---
base_model:
- Lightricks/LTX-2.3
base_model_relation: adapter
license: other
license_name: ltx-2-community-license
license_link: https://github.com/Lightricks/LTX-2/blob/main/LICENSE
language:
- en
tags:
- ltx-2
- ltx-video
- ic-lora
- video-to-video
- lora
pipeline_tag: text-to-video
library_name: diffusers
extra_gated_description: >-
  By clicking "Agree and Access" you acknowledge the [Privacy
  Policy](https://static.lightricks.com/legal/Privacy%20Policy%20-%20LTX%20Platform.pdf) 
  and consent to receive offers and updates. You can unsubscribe at any time.
extra_gated_button_content: Agree and Access
---

# LTX-2.3 22B IC-LoRA CROOKED

IC-LoRA adapter for LTX-2 that transforms close-up portrait videos from **straight eyes** to **permanent convergent strabismus** (inward-turned / crossed eyes), while preserving facial expression, head movement, lighting, and camera framing.

## Model Details

- **Base model:** `ltx2.3-20b-20k-step-2362000-v4.safetensors` ([Lightricks LTX-2](https://huggingface.co/Lightricks/LTX-2))
- **Type:** IC-LoRA (video-to-video)
- **Training steps:** 3000
- **LoRA rank / alpha:** 32 / 32
- **Resolution buckets:** `960×544×49` (landscape), `544×960×49` (portrait)

## Requirements

This LoRA is an **adapter**, not a standalone model. You also need:

1. Base checkpoint: `ltx2.3-20b-20k-step-2362000-v4.safetensors`
2. Text encoder: `gemma-3-12b-it-qat-q4_0-unquantized`
3. A **reference video** with straight eyes

Frame count must satisfy `frames % 8 == 1` (e.g. 49, 145, 201). Use `544×960` for portrait clips.

## Usage

### LTX Trainer

```bash
cd LTX-2/packages/ltx-trainer

uv run python scripts/inference.py \
  --checkpoint path/to/ltx2.3-20b-20k-step-2362000-v4.safetensors \
  --text-encoder-path path/to/gemma-3-12b-it-qat-q4_0-unquantized \
  --lora-path path/to/lora_weights_step_03000.safetensors \
  --reference-video path/to/straight_eye_clip.mp4 \
  --prompt "A close-up portrait video of a man with permanent severe convergent strabismus. Both eyes are continuously turned inward toward the nose throughout the shot. The subject maintains natural facial expression, head movement, lighting, and camera framing." \
  --negative-prompt "normal eyes, straight eyes, corrected eyes, gaze correction, blurry, jittery" \
  --width 960 --height 544 \
  --num-frames 49 \
  --num-inference-steps 30 \
  --guidance-scale 4.0 \
  --stg-scale 1.0 --stg-blocks 29 --stg-mode stg_v \
  --skip-audio \
  --output crooked.mp4
```

### ComfyUI

1. Place `lora_weights_step_03000.safetensors` in `ComfyUI/models/loras/`.
2. Load the matching LTX 2.3 base checkpoint.
3. Use an [IC-LoRA (video-to-video) workflow](https://github.com/Lightricks/ComfyUI-LTXVideo/blob/master/example_workflows/2.3/LTX-2.3_V2V_ICLoRA_Single_Stage_Distilled.json) with a reference video input.

Reference workflows: [LTX-2 repository](https://github.com/Lightricks/LTX-2).

## Example Prompts

**Positive:**

> A close-up portrait video of a man with permanent severe convergent strabismus. Both eyes are continuously turned inward toward the nose throughout the shot. The subject maintains natural facial expression, head movement, lighting, and camera framing.

**Negative:**

> worst quality, inconsistent motion, blurry, jittery, distorted, normal eyes, straight eyes, corrected eyes, gaze correction

## License

See the **[LTX-2-community-license](https://github.com/Lightricks/LTX-2/blob/main/LICENSE)** for full terms.

## Acknowledgments

- Base model by **[Lightricks](https://ltx.io/)**
- Training infrastructure: **[LTX-2 Community Trainer](https://github.com/Lightricks/LTX-2/tree/main/packages/ltx-trainer)**