So yes, a lot custom nodes, Sage Attention and Torch comp (you can disable), heavy cause Wan2.2 but i run it with RTX4080.The full workflow : txt to img + Refiner + Detailers hands + Detailer face, take for me around 10 min. But 2K Image with really good quality, you can do Anime or realistic with this settings both work great. And it NEVER failed, i mean let's the full run and you'll never see a bad hands or other ai mistake.You can accelerate the process by passs the Detailer, btw the hands or face will skip if none is detected (for a landscape by example). There is 3 WF in this :1 - WAN2.2 T2i Detailers.json (Without LoRA section and my custom node)2 - WAN2.2 T2i Detailer FULL ok.json 3 - WAN2.2 Refiner and Detailer Only.json (Only Refiner and Detailer)(4 - Custom node for the "FULL") the custom node work like a wildcard, if you let the field of the categories empty it will randomise it

Files with SHA256: 0855e6208c4b882da75783b9f9ac63b136270ace73afaed7a20d4cf24225b199