Text-to-Video
MiniMax H3
lora
character
consistency
mila
ltx-2.3
z-image-turbo
krea-2-turbo
text-to-image
image-to-video
realistic
cinematic
fashion
lifestyle
Instructions to use SOLRICKS/Mila with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Inference
- Notebooks
- Google Colab
- Kaggle
Update README.md
Browse files
README.md
CHANGED
|
@@ -2,6 +2,7 @@
|
|
| 2 |
license: other
|
| 3 |
base_model:
|
| 4 |
- Lightricks/LTX-2.3
|
|
|
|
| 5 |
- Tongyi-MAI/Z-Image-Turbo
|
| 6 |
- krea/Krea-2-Turbo
|
| 7 |
tags:
|
|
@@ -11,6 +12,7 @@ tags:
|
|
| 11 |
- consistency
|
| 12 |
- mila
|
| 13 |
- ltx-2.3
|
|
|
|
| 14 |
- z-image-turbo
|
| 15 |
- krea-2-turbo
|
| 16 |
- text-to-image
|
|
@@ -48,7 +50,7 @@ pipeline_tag: text-to-video
|
|
| 48 |
|
| 49 |
This repository contains dedicated Mila LoRA weights for:
|
| 50 |
|
| 51 |
-
**LTX-2.3 · Z-Image Turbo · Krea 2 Turbo**
|
| 52 |
|
| 53 |
Each LoRA is trained specifically for its corresponding base model and should only be used with that architecture.
|
| 54 |
|
|
@@ -56,109 +58,14 @@ Each LoRA is trained specifically for its corresponding base model and should on
|
|
| 56 |
|
| 57 |
---
|
| 58 |
|
| 59 |
-
## Model Overview
|
| 60 |
-
|
| 61 |
-
| Model | Base Model | Trigger | LoRA Weight |
|
| 62 |
-
|---|---|---|---|
|
| 63 |
-
| **LTX-2.3** | `Lightricks/LTX-2.3` | `mila` | `models/ltx-2.3/mila_ltx23_lora.safetensors` |
|
| 64 |
-
| **Z-Image Turbo** | `Tongyi-MAI/Z-Image-Turbo` | `mila_v1` | `models/z-image-turbo/mila_zimage_turbo_lora.safetensors` |
|
| 65 |
-
| **Krea 2 Turbo** | `krea/Krea-2-Turbo` | `mila` | `models/krea-2-turbo/mila_krea2_lora.safetensors` |
|
| 66 |
-
|
| 67 |
-
---
|
| 68 |
-
|
| 69 |
-
# LTX-2.3
|
| 70 |
-
|
| 71 |
-
### Character Video LoRA
|
| 72 |
-
|
| 73 |
-
**Base Model:** `Lightricks/LTX-2.3`
|
| 74 |
-
**Trigger:** `mila`
|
| 75 |
-
**Recommended LoRA Strength:** `0.8`
|
| 76 |
-
**Training Steps:** `2000`
|
| 77 |
-
|
| 78 |
-
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.
|
| 79 |
-
|
| 80 |
-
It works particularly well for:
|
| 81 |
-
|
| 82 |
-
- Lifestyle videos
|
| 83 |
-
- Fashion content
|
| 84 |
-
- Travel scenes
|
| 85 |
-
- Social media
|
| 86 |
-
- Reels-style shots
|
| 87 |
-
- Fitness scenes
|
| 88 |
-
- Luxury environments
|
| 89 |
-
- Natural facial expressions
|
| 90 |
-
- Cinematic character shots
|
| 91 |
-
|
| 92 |
-
### Example Prompt
|
| 93 |
-
|
| 94 |
-
```text
|
| 95 |
-
mila, consistent facial identity, blonde ponytail,
|
| 96 |
-
soft facial features, expressive eyes, natural skin texture,
|
| 97 |
-
cream outfit, luxury coastal background, warm sunlight,
|
| 98 |
-
realistic cinematic lifestyle video, natural movement
|
| 99 |
-
```
|
| 100 |
-
---
|
| 101 |
-
|
| 102 |
-
# Z-Image Turbo
|
| 103 |
-
|
| 104 |
-
### Character Image LoRA
|
| 105 |
-
|
| 106 |
-
**Base Model:** `Tongyi-MAI/Z-Image-Turbo`
|
| 107 |
-
**Trigger:** `mila_v1`
|
| 108 |
-
**Training Steps:** `2000`
|
| 109 |
-
|
| 110 |
-
The Z-Image Turbo version focuses on high-quality realistic still images while maintaining Mila's recognizable facial identity.
