwear) can start to change if the lora strength is too low.
Character requests available, leave them in the comments and I'll see what I can do :)
Description
Grace Ashcroft from Resident Evil 9 Requiem. V1.0 includes her brown jacket and black jeans outfit as well as her naked.
This version has improved motion over the other character LORAs and has seperate high and low LORAs for WAN 2.2. When the character is off-screen the LORA calls the photoreal version of her. Starting with an animated or CG image and using the LORA will maintain the character consistency, but lower the strengths to avoid turning her photoreal.
Call the character with "grace ashcroft, short wavy light blonde hair, pale grey-blue eyes, fair skin, brown leather biker jacket, black hoodie, black skinny jeans, brown boots," for a photoreal version. For animated versions starting with an image and lowering lora strength maintains consistency.
Thank you to the generous supporters who made this training possible.
FAQ
Comments (16)
Why do the last to preview videos look more like source filmmaker/blender than AI generated?
Check the usage instructions for the version. The LORA calls a photoreal style character, but it can also be used on animated or CG starting frames as well. Many sites don't allow photoreal style so I included a mix of photoreal images and screenshots from the games.
I'm still working on getting the LORA to actually call the more animated style of images. V1.0 will only produce the photoreal character if she's off screen (as far as I know, if you crack this please let me know). But if you start with an animated style image or a screenshot from the game, the LORA will maintain her consistency and details, just lower the strengths to avoid it turning her from the animated style into the photoreal version.
@GlowingGuardianGirl I'm not even going to bother with this attitude
@MiddlingMaker From what I can see is that the very last cgi video involves at least 4 different loras unzipping the pants, the penis itself, blowjob, facial etc. I just can't get my head around the fact that the whole video has only like two lines of prompt. I initially just wanted to know if the prompts you used for the two last videos are even the real prompts or if you just pasted something manually written afterwards. Because there is no reference to a facial in the prompt, yet I'm sure that it has the anime cumshot aesthetics lora included.
@fenasikerim of course I use other LORAs, mine only has Grace's appearance. The cumshot specifically was done by editing an image for the final frame and adding the fluid to her and letting the checkpoint I use handle it with it's built-in LORAs (though I do often use LORAs for it specifically). The two different colors of fluids are a bit of a giveaway.
@MiddlingMaker Alright, things are clear now. Are you looking for something like this? https://streamable.com/frgkbp
@fenasikerim I'm not sure what you mean
@MiddlingMaker "will only produce the photoreal character if she's off screen (as far as I know, if you crack this please let me know)"
Incredible work, brother, thanks for sharing it with us.
Aerith:
The pink bow is really strong 0_0 , if I put her with anyone, even a blue skinned Asari it adds a pink bow to both, Asari don't even have hair to add the bow lol, negative prompts don't even help ... even if its not mentioned , and on any sampler, the Aerith one for must include a pink bow lol, Otherwise perfect.
Try calling them into the scene one at a time and use "aerith with brown braided hair, and another woman with blue skin and no hair" or whatever the other person looks like, it might help. I'll work on the character separation for the next version
Or if you're able call them into the scene one at a time in two separate shots, calling multiple characters in a single prompt is tough
I didn't know you can train character with clothes at same time. In Civitai, there are 2 different sections (character and clothing). Did you do it locally ? How does it need to be proceeded?
I use musubi tuner and caption the outfits specifically, unless its a characters iconic look you expect them in every time then I usually dont caption that so that the lora associates it with the trigger word. I'm still experimenting a bit with what works better, describing the outfits or letting it learn them.
I don't know too yet. For the moment, I'm spending time to generate my character with same outfit each time. In theory I should save time by making a lora with character + outfit + accessories to get the consistency each time. I noticed on your videos sometime the hoodie or the shirt doesn't have the exact same color.
for your consideration, damiane from crimson desert!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.