---
license: mit
library_name: diffusers
pipeline_tag: text-to-image
base_model:
- Lightricks/LTX-2.3
base_model_relation: adapter
language:
- en
tags:
- ltx-video
- ltx-2.3
- text-to-video
- video-to-video
- lora
- ic-lora
- template:diffusion-lora
instance_prompt: A person talking directly to the camera against a flat green chroma-key background.
pinned: true
widget:
- text: A person talking directly to the camera against a flat green chroma-key background. Natural lip sync, occasional blinks, subtle head movements. Soft natural lighting, photorealistic, sharp focus, stable framing.
  output:
    url: samples/sample_0.gif
- text: A person talking directly to the camera against a flat green chroma-key background. Natural lip sync, occasional blinks, subtle head movements. Soft natural lighting, photorealistic, sharp focus, stable framing.
  output:
    url: samples/sample_1.gif
- text: A person talking directly to the camera against a flat green chroma-key background. Natural lip sync, occasional blinks, subtle head movements. Soft natural lighting, photorealistic, sharp focus, stable framing.
  output:
    url: samples/sample_2.gif
---

<div align="center">

# LTX-2.3 Greenscreen Avatar — IC-LoRA

**Vertical talking-head avatars on a flat green chroma-key background — ready to composite.**

[![Base model](https://img.shields.io/badge/base-LTX--2.3-1f6feb?logo=huggingface&logoColor=white)](https://huggingface.co/Lightricks/LTX-2.3)
[![Adapter](https://img.shields.io/badge/adapter-IC--LoRA-8957e5)](#)
[![Task](https://img.shields.io/badge/task-video--to--video-fb8500)](#)
[![Orientation](https://img.shields.io/badge/orientation-vertical-2a9d8f)](#)
[![License](https://img.shields.io/badge/license-MIT-2ea043)](#license)

</div>

---

An **In-Context LoRA (IC-LoRA)** for [`Lightricks/LTX-2.3`](https://huggingface.co/Lightricks/LTX-2.3),
trained on top of the `ltx-2.3-22b-dev.safetensors` checkpoint. It specializes the base
model in generating **vertical talking-head avatars on a flat green chroma-key background**,
ready for compositing in downstream video pipelines.

| Property | Value |
|---|---|
| **Base model** | [`Lightricks/LTX-2.3`](https://huggingface.co/Lightricks/LTX-2.3) (`ltx-2.3-22b-dev.safetensors`) |
| **Adapter type** | IC-LoRA (in-context LoRA) |
| **Orientation** | Vertical (portrait) |
| **Task** | Video-to-video / text-to-video |
| **Weights** | `lora_weights_step_05000.safetensors` |

---

## Samples

<div align="center">

<table>
<tr>
<td><img src="./samples/sample_0.gif" width="240" alt="sample 0" /></td>
<td><img src="./samples/sample_1.gif" width="240" alt="sample 1" /></td>
<td><img src="./samples/sample_2.gif" width="240" alt="sample 2" /></td>
</tr>
</table>

<details>
<summary><b>Prompt used for all samples</b></summary>

<br>

*"A person talking directly to the camera against a flat green chroma-key background. Natural lip sync, occasional blinks, subtle head movements. Soft natural lighting, photorealistic, sharp focus, stable framing."*

</details>

<br>

**High-quality example** — input generation → green-screen output (side-by-side comparison):

<video src="https://huggingface.co/OmerHagage/ltx2-greenscreen-avatar-ic-lora-vertical-v1/resolve/main/samples/hq/sample.mp4" controls width="480" muted loop></video>

</div>

---

## Model Details

- **Developed by:** Lightricks (fine-tune)
- **Base model:** [`Lightricks/LTX-2.3`](https://huggingface.co/Lightricks/LTX-2.3) — `ltx-2.3-22b-dev.safetensors`
- **Adapter type:** In-Context LoRA (IC-LoRA)
- **Trainer:** [LTX-2](https://github.com/Lightricks/LTX-2)

### Training

| Hyperparameter | Value |
|---|---|
| Training type | IC-LoRA fine-tuning |
| Training steps | 5,000 |
| Learning rate | 2e-4 |
| Batch size | 1 |

---

## Intended Use

This IC-LoRA is built for producing **vertical talking-head avatars on a clean green
chroma-key background**, so the subject can be keyed out and composited onto any
background. Use cases include:

- **Avatar & presenter clips** — short, front-facing talking-head shots with natural
  lip sync, blinks, and subtle head movement.
- **Greenscreen compositing** — the flat chroma-key background makes keying and
  background replacement straightforward in any editor.
- **Vertical / portrait content** — sized for mobile-first formats such as Reels,
  Shorts, and TikTok.

---

## Usage

This adapter is designed for the LTX-2.3 (Lightricks audio-video) pipeline and can be used
with the standard LTX-2.3 tooling (ComfyUI or `diffusers`).

### 🔌 Using Trained LoRAs in ComfyUI

In order to use this LoRA in ComfyUI, follow these steps:

1. Copy `lora_weights_step_05000.safetensors` to the `models/loras` folder in your ComfyUI installation.
2. In your ComfyUI workflow:
    - Add the "Load LoRA" node and select `lora_weights_step_05000.safetensors`
    - Connect it to the "Load Checkpoint" node (`ltx-2.3-22b-dev.safetensors`) to apply the LoRA to the base model

You can find reference Text-to-Video (T2V) and Image-to-Video (I2V) workflows in the
official [LTX-2 repository](https://github.com/Lightricks/LTX-2).

---

## License

This adapter is released under the MIT license and is intended to be used together with
the base model, [`Lightricks/LTX-2.3`](https://huggingface.co/Lightricks/LTX-2.3). Please
review and comply with the base model's own license before use.

---

## Acknowledgments

- Base model: [Lightricks/LTX-2.3](https://huggingface.co/Lightricks/LTX-2.3)
- Trainer: [LTX-2](https://github.com/Lightricks/LTX-2)