README.md

---
license: other
base_model:
- Lightricks/LTX-2.3
- Tongyi-MAI/Z-Image-Turbo
- krea/Krea-2-Turbo
tags:
- lora
- character
- character
- consistency
- mila
- ltx-2.3
- z-image-turbo
- krea-2-turbo
- text-to-image
- text-to-video
- image-to-video
- realistic
- cinematic
- fashion
- lifestyle
widget:
- text: Mila — realistic cafe portrait
  output:
    url: models/z-image-turbo/mila-cafe-portrait.png
- text: Mila — portrait on the Amalfi Coast
  output:
    url: models/z-image-turbo/mila-amalfi-coast.png
- text: Mila — Positano balcony video
  output:
    url: models/ltx-2.3/mila-positano-balcony.mp4
- text: Mila — cafe lifestyle video
  output:
    url: models/ltx-2.3/mila-cafe-lifestyle.mp4
- text: Mila — grocery shopping video
  output:
    url: models/ltx-2.3/mila-grocery-shopping.mp4
- text: Mila — luxury car video
  output:
    url: models/ltx-2.3/mila-luxury-car.mp4
pipeline_tag: text-to-video
---

# Mila — Multi-Model Character LoRAs

**Mila** is a fictional AI character developed for consistent identity generation across multiple image and video model architectures.

This repository contains dedicated Mila LoRA weights for:

**LTX-2.3 · Z-Image Turbo · Krea 2 Turbo**

Each LoRA is trained specifically for its corresponding base model and should only be used with that architecture.

<Gallery />

---

## Model Overview

| Model | Base Model | Trigger | LoRA Weight |
|---|---|---|---|
| **LTX-2.3** | `Lightricks/LTX-2.3` | `mila` | `models/ltx-2.3/mila_ltx23_lora.safetensors` |
| **Z-Image Turbo** | `Tongyi-MAI/Z-Image-Turbo` | `mila_v1` | `models/z-image-turbo/mila_zimage_turbo_lora.safetensors` |
| **Krea 2 Turbo** | `krea/Krea-2-Turbo` | `mila` | `models/krea-2-turbo/mila_krea2_lora.safetensors` |

---

# LTX-2.3

### Character Video LoRA

**Base Model:** `Lightricks/LTX-2.3`  
**Trigger:** `mila`  
**Recommended LoRA Strength:** `0.8`  
**Training Steps:** `2000`

The LTX-2.3 version is designed for realistic character-focused video generation while maintaining Mila's facial identity across motion, expressions, environments and camera compositions.

It works particularly well for:

- Lifestyle videos
- Fashion content
- Travel scenes
- Social media
- Reels-style shots
- Fitness scenes
- Luxury environments
- Natural facial expressions
- Cinematic character shots

### Example Prompt

```text
mila, consistent facial identity, blonde ponytail,
soft facial features, expressive eyes, natural skin texture,
cream outfit, luxury coastal background, warm sunlight,
realistic cinematic lifestyle video, natural movement
```
---

# Z-Image Turbo

### Character Image LoRA

**Base Model:** `Tongyi-MAI/Z-Image-Turbo`  
**Trigger:** `mila_v1`  
**Training Steps:** `2000`

The Z-Image Turbo version focuses on high-quality realistic still images while maintaining Mila's recognizable facial identity.

Recommended use cases include:

- Portrait photography
- Lifestyle imagery
- Fashion
- Luxury environments
- Travel photography
- Social media content
- Full-body character shots
- Cinematic portraits

### Example Prompt

```text
mila_v1, blonde ponytail, soft facial features,
expressive eyes, natural skin texture,
premium lifestyle photography,
warm natural sunlight, realistic skin detail,
cinematic photography
```

---

# Krea 2 Turbo

### Character Image LoRA

**Base Model:** `krea/Krea-2-Turbo`  
**Trigger:** `mila`

The Krea 2 Turbo version is designed for high-quality character rendering with emphasis on natural photography, skin detail, fashion imagery and premium cinematic aesthetics.

Recommended use cases include:

- Photorealistic portraits
- Fashion photography
- Editorial imagery
- Lifestyle scenes
- Travel content
- Luxury environments
- Full-body compositions

### Example Prompt

```text
mila, blonde woman, soft facial features,
expressive eyes, natural skin texture,
premium fashion photography,
warm cinematic daylight,
realistic lifestyle aesthetic
```

---

# About Mila

**Mila is a fictional AI-generated character.**

The project explores consistent character identity across independent generative model architectures while preserving flexibility in:

- Clothing
- Hairstyle
- Pose
- Expression
- Environment
- Lighting
- Camera framing
- Image and video generation

Rather than locking the character to a single scene or aesthetic, the LoRAs are intended to preserve Mila's recognizable identity across a wide range of realistic scenarios.

---

# Licensing

This repository contains LoRA adapters trained for multiple independent base-model families.

Use and redistribution of each LoRA must comply with the license and usage terms of its corresponding base model:

- `Lightricks/LTX-2.3`
- `Tongyi-MAI/Z-Image-Turbo`
- `krea/Krea-2-Turbo`

Please review the original base-model licenses before redistribution, deployment or commercial use.

---

## SOLRICKS

**Generative AI · Models · Workflows · Tools**

Built for experimentation with high-quality image and video generation.