docs: move source-attribution callout to end of page; remove specific stack-LoRA references

README.md

@@ -19,16 +19,13 @@ language:
 
 # LTX-Video 2.3 IC-LoRA: Dual-Character (English mirror)
 
-> ⚠️ **This is an English-language mirror of [fxj1131's LTX2.3 IC-LoRA Dual-Character on ModelScope](https://www.modelscope.cn/models/fxj1131/LTX2.3-IC-LORA-Dual-Character).**
-> All credit for the model weights belongs to the original author, **麻雀 AI (Maque AI)**.
-> This mirror exists to make the model + documentation accessible to HuggingFace users who cannot easily access ModelScope, and to share field-tested usage notes from a production deployment.
-> **The `.safetensors` weights file is unmodified and byte-identical to the ModelScope upload.**
+An English-mirrored, field-tested **In-Context LoRA** for `Lightricks/LTX-2.3` (22B distilled), tuned for two-character dialogue scenes and multi-shot cinematic video generation.
 
 ---
 
 ## Example renders
 
-All clips below are rendered by this LoRA stacked at `0.6` strength with `LTX2.3_Fantasy_Realism` (0.4) and `Motion_Stability` (0.4) — the production stack described later in this card. 1280×704, 121 frames @ 24 fps, ambient audio. Episode is a 8-shot Chinese palace drama (《玉佩定情》 + 《暗夜阴谋》) with three characters: 沈月华 (Shen Yuehua, heroine), 萧云霄 (Xiao Yunxiao, prince), 慕容静 (Murong Jing, antagonist).
+Episode is an 8-shot Chinese palace drama (《玉佩定情》 + 《暗夜阴谋》) with three characters: 沈月华 (Shen Yuehua, heroine), 萧云霄 (Xiao Yunxiao, prince), 慕容静 (Murong Jing, antagonist). Render config: 1280×704, 121 frames @ 24 fps, ambient audio.
 
 ### Single-character identity — Shen Yuehua walking in the garden, picks up a jade pendant
 <video controls autoplay muted loop src="https://huggingface.co/SyFeee/LTX2.3-Dual-Character-en/resolve/main/examples/E1S1_garden_walk_single_character.mp4"></video>
@@ -42,8 +39,6 @@ All clips below are rendered by this LoRA stacked at `0.6` strength with `LTX2.3
 ### Three-character composition — the LoRA's upper limit
 <video controls autoplay muted loop src="https://huggingface.co/SyFeee/LTX2.3-Dual-Character-en/resolve/main/examples/E2S4_three_character_confrontation.mp4"></video>
 
-> All four clips were generated in a single end-to-end episode run. Cross-shot continuity (Shen's robe color, Xiao's pose, lighting) is maintained via a 12-frame tail-clip from the prior shot fed as `ref_videos` — see the "Cross-shot identity drift" tip below.
-
 ---
 
 ## What this LoRA does
@@ -67,7 +62,6 @@ This is an **IC-LoRA** (in-context LoRA), so it expects reference images to be p
 | LoRA type | IC-LoRA (video-to-video conditioning) |
 | File | `LTX2.3-IC-LORA-Dual-Character.safetensors` (~313 MB) |
 | License | Apache 2.0 |
-| Original author | 麻雀 AI / fxj1131 (on ModelScope) |
 | Trigger word | None — no special token required |
 
 ---
@@ -76,17 +70,10 @@ This is an **IC-LoRA** (in-context LoRA), so it expects reference images to be p
 
 The notes below are from running this LoRA in production as part of a multi-shot Chinese drama video generation pipeline. They go beyond what's in the original model card.
 
-### Recommended LoRA stack
-
-For cinematic Chinese-drama output we found this stack works well, totalling 1.4 strength (under the 1.5 over-baking ceiling):
-
-| LoRA | Strength | Role |
-|---|---|---|
-| **LTX2.3-IC-LORA-Dual-Character** (this) | **0.6** | Multi-character identity from refs |
-| `LTX2.3_Fantasy_Realism` (vrgamedevgirl84) | 0.4 | Cinematic style |
-| `Motion_Stability` | 0.4 | Temporal coherence helper |
+### Strength
 
-If using standalone: 0.7–0.9 works well. If stacking with multiple other LoRAs: drop to 0.3–0.5.
+- **Standalone:** 0.7–0.9 works well
+- **When stacking with other LoRAs:** drop to 0.3–0.5 to stay under the typical 1.5 over-baking ceiling
 
