FIRST UPLOAD, SO IF YOU USE THIS LORA, PLEASE POST YOUR GENERATIONS! I'D LIKE TO SEE WHERE IT COULD BE BETTER! 🙏
⚠️ SORRY FOR LARGE FILE SIZE - WILL RETRAIN AND REUPLOAD IF QUALITY IS OK
I know this most likely doesn't have any real world use case, it was meant as a proof of concept for me for now.
Tested Settings (but feel free to play around):
Width x Height: 1216 x 811
Method: Euler a
LoRA Weight: 0.8 - 1 (depends on prompts used)
Trigger prompt: Head Texture
Sampling steps: 20+
CFG Scale: 5-9 (depends on prompts used)
Positive Prompt Template:
Head Texture, natural skin texture, face center, (any additional facial features), <lora:head_texturesXL_V1:1>
Negative Prompt Template:
layman work, worst quality, poorly drawn, bald, bald spot, earrings, teeth, smile, open mouth, ears covered, forehead covered, hair on forehead, bangs
Observations:
This model seems to have a tendency to make male head textures with a bald spot, so if you find that happening, try pulling the lora weight down a bit.
Alternatively, if you find the lora starts to generate normal heads (not a head texture), just bring the lora weight up. It seems to vary between 0.8 to 1 depending on the prompt you use.
I also find it's sometimes better to leave out the description of the hair, unless you really want something specific, otherwise it may generate too much hair that cover the ears.
Lora also seems to work well with DreamShaperXL if you want less humanlike generations (e.g. vampire, troll, clown, etc.).
Description
FAQ
Comments (17)
UV map for 3D？
No. It literally says texture map. UV is a coordinate system.
@olivetty The texture is mostly pointless without the UV map layout, not to mention you'd have to match the UVs on your own models (unless it's a prefab model from some popular software, which is why everyone is asking for the UV layout info...). I guess it ight be ok if you don't care much about the accuracy in your 3D model...
@Alverik bro, use substance painter.
Great idea, but a color texture map and the corresponding UV map would probably have to be generated together, so that they match. I don't know if Stable Diffusion can output two related bitmaps?
❤🧡💛💚💙💜🤎
Great idea! I would be interested in seeing the 3d models you can make with this
This is a great idea. I tried to get rid of glossy reflections and other things you don't want on a head texture. This is the really hard part, when you create a texture manually. Unfortunately, this didn't work. Anyway, great concept.
Thanks for trying it out! It definitely has some issues, but hopefully I'll have some time to work those out at some point. The hardest part is finding decent training images of head textures..
Wow that's amazing I not expect that is really cool
what UV map and Model are these trained for?
its pretty obvious it isnt trained for a specific uv map or model - you good?
InvokeAI dont see your Lora if the name contains spaces or capital letters :(
If i change name to correct, Invoke AI has error lora loader :(
Can you cgange name of this Lora correct? (no spacec and little letters please)
tried to use it. it tried to contact huggingface
lovely work. now if we only can do bodies :)
Train this on real headscans where the lighting info isnt baked into the the diffuse. then itll actually be useful! I've been looking for something like this for a while but it has to be removed of lighting info to be useful
i think its good for making mod for unreal toranment game ,, may be you dont know it
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.