Daankular commited on
Commit
7065e43
Β·
verified Β·
1 Parent(s): 8b38980

Add identity-preserving BLINK LoRAs (NSFWcode, lopi999, Outfit), fix -H-/-L- pairing

Browse files
Files changed (1) hide show
  1. loras.txt +28 -0
loras.txt CHANGED
@@ -86,6 +86,34 @@ I2Pee-V1 [obsxrver] | all_releases/Version_1/I2Pee-V1_Wan2.2_high.safetensors |
86
  # UnifiedHorusRA/WAN_2.2_i2v_LORA_-_Mastram_-_...
87
  # INVALID repo ID (too long / illegal chars) β€” cannot be fetched via HF API
88
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
89
  # ── TO ADD ────────────────────────────────────────────────────────────────────
90
  # Add new entries here in the same format, then push loras.txt + update
91
  # EXTRA_LORA_REPOS list in app.py if the repo isn't already there.
 
 
 
 
 
 
 
 
 
86
  # UnifiedHorusRA/WAN_2.2_i2v_LORA_-_Mastram_-_...
87
  # INVALID repo ID (too long / illegal chars) β€” cannot be fetched via HF API
88
 
89
+ # ── NSFWcode-com β€” BLINK / IDENTITY-PRESERVING ────────────────────────────────
90
+ # These train cross-attention layers (same IIAM methodology as obsxrver/I2Pee).
91
+ # Trigger: "a [man|woman], jumpcut, after the transition, [he|she] is [action]"
92
+
93
+ NSFWcode_BLINK_Handjob_I2V_V01 [BLINK] | NSFWcode_com_WAN22_BLINK_HANDJOB_I2V_V01_HIGH.safetensors | NSFWcode_com_WAN22_BLINK_HANDJOB_I2V_V01_LOW.safetensors | NSFWcode-com/NSFWcode_com_WAN22_BLINK_HANDJOB_I2V_V01
94
+ NSFWcode_Oral_Insertion_V01 | NSFWcode_com_WAN22_ORAL_INSERTION_V02_HIGH.safetensors | NSFWcode_com_WAN22_ORAL_INSERTION_V01_LOW.safetensors | NSFWcode-com/NSFWcode_com_WAN22_ORAL_INSERTION_V01
95
+
96
+ # ── lopi999 β€” GENERAL NSFW (56 likes, 13k downloads) ─────────────────────────
97
+ # Highest-download general NSFW enhancer for Wan 2.2 I2V 14B.
98
+ # Keys: NSFW-22-H-e8 / NSFW-22-L-e8 β€” now correctly paired with updated -H-/-L- detection.
99
+ # Trigger: "nsfwsks"
100
+
101
+ lopi999_General_NSFW | NSFW-22-H-e8.safetensors | NSFW-22-L-e8.safetensors | lopi999/Wan2.2-I2V_General-NSFW-LoRA
102
+
103
+ # ── UnifiedHorusRA/Outfit_Transformations ─────────────────────────────────────
104
+ # Identity-preserving by design β€” person stays, outfit/clothing changes (strips).
105
+ # Single-file LoRA (Wan 2.2, high-noise transformer only in name but treated as single).
106
+
107
+ Outfit_Transformations [single] | RX_Wan2.2_HightNoise_TranKoolth_epoch64.safetensors | RX_Wan2.2_HightNoise_TranKoolth_epoch64.safetensors | UnifiedHorusRA/Outfit_Transformations
108
+
109
  # ── TO ADD ────────────────────────────────────────────────────────────────────
110
  # Add new entries here in the same format, then push loras.txt + update
111
  # EXTRA_LORA_REPOS list in app.py if the repo isn't already there.
112
+ #
113
+ # Authors to watch for new identity-preserving BLINK LoRAs:
114
+ # obsxrver, NSFWcode-com, PetruZetta (mirrors lkzd7 iGoon files though)
115
+ #
116
+ # Identity preservation mechanism: LoRAs that train cross_attn layers
117
+ # (lora_unet_blocks_N_cross_attn_q/k/v/o) bind the input image's appearance
118
+ # to the generated frames. LoRAs that only train self_attn/ffn modify motion
119
+ # but don't strongly preserve the input image person's identity.