👋 If you like what I do and want to support the development, feel free to buy me a coffee:
Hello, you’re probably wondering: why so many versions?
Well… I’d be asking the same thing if I were in your place. The reason is simple: it’s designed this way to offer more control, since, unlike normal LoRAs, DMD2 works best at its maximum strength.
For example:
• HD 1 CFG Scale has “diluted” strength, so it requires the help of triggers or manually increasing its LoRA strength. This makes it very useful for combining with PDXL LoRAs in Illustrious, since you can simply raise the strength without losing details.
• DPM A1 and DPM A15 already come with boosted strength and detail, so they don’t require triggers. A1 is the standard strength, while A15 adds an extra +15%.
• V4 is an experiment to generate images in 2 steps. It was created in the opposite way to HD 1 CFG: instead of reducing strength to improve stability, V4 increases strength by 1.35 ratios (20 more than DPM A15).
•
• V5 Enhancing checkpoint details, emphasizing styles, and making prompts more effective without changing your model.
•
• V6 It improves colors and details without changing the style; its sweet spot is cfg scale 1.
• V7 (Visual Only): Built upon the foundations of V6 but heavily cleaned up to focus strictly on visual aesthetics. It sacrifices some cognitive prompt understanding (intelligence) to prioritize pure style.
• V7.5 FIX (The Structural Update): The definitive, fully integrated version. It restores the deep geometric understanding and strict prompt obedience that the basic variants lacked. It features a massive parameter increase to fix the "blindness" of earlier versions, making it highly accurate for complex prompts. (Tip: Because of its dense structure, if you notice issues with hands at extremely low step counts like 3, simply increase your steps to 4-6 or lower your CFG scale to 0.6 - 1.4).
•
In short: it all depends on your taste and goal. For example, V4 will produce more “noise