hinablue/[Krea2 Turbo] AsianMix Lora

Recommend: Sampler: er_sde Scheduler: simple/sgm_uniform/beta Steps: 8 - 12 v5.0 - CPO 加入 1,200 張新資料，使用 TQD 訓練後，與 v4.0 權重做 SVD 融合，再使用 FlowCPO 做 50 組小資料集的後訓練。整體表現與 v4.0 水準差不多，但在色彩、背景、結構上有小幅度的優化。 After incorporating 1,200 new dataets and training with TQD, the model was merged with the v4.0 weights via SVD fusion, followed by post-training on 50 small paired datasets using FlowCPO. Overall performance is comparable to v4.0, though there are slight improvements in color, background, and structure. 1,200件の新規データセットを組み込み、TQDを用いて学習を行った後、SVD 融合（SVD fusion）によってモデルをv4.0の重みと統合しました。続いて、FlowCPOを用いて小規模なペアデー

<iframe src="https://silytics.si/embed/item/6ba96456-0b0b-498e-b91d-870370a214d6" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Civitai and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.