Training SDXL to Generate Text Using IA3 LoRA | It's like Kai's Power Tools, I Guess? скачать в хорошем качестве

Training SDXL to Generate Text Using IA3 LoRA | It's like Kai's Power Tools, I Guess?

#sdxl #stablediffusionart #generativeai Can you reliably generate text with Stable Diffusion XL? Maybe? Let's explore this "new" method. I try training a SDXL IA3 LoRA using Prodigy to guide text generation while trying to minimize the number of parameters disrupted by training. [00:00] Intro/topics [00:28] Image generation flaws/difficulties [00:36] Other text generation methods [01:05] LoRA types VS Dreambooth, and my limitations [01:40] SD1.5 VS SDXL [02:00] Briefly discussing the project [03:20] Going over the dataset [05:50] Python script for generating PNGs from TTF files [08:16] Dataset pruning ]09:11] Test run 1 [11:13] Test run 2 [12:23] Test run 3 [12:30] Summary [14:00] Using the IA3 LoRA in the Auto1111 WebUI with some before/after comparisons Test IA3 Text-Image Adapters: https://huggingface.co/AOLCDROM/iA3-a... More about the Prodigy optimizer: https://github.com/konstmish/prodigy More about the IA3 PEFT adapter: https://huggingface.co/docs/peft/conc... T-shirts dataset: https://www.kaggle.com/datasets/sunny... More on Pangrams: https://www.prdaily.com/16-clever-pan... Font to Image Python Script: http://nanonomad.com/2024/05/07/train... #sdxl #stablediffusionart #generativeai