🙏👍*** If you like it, please like it! ***👍🙏
A model of Twitch Streamer and OF model, Hannahowo. This is based on her more ahem 'enhanced' appearance. This was trained with around 50 close-up face images to ensure her likeness remains intact. Like most Lora with physiques, the body can be stubborn to trigger (especially in certain clothes), so I made sure the portrait would always be accurate like a normal face Lora.
Please let me know in the comments if you run into any issues! I'm always looking to improve.
Best Results (so far) = Euler/beta (my preferred) or Euler/normal ; 25 to 30 steps; Guidance 3.5 (approx.); CFG 3.5 (approx.)
I haven't tested other Samplers/Schedulers much, so let me know if you produce good results!
IMPORTANT/KNOWN ISSUES:
Body Conflicts - Because this Lora has a body, it can conflict with other Lora that have bodies as well. To combat this, simply lower the strength on the second Lora. A strength around .6 to .9 worked on most secondary Lora I tried, so staying below 1.0 on secondary body Lora is recommended.
Sitting Tendancy - If you don't use much description, in certain types of images the Lora likes to sit by default. This is because many of her pics in general are seated, so the model has a tendancy to appear seated as well. You can typically get around this by adding to the prompt things like mentioning her tattoos (see Trigger Words below), adding footwear, and (most important) describing the pose. It's an annoyance, but not Lora breaking.
Selfie Tendancy - Similar to the Sitting Tendancy above, many of the training photos are selfies because those are primarily what's available. Like sitting, it can be prompted out typically just by describing the type of photo you want. Close-ups especially tend to do this, so prompting is key. If you want a full body shot, 3/4 body, etc; you may need to add it to the description in some cases.
Abodmen Tattoos - This was trained with her lower abdomen tribal tattoos. However, tattoos are very difficult to replicate. Her lower abdomen tattoos typically appear without prompting, but are more accurate when prompting for them. So I suggest prompting for them when needed. If they start appearing in other places, I suggest not prompting. See Trigger Words below for more.
Thigh Tattoos - Her thigh tattoos I haven't had much success with, so I don't suggest prompting for them. For those, a tattoo appears in the right place, but the checkpoint will just fill in a generic tattoo or try to replicate the tribal tattoos.
Septum Ring - Her septum ring will often appear in pictures because there is rarely a photo without her wearing it. If you aren't prompting for it, and you have a strange dent in her nose/lip where it should appear, or a generic stud piercing in her upper philtrum, I suggest prompting for it. It should give a clear result. If you're okay editing that out, go for it.👍See Trigger Words below for more.
Describe, Describe, Describe - That is the primary key to getting good results. If you use short descriptions or don't add much detail, the less likely you will get good results. See my image data for examples.
TRIGGER WORDS (also important)
"Hannahowo" = duh 😜
"Freckles" or "Freckles across her nose and cheeks" = Her freckles tend to appear regardless, but this will ensure they do.
"Egirl makeup" = This should trigger more pronounced makeup with long lashes and winged eyeliner. It already appears much of the time, but this ensures it.
"Small" or "Horseshoe" + "Septum Ring" or "Septum Piercing" = These usually result in her septum ring appearing much more prounced and complete. She has two kinds (eg. "Small" or "Horseshoe" - but the prompting doesn't always produce the one you want. Instead, I use the descriptors to ensure a complete ring appears. (see Important/Known Issues for more)
"Ample Breasts" or "Medium Breasts" = If I have to explain it to you, you don't need to know.🤐
"Tribal tattoos on her lower abdomen" = Usually (not always) produces a pretty good likeness in Flux 1.D.