digraph G {
    newrank=true
    compound = true
    
    start [label="Hires fix: which pass to enable refiner for",shape=none]
    
    first [label="1st pass" shape=none]
    hires [label="2nd pass" shape=none]
    
    ckpt1 [label="ckpt"]
    ckpt2 [label="ckpt"]
    ckptB [label="ckpt"]
    refck1 [label="Refiner ckpt"]
    refck2 [label="Refiner ckpt"]
    refckB1 [label="Refiner ckpt"]
    refckB2 [label="Refiner ckpt"]
    hrck1 [label="HiRes ckpt"]
    hrck2 [label="HiRes ckpt"]
    hrckB [label="HiRes ckpt"]
    
    {rank = same; first; ckpt1; ckpt2; ckptB; }
    {rank = same; hires; hrck1; hrck2; hrckB; }
    
    start -> ckpt1 [lhead = "cluster_1set"]
    start -> ckpt2 [lhead = "cluster_2set"]
    start -> ckptB [lhead = "cluster_Bset"]
    
    subgraph cluster_1set {
        label="first pass"
        ckpt1, refck1, hrck1
    }
    subgraph cluster_2set {
        label="second pass"
        ckpt2, hrck2, refck2
    }
    subgraph cluster_Bset {
        label="both pass"
        ckptB, refckB1, hrckB, refckB2
    }
        ckpt1 -> refck1 [label=n>]
        refck1 -> hrck1
        
        ckpt2 -> hrck2
        hrck2 -> refck2 [label=n>]
        
        ckptB -> refckB1 [label=n>]
        refckB1 -> hrckB
        hrckB -> refckB2 [label=n>]
        
}

アニメモデルで出してリアルモデルでHires.fixする。ついでにHiRes PromptsのほうだけでLoRAを切るとかもできる。ノイズ除去強度を高くすればHiresモデル寄りにもできる

ほぼ城ヶ崎美嘉なんだからmedium breastsでいけるって！　って囁くのよ、私のゴーストが⋯⋯