>>67397707I'll deal with this later
Now I will attempt to further prove a lora building method I've already done on two OC loras I won't post; boosting a dataset WITH AI art. This was limited and with pretty awful results on NAI loras but seems to be legitimately worth considering if you're using EF+hll, randomizing artists, and training on fluffyrock; at least, if the character is relatively simple. This can be done with lower denoise img2img or making an initial lora and using its outputs.
>>67398014maybe try asking the /trash/ thread