This workflow contains 4 groups for different use-cases. To activate one, right-click the group and click "Set group nodes to Always".You will need to install many custom nodes, you might want to read the docs of the following repositories :<ul><li><a target="_blank" rel="ugc" href="https://github.com/filliptm/ComfyUI_FL-Trainer">https://github.com/filliptm/ComfyUI_FL-Trainer</a></li><li><a target="_blank" rel="ugc" href="https://github.com/pythongosssss/ComfyUI-WD14-Tagger">https://github.com/pythongosssss/ComfyUI-WD14-Tagger</a></li><li>For PNG conversion install manually : <a target="_blank" rel="ugc" href="https://github.com/doomy23/ComfyUI-D00MYsNodes">https://github.com/doomy23/ComfyUI-D00MYsNodes</a></li></ul><h2 id="group-1-:-png-conversion-049nrlcch">Group 1 : PNG Conversion</h2>Inputs :<ul><li>Path of the images to convert</li><li>Path of the outputs</li><li>Format (set to PNG)</li></ul><h2 id="group-2-:-captioning-qpd95wnfy">Group 2 : Captioning</h2>Inputs :<ul><li>Path to the dataset folder containing all the PNGs (IMPORTANT : Must be PNG, not JPEG or anything else).</li><li>Trigger words and appended words to the results.</li></ul>It will tag all the images one by one and create a TXT file with the same name as the PNG image.<h2 id="group-3-:-training-configuration-iciicavxh">Group 3 : Training configuration</h2>Inputs :<ul><li>Path to the dataset folder with the PNGs and TXT files.</li><li>Training prompt.</li></ul>I've set the "FL Khoya Adv Config" settings to the same used by <a target="_blank" rel="ugc" href="http://CIVITAI.com">CIVITAI.com</a> in the "Train a LORA" function of the site but feel free to add/remove network dim and network alpha, try different optimizers and so on...<h2 id="group-4-:-testing-the-model-kl4oo0ay9">Group 4 : Testing the model</h2>Inputs :<ul><li>Your checkpoint</li><li>Your new LORA model</li><li>Prompts</li><li>Size and batch</li></ul>It is setup for SD15 models but can easily be changed to SDXL model loaders.

Auto-caption and FL Trainer for LORA

<ul><li>SD15 LORA training works with 10 epoch or so (tested)</li></ul>

Files with SHA256: 6440f0c953decaa2b35cabdc011acc548de797e68f86e72015fea8f10241e998