Why check the so called "vpred" placeholder layer, where entire WebUI never use them?
Just use the vpred node, instead of relying the so called "auto detection" (github issue).
For all ComfyUI users, it is your responsibilty to adapt my model in 2026, it is not my responsibilty to insert placeholders to entertain your workflow.
Whatever how "experienced" you are, there are always new models to screw you up and force you to adapt along with ComfyUI.
Pay attention to the "some anime models", Astolfo is different. Astolfo is the one who making history. I trained my model via Kohyas intead of NaifuDiffusion, which Kohyas obviously don't have this layer.
Moreover, there are SDXL RF models which already required SD3 node. I think I can require the vpred node.
Therefore, All my models will not support ComfyUI. Long live AUTOMATIC1111.
PS: Generated by AstolfoGPT with context windows of 2 brain cells.