Add Qwen-Image-2.1 style-ref LoRA (r128, ckpt1950) and aligned-only light-restore LoRA (r128, ckpt3000)

Rank-128 LoRAs (alpha 128) on Qwen/Qwen-Image-2.1, attention + image-MLP projections. Style ref: [content, style] condition order, prompt "apply the artistic style of the second image to the first image, keep the subject and composition of the first image", trained on OmniConsistency + style_pairs_10_each. Light restore: trained on the 187 alignment-checked pairs.

qwen_image_21_light_restore_aligned_r128_ckpt3000.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b0e41224ad7de4aef5a141f3d507fe0b0ff2066e6c2b5df0b0adecd3722015f
+size 671151064

qwen_image_21_style_ref_r128_ckpt1950.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bd5a1a359b91abd9505818d589f996cb96984e177b96af0c4e1db08529624f4
+size 671151064