[LoRA] TigerstripeAI
模型介绍：This is my very first LoRA attempt. I got up the nerve and tried my hand at making one thanks to the wonderful Konyconi-LoRA tutorial. If you haven't already seen their LoRAs, you owe it to yourself to check them out, they're fantastic and I always look forward to their next release. I also used the excellent guide and training Colab from holostrawberry.
About : My LoRA "Tigerstripe" will do just that, add tiger stripes to various objects in the image reasonably well. It excels with cars and trains and buildings but even then I find I might have to add the word "painted" or "colored" sucn as "TigerstripeAI painted dirigible" to get it to work well. can be a bit cantankerous when it comes to certain objects such as rifles or toilets or just about any object where SD expects to see a human holding it or on it. It will put one there (most often a human with tiger stripes and a tiger's head) most of the time unless you use the negative prompt to block it out. Sometimes a tiger tail might appear. I am not sure if this is SD or my failure somewhere in the settings during training - there were no images in the training set with tiger heads or tails, but I only have a thin grasp on what I am doing so it's likely me. If anyone has any ideas on what I did wrong, let me know. Trained using 100 images for 4 repeats, 20 epochs, and 2 batches.
It works best in most cases with a strength between 0.6 and 0.9
~~~~~~~~~~~~~~~~~~~~~~~o(*￣︶￣*)o~~~~~~~~~~~~~~~~~~~~~~~
示例提示词和参数：
Prompt: <lora:TigerstripeAI-15:0.8>TigerstripeAI coffee machine
Negative prompt: EasyNegative, sepia,
Steps: 45, CFG scale: 7, Sampler: DPM++ 2M Karras, Seed: 1778609929, ENSD: 31337, Size: 512x512, Model hash: 4199bcdd14, Hires steps: 15, Hires upscale: 2, Hires upscaler: R-ESRGAN 4x+, Denoising strength: 0.45, Clip skip: 2
页:
[1]