|
| 111 |
-
|
| 112 |
-
Recommended use cases include:
|
| 113 |
-
|
| 114 |
-
- Portrait photography
|
| 115 |
-
- Lifestyle imagery
|
| 116 |
-
- Fashion
|
| 117 |
-
- Luxury environments
|
| 118 |
-
- Travel photography
|
| 119 |
-
- Social media content
|
| 120 |
-
- Full-body character shots
|
| 121 |
-
- Cinematic portraits
|
| 122 |
-
|
| 123 |
-
### Example Prompt
|
| 124 |
-
|
| 125 |
-
```text
|
| 126 |
-
mila_v1, blonde ponytail, soft facial features,
|
| 127 |
-
expressive eyes, natural skin texture,
|
| 128 |
-
premium lifestyle photography,
|
| 129 |
-
warm natural sunlight, realistic skin detail,
|
| 130 |
-
cinematic photography
|
| 131 |
-
```
|
| 132 |
-
|
| 133 |
-
---
|
| 134 |
-
|
| 135 |
-
# Krea 2 Turbo
|
| 136 |
-
|
| 137 |
-
### Character Image LoRA
|
| 138 |
-
|
| 139 |
-
**Base Model:** `krea/Krea-2-Turbo`
|
| 140 |
-
**Trigger:** `mila`
|
| 141 |
-
|
| 142 |
-
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.
|
| 143 |
-
|
| 144 |
-
Recommended use cases include:
|
| 145 |
-
|
| 146 |
-
- Photorealistic portraits
|
| 147 |
-
- Fashion photography
|
| 148 |
-
- Editorial imagery
|
| 149 |
-
- Lifestyle scenes
|
| 150 |
-
- Travel content
|
| 151 |
-
- Luxury environments
|
| 152 |
-
- Full-body compositions
|
| 153 |
-
|
| 154 |
### Example Prompt
|
| 155 |
|
| 156 |
```text
|
| 157 |
-
|
| 158 |
-
|
| 159 |
-
|
| 160 |
-
warm
|
| 161 |
-
|
| 162 |
```
|
| 163 |
|
| 164 |
---
|
|
@@ -176,7 +83,7 @@ The project explores consistent character identity across independent generative
|
|
| 176 |
- Environment
|
| 177 |
- Lighting
|
| 178 |
- Camera framing
|
| 179 |
-
-
|
| 180 |
|
| 181 |
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.
|
| 182 |
|
|
@@ -191,6 +98,7 @@ Use and redistribution of each LoRA must comply with the license and usage terms
|
|
| 191 |
- `Lightricks/LTX-2.3`
|
| 192 |
- `Tongyi-MAI/Z-Image-Turbo`
|
| 193 |
- `krea/Krea-2-Turbo`
|
|
|
|
| 194 |
|
| 195 |
Please review the original base-model licenses before redistribution, deployment or commercial use.
|
| 196 |
|
|
|
|
| 2 |
license: other
|
| 3 |
base_model:
|
| 4 |
- Lightricks/LTX-2.3
|
| 5 |
+
- MiniMaxAI/MiniMax-H3
|
| 6 |
- Tongyi-MAI/Z-Image-Turbo
|
| 7 |
- krea/Krea-2-Turbo
|
| 8 |
tags:
|
|
|
|
| 12 |
- consistency
|
| 13 |
- mila
|
| 14 |
- ltx-2.3
|
| 15 |
+
- minimax-h3
|
| 16 |
- z-image-turbo
|
| 17 |
- krea-2-turbo
|
| 18 |
- text-to-image
|
|
|
|
| 50 |
|
| 51 |
This repository contains dedicated Mila LoRA weights for:
|
| 52 |
|
| 53 |
+
**MiniMax H3 · LTX-2.3 · Z-Image Turbo · Krea 2 Turbo**
|
| 54 |
|
| 55 |
Each LoRA is trained specifically for its corresponding base model and should only be used with that architecture.
|
| 56 |
|
|
|
|
| 58 |
|
| 59 |
---
|
| 60 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 61 |
### Example Prompt
|
| 62 |
|
| 63 |
```text
|
| 64 |
+
mila_v1, a blonde adult woman with softly defined facial features, natural skin texture and loose strands framing her face.
|
| 65 |
+
Seated at an outdoor cafe table, wearing a fitted black scoop-neck top, a small pendant and a wristwatch, with her blonde hair tied back.
|
| 66 |
+
She glances downward, lifts her face, smiles broadly and raises a hand beside her hair.
|
| 67 |
+
A white cup rests near the table. Plants and warm restaurant lights form a softly blurred background.
|
| 68 |
+
sound: silent video; no audible speech, music or ambient sound.
|
| 69 |
```
|
| 70 |
|
| 71 |
---
|
|
|
|
| 83 |
- Environment
|
| 84 |
- Lighting
|
| 85 |
- Camera framing
|
| 86 |
+
- T2V and I2V generation
|
| 87 |
|
| 88 |
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.
|
| 89 |
|
|
|
|
| 98 |
- `Lightricks/LTX-2.3`
|
| 99 |
- `Tongyi-MAI/Z-Image-Turbo`
|
| 100 |
- `krea/Krea-2-Turbo`
|
| 101 |
+
- `MiniMaxAI/MiniMax-H3`
|
| 102 |
|
| 103 |
Please review the original base-model licenses before redistribution, deployment or commercial use.
|
| 104 |
|