Greetings everyone, IMPORTANT NOTICE! Before starting, download and set the DMD2 lora correctly! I added notes next to the lora stack and the EasyLoader nodes. (Unless you want to up the step count and forego the DMD2 lora altogether). TLDR; - Quick generation times (3 steps+ depending on the sampler and settings) - Prompt and go - Two step inference - Illustrious, SDXL 1.0 and PonyXL compatible (others also possible, if DMD2 is dropped) - Split lora stacks for each sampler - Separated prompts - 2 sliders per sampler - Embedded Civitai metadata - Based on DMD2 usage, but not required (open the EasyLoader (full) node and set the lora to none) - RES4LYF node pack required - I tried to keep the required node packs to a minimum This workflow aims to provide an intuitive interface with easy to understand controls, that modify the DMD2 weights and the noise injection between steps. No need to understand what any of that means, just play with the 4 included sliders. This workflow also features a two step sampling process on the same latent without decodes in between. This provides a way to vary the intensity of parts of the inference process, while still leaving room for a "tame" last few steps of the inference process. CFG is set to one for lightning fast performance. Adjust if you like, but that will make it more complex. The main goal is to have just 4 values and the step distribution, plus prompt to worry about. I grouped all the most important controls to the left of the graph for easy access. Modify the folder/filename to whatever you like. Optimized for 1:1 or 2:3 aspect ratios—drastic resolution changes might require shifting the step splits and base ranges of the sliders. Check the formula next to the slider for that.

Files with SHA256: 346a5ce88456e0682bfa9354422a9f677f362098fb746cadbf0070fda5364770