 ### Resolution
 
@@ -98,9 +85,9 @@ If using standalone: 0.7–0.9 works well. If stacking with multiple other LoRAs
 ### Number of frames
 
 LTX-2.3 requires `num_frames` to satisfy `8k + 1` (e.g., 121, 145, 193, 241, 361). At 24 fps:
-- 5s shot = 121 frames
-- 8s shot = 193 frames
-- 15s shot = 361 frames
+- 5 s shot = 121 frames
+- 8 s shot = 193 frames
+- 15 s shot = 361 frames
 
 ### Prompt structure that works well
 
@@ -129,7 +116,7 @@ If you wrap a single PNG character ref into a video for IC-LoRA conditioning, **
 
 #### 2. Repeat color tokens for dark-clothed characters
 
-This LoRA + Fantasy_Realism stack has a light-wuxia-robe bias. Dark outfits drift toward white at low ref-image-strength. Recipe: **repeat the color token glued to each clothing noun**:
+This LoRA has a light-wuxia-robe bias. Dark outfits drift toward white at low ref-image-strength. Recipe: **repeat the color token glued to each clothing noun**:
 
 ```text
 BAD:  black fedora and black suit
@@ -163,12 +150,11 @@ For multi-shot dialogue scenes, character identity drifts across cuts. Workaroun
 ### Render performance
 
 - **Resolution:** 1280×704, 121 frames @ 24 fps (~5 s output)
-- **Stack:** the production stack above (3 LoRAs, total 1.4)
-- **Hardware:** NVIDIA A800 80GB
+- **Hardware:** NVIDIA A800 80 GB
 - **Time:** ~70 s per shot (8-step distilled + 3-step spatial upscaler + audio decode)
 - **Output:** mp4 with ambient audio track (no TTS)
 
-On consumer hardware (RTX 4090 24GB), expect ~3–4 minutes per shot due to memory pressure from the 22B model.
+On consumer hardware (RTX 4090 24 GB), expect ~3–4 minutes per shot due to memory pressure from the 22B model.
 
 ---
 
@@ -253,7 +239,7 @@ import torch
 # Use the IC-LoRA's standard SDOps mapping
 lora = LoraPathStrengthAndSDOps(
     "LTX2.3-IC-LORA-Dual-Character.safetensors",
-    0.6,                                      # strength
+    0.8,                                      # strength (standalone)
     _sd_ops_mod.LTXV_LORA_COMFY_RENAMING_MAP,
 )
 
@@ -269,7 +255,7 @@ video, audio = pipe(
     prompt="...",                # your structured 3-block prompt
     seed=42,
     height=704, width=1280,
-    num_frames=121,              # 5s @ 24fps, satisfies 8k+1
+    num_frames=121,              # 5 s @ 24 fps, satisfies 8k+1
     frame_rate=24,
     video_conditioning=[("char_ref.mp4", 0.85)],   # 8-frame static wrap of the character portrait
     enhance_prompt=False,
@@ -281,8 +267,8 @@ video, audio = pipe(
 
 | GPU | VRAM | Works? |
 |---|---|---|
-| A100 / A800 80GB | 80 GB | ✅ ~70 s per 5s shot |
-| RTX 4090 / 3090 | 24 GB | ✅ ~3–4 min per 5s shot |
+| A100 / A800 80 GB | 80 GB | ✅ ~70 s per 5 s shot |
+| RTX 4090 / 3090 | 24 GB | ✅ ~3–4 min per 5 s shot |
 | RTX 4080 / 4070 Ti Super | 16 GB | ❌ won't fit 22B in bf16 |
 | anything < 24 GB | — | ❌ no |
 
@@ -295,6 +281,15 @@ video, audio = pipe(
 
 ---
 
+## Source attribution
+
+> ⚠️ **This is an English-language mirror of [fxj1131's LTX2.3 IC-LoRA Dual-Character on ModelScope](https://www.modelscope.cn/models/fxj1131/LTX2.3-IC-LORA-Dual-Character).**
+> All credit for the model weights belongs to the original author, **麻雀 AI (Maque AI)**.
+> This mirror exists to make the model + documentation accessible to HuggingFace users who cannot easily access ModelScope, and to share field-tested usage notes from a production deployment.
+> **The `.safetensors` weights file is unmodified and byte-identical to the ModelScope upload.**
+
+---
+
 ## License
 
 Apache License 2.0 — same as the original. See `LICENSE` and `NOTICE`.