This is my first public lora which is generated on a RTX 5090 with 1080x1920 resolution. I figured out that using a higher image resolution then training on a lower resolution, helps image quality. The detail of the skin is WOAH. Also, I just realized that image artifacts/compression affects the quality HUGE, dramatically huge. Hope this info helps.
V0.5 (Recommended)
prompt: dragon, female, red scales
V0.6b (Testing Phase, more stable???)
Using Vinci194 in the prompt isn't necessary
prompt: female, red and black scales, orange hair, long snout
Options Version (75 Images x 10 repeats)
V0.5 & V0.6b
(78 images x 10 repeats)
Prodigy is king
Basically default settings prodigy with cosine
bfloat 16 for training data type
Lora rank 128
Recommended Model (https://civitai.com/models/97479?modelVersionId=1323509)
model base (https://civitai.com/models/1369089?modelVersionId=1546777)
Around 1500 to 6200 steps
Original Creator: https://x.com/vinci194 (Vinci194)
3D Model by: FoxiPaws
Follow the original creator if you enjoy this lora.
They have NSFW in their patreon page here (https://www.patreon.com/VinciTheDragon)
Description
V0.5
Trained on a 34,320 step count (I'm pretty sure this amount of steps are used for training models with like a 2,000 images or so...).
I know it's overkill, but the images I've seen produced from it are almost perfect, it's just missing some minor details.
For now, this Lora is recommend for single character use only.
It took 10 hours to generate (RTX 5090 limited to 400watts or 67% power limit).
I'm actually surprised it worked at that high of a step and didn't seem to overfit the Lora much.
It's missing scale detail on the chest area which is a bit annoying, can be hit or miss.
This lora is unstable due to the high steps and must be used in 1080x1920 and use a strength of 0.65 - 0.70. If you use anything else it will heavily depended on the model.
This Lora might generate duplicate character's or generate a pet creature, getting rid of the word dragon in the prompt should reduce the chance of it reoccurring.
This version aims to increase visual quality.
Tested using Lora Strength 0.65.
This is lora can be unstable on Lora Strength 1.0
Using 78 images x 20 repeats with 22 Epochs
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.