PsyOpsAI art
patreon
Failures (luxillia, liliraka, lilibriar)
🕑 Added 2025-04-09 19:33:03 +0000 UTCNormally I'd delete these, but I figured some of you might still enjoy seeing them.
~~~~~~~~~~
So, duo gens, as you can tell, fail a lot. Especially when you try to automate them in bulk with a streamlined workflow. And while I know exactly why they fail and how to fix it, I often don't… because, weirdly enough, this kind of instability is part of the intended design.
I'm not aiming for just one or two images like most AI artists doing duo commissions. I aim for a lot of high quality results, which takes a clever, headache inducing setup. Usually a precise cocktail of extensions (Forge couple, ControlNet), prompts, and sensitive settings.
Pro: When it works, it really works. Duo gens with 10+ images, often with little to no inpainting needed.
Con: When even one setting is off, it breaks. Also some character combos just don’t play nice, especially complex or unique ones with tricky LoRA interactions. It slows everything down.
So what’s next? I’m considering switching to a more traditional approach: one or two images per duo, but much more manual. High res, detailed inpainting, complex scenes, custom backgrounds. Just more focused, deliberate art.
Why not do both? Because clarity matters. Let’s be real, not many people read long posts (I barely do either), and I’d rather keep things simple and clear.
I'll do a poll about it later
Comments
Zachary Fitzgerald
Dope
