SyFeee/ltx2.3-chinese-drama-iclora-canny

LTX-2.3 Chinese Drama IC-LoRA — Canny Edge Control

An IC-LoRA (In-Context LoRA) for LTX-Video 2.3 (22B) that conditions video generation on a Canny edge / line-art video so the generated scene's composition follows a user-supplied reference outline. Trained on the same 78-episode Chinese historical drama corpus as the character LoRA, with Canny edges extracted per-frame from the source clips.

Model details

Training data

Identical corpus + caption format to the character LoRA, plus per-clip Canny edge videos extracted via OpenCV at the source resolution. Canny edges capture composition (where things are) and silhouette (subject outlines) without committing to specific identity or texture.

Usage

IC-LoRA inference requires a Canny edge reference video at the target resolution + an inline-weave prompt. You can generate the canny reference from any source video with OpenCV.

LTX ltx_pipelines.ic_lora invocation

python -m ltx_pipelines.ic_lora \
    --prompt "char_0_person. Framed in a wide eye level shot, on a 24mm wide lens, with natural light. Set in a torch-lit Han dynasty palace courtyard, the subject walks slowly forward. Live-action photorealistic, cinematic Chinese drama." \
    --negative-prompt "no CGI, no animation, no illustration, no painterly style, no anime" \
    --lora <path_to>/lora_weights_step_06000.safetensors 1.0 \
    --video-conditioning <canny_reference>.mp4 1.0 \
    --width 1280 --height 544 --num-frames 89 \
    --guidance-scale 4.0 --num-inference-steps 20 \
    --skip-stage-2

Generating a Canny reference

import cv2

cap = cv2.VideoCapture("source.mp4")
fps = cap.get(cv2.CAP_PROP_FPS)
w, h = int(cap.get(3)), int(cap.get(4))
out = cv2.VideoWriter("canny_ref.mp4", cv2.VideoWriter_fourcc(*"mp4v"), fps, (w, h))

while True:
    ret, frame = cap.read()
    if not ret:
        break
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    gray = cv2.GaussianBlur(gray, (3, 3), 1.0)
    edges = cv2.Canny(gray, 100, 200)
    out.write(cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR))

cap.release()
out.release()

Recommended strengths

When to use this vs the other adapters

Stack with the character LoRA for identity + composition. Validated stack: char 0.9 + canny 1.0.

What this LoRA does well

• —Faithful composition transfer — generated scene preserves the layout of subjects + props from the canny reference.
• —Strong on architectural references (palace pillars, courtyard structures) where edges are dense and informative.
• —Works well as a storyboard-to-video bridge: hand-draw rough line art, get a Han dynasty scene that matches.

What it does NOT do

• —No identity — canny IC-LoRA controls composition, not who the subject is. Stack with the character LoRA for that.
• —Edge density matters — very sparse canny references (mostly black) leave the model under-constrained and outputs drift away from the reference.
• —Canny reference resolution must match target output resolution.

Related models

• —`SyFeee/ltx2.3-chinese-drama-charlora` — character / style LoRA.
• —`SyFeee/ltx2.3-chinese-drama-iclora-pose` — pose-controlled IC-LoRA.
• —`SyFeee/ltx2.3-chinese-drama-iclora-depth` — depth-controlled IC-LoRA.

License

Apache 2.0. See LICENSE for terms.

Attribution: SyFe.