tutorials VS defuser

Having been generating very large batches for character training (per this tutorial which worked really well for me the first time), it occurs to me that the lack of interactivity of the process might make it an ideal use case for ComfyUI, and the lower overhead of ComfyUI might make it a bit quicker.

I created a TI to produce character images for a TTRPG campaign following this great tutorial by u/BelieveDiffusion. I didn't aim at an esotheric level of consistency just enough that seeing a couple of images would show a recognizable person. This will not hold up to close inspection or side by side comparision but I didn't even try to because a) we're playing TTRPG not image coparision b) people change over time and c) it's out of my skill level.

See: https://github.com/BelieveDiffusion/tutorials/tree/main/consistent_character_embedding

Hi guys, I'm very new to this (set up A1111 yesterday) and I'm trying to train the AI into generating pictures of myself for entertainment. I have collected a bunch of pictures and followed this guide but none of the pictures it generated match (probably because that's not the inteded use of the guide). I've googled a bunch and found guides on using LORA an Dreambooth but I've got no clue on which ones are up to date and best to follow. Would love some advice here! I have 25-ish pictures like the guide I followed suggested, if that matters for the recommendation. Thanks!

To create a consistent character, the two primary methods are creating a LORA or a Textual Inversion. I will not go into detail for this process, but instead focus on what you can do to get the most out of an existing Textual Inversion, which is the method I use. This will also be applicable to LORAs. For a guide on creating a Textual Inversion, I recommend BelieveDiffusion’s guide for a straightforward, step-by-step process for generating a new “person” from scratch. See it on Github.