Part 1 of my Lab series. Checkout my other workflows on my profile, they can all be used in conjunction!<h2 id="3-step-auto-lora">3 Step Auto Lora</h2><ol><li>Set your batch number hit run and generate 100s of hi-res images with auto captions!</li><li>Automatically store them in a consolidated folder</li><li>Feed it to a Lora Trainer!</li></ol>I recommend Auto MED. Fully automatic, no manual packaging, no real loss in accuracySemi-Auto HI allows for much more manual interaction for those willing to put in the work for a slight increase in caption qualityAuto LO is for those with a low VRAM GPU. It will require loading one ksampler model and one interpreter model (Florence-Basic or Large)<a target="_blank" rel="ugc" href="https://github.com/anyantudre/Florence-2-Vision-Language-Model">https://github.com/anyantudre/Florence-2-Vision-Language-Model</a><h3 id="running-versions">Running Versions</h3>____________________VRAM ______ACCURACY_____ MANUAL WORKSemi-Auto HI _________HI_____________HI______________ HIAuto MED ____________HI- ___________HI- _____________LOWAuto LO ____________MED __________MED _____________LOWSemiAuto HI: Dual inference models gets higher accuracy in prompts, but requires manually consolidating your folder and choosing the best prompt for each image. Dual Face refiner and refiner base model with 4x upscale creates hi-realism and accuracy.Auto MED: Single inference model allows for auto compilation and requires no manual package construction. However it will slightly reduce accuracy (not noticeably IMO) in auto prompts. Dual Face refiner and refiner base model with 4x upscale creates hi-realism and accuracy.Auto LO: Single inference model allows for auto compilation and requires no manual package construction. However it will slightly reduce accuracy (not noticeably IMO) in auto prompts. No Face refiner and refiner base model with 2x upscale creates med-realism and accuracy.When filling in the prompt buckets DO NOT PUT A SPACE BETWEEN COMMAS.(trait1,trait2,trait3)<h2 id="how-it-works-(semi-auto-hi):">How it works (Semi-Auto HI):</h2>Workflow Logic<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/389e518c-ef7f-44b5-b5ec-e68cd043605c/width=525/389e518c-ef7f-44b5-b5ec-e68cd043605c.jpeg" />Real Image Example<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/de346696-c181-44bb-9fae-367c1fbf7002/width=525/de346696-c181-44bb-9fae-367c1fbf7002.jpeg" />Workflow Prompt buckets<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5656b835-660b-4281-a51c-2c700b460985/width=525/5656b835-660b-4281-a51c-2c700b460985.jpeg" />Workflow results 1 (Random Prompt, Random Orientation, Florence and MAI Results)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ebd019ec-8a6f-458b-a774-e0b083a6f445/width=525/ebd019ec-8a6f-458b-a774-e0b083a6f445.jpeg" />Workflow results 2 (Random Prompt, Random Orientation, Florence and MAI Results)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ca3a19f6-0157-4ea7-a2a6-5231b2761373/width=525/ca3a19f6-0157-4ea7-a2a6-5231b2761373.jpeg" /><h3 id="henry-images-+-best-prompts-auto-lora-data">Henry Images + Best Prompts = Auto Lora Data</h3><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/cb0e4f65-78df-4108-a41e-ca8c4220f39b/width=525/cb0e4f65-78df-4108-a41e-ca8c4220f39b.jpeg" /><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/505a676d-4cba-4b18-9774-c0026b23ab58/width=525/505a676d-4cba-4b18-9774-c0026b23ab58.jpeg" />Lora Training Images saved at: /output/train/image/Lora Text saved at: /output/text/Florence and /output/text/MAIUp-scaled Images saved at: /output/train/source/<h2 id="system-specs:">System Specs:</h2>This workflow is High VRAM out of the boxIf there is enough support/demand I will create Med VRAM and Low VRAM versionsHigh VRAM: Use the large model and fp16/sdpa where you canLow VRAM: Use base models and fp8/flash attention where you can, skip the refiner, skip the upscale, and skip the detailers (might be tricky to re-wire)<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/2e31eb48-7671-4916-a975-6834827341f0/width=525/2e31eb48-7671-4916-a975-6834827341f0.jpeg" /><h2 id="prerequisites:">Prerequisites:</h2>Necessary Custom Nodes: rgthree, easy-use, impact pack, impact subpack, florence 2, MiaoshouAI, basic math, was ns, Crystools, kj nodes, StringOps (Thank you devs!)When you open the workflow ComfyUI-Manager should let you know you are missing nodesClick install missing nodes, restart comfyui, and refresh browserIf there are still missing nodes search for them in the Custom Nodes menu in the Manager windowI also had to install flash attentionEasiest way is to look up your CUDA version (nvidia-smi in cmd) and download the prebuilt whl file here that matches those verions: <a target="_blank" rel="ugc" href="https://github.com/mjun0812/flash-attention-prebuild-wheels?tab=readme-ov-file">https://github.com/mjun0812/flash-attention-prebuild-wheels?tab=readme-ov-file</a><h2></h2>

Lora Lab: Press Start, Auto Gen, Auto Label, Auto Compile

This version removes the dual interpreter build of HI VRAM and auto consolidates the training data into a single folder

Files with SHA256: 11f90b39d27f4739c0aa9ddc680c5352886d5bb34d0eb80564b7079437180e76