Upload 2787085/3140313/README.md with huggingface_hub
Browse files- 2787085/3140313/README.md +77 -0
2787085/3140313/README.md
ADDED
|
@@ -0,0 +1,77 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
tags:
|
| 4 |
+
- kiss
|
| 5 |
+
- kissing
|
| 6 |
+
- style
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# cxy_kiss - v0.1 - 3140313
|
| 10 |
+
|
| 11 |
+
**Model Type**: LORA
|
| 12 |
+
|
| 13 |
+
**Base Model**: Wan Video 2.2 I2V-A14B
|
| 14 |
+
|
| 15 |
+
**Trigger Words**: None
|
| 16 |
+
|
| 17 |
+
**Tags**: kiss, kissing, style
|
| 18 |
+
|
| 19 |
+
## Gallery
|
| 20 |
+
|
| 21 |
+
<table>
|
| 22 |
+
<tr>
|
| 23 |
+
<td><video src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/0132b799-2d9f-4715-b65b-cf75ab1914d9/original=true/137043344.mp4" width="200" controls muted autoplay loop></video></td>
|
| 24 |
+
</tr>
|
| 25 |
+
</table>
|
| 26 |
+
|
| 27 |
+
## Description
|
| 28 |
+
|
| 29 |
+
My First Wan 2.2 LoRA
|
| 30 |
+
---------------------
|
| 31 |
+
|
| 32 |
+
This is my first attempt at training a video LoRA.
|
| 33 |
+
|
| 34 |
+
It is not perfect, but it works—and for me, that already makes it a meaningful first step. The LoRA was trained on **Wan 2.2**, which I still find to be a reliable foundation for character-focused video generation.
|
| 35 |
+
|
| 36 |
+
I also experimented with **LTX 2.3**, but in my experience, it still struggles with character consistency and identity preservation. Facial features can become unstable or distorted, especially during larger movements or longer generations. Because of this, I decided to continue training with Wan 2.2 for now.
|
| 37 |
+
|
| 38 |
+
Hopefully, more powerful and consistent open video models will arrive in the future. I am looking forward to seeing how the ecosystem develops—and to training better versions as I gain more experience.
|
| 39 |
+
|
| 40 |
+
### Trigger Words
|
| 41 |
+
|
| 42 |
+
**No trigger word is required.**
|
| 43 |
+
|
| 44 |
+
### Prompting
|
| 45 |
+
|
| 46 |
+
The LoRA is designed around scenes featuring **two people**.
|
| 47 |
+
|
| 48 |
+
And the prompting just : kiss .
|
| 49 |
+
|
| 50 |
+
Example prompt:
|
| 51 |
+
|
| 52 |
+
`two people kiss.`
|
| 53 |
+
|
| 54 |
+
`They are kissing.`
|
| 55 |
+
|
| 56 |
+
For better results, try adding:
|
| 57 |
+
|
| 58 |
+
* Detailed descriptions of both characters
|
| 59 |
+
* Clothing, hairstyle, and facial features
|
| 60 |
+
* Camera movement and shot composition
|
| 61 |
+
* Lighting and environment
|
| 62 |
+
* The specific interaction or action
|
| 63 |
+
* Mood, pacing, and visual style
|
| 64 |
+
|
| 65 |
+
Since this is my first LoRA, results may vary depending on the base model, resolution, sampler, LoRA strength, and generation workflow.
|
| 66 |
+
|
| 67 |
+
Thank you for trying it. Feedback, example videos, and workflow suggestions are always welcome.
|
| 68 |
+
|
| 69 |
+
---
|
| 70 |
+
|
| 71 |
+
Author: [CuiGodKiss](https://civitai.com/user/CuiGodKiss)
|
| 72 |
+
|
| 73 |
+
Model: [CivitAI Model Page](https://civitai.com/models/2787085?modelVersionId=3140313)
|
| 74 |
+
|
| 75 |
+
Archive: [CivArchive Page](https://civarchive.com/models/2787085?modelVersionId=3140313)
|
| 76 |
+
|
| 77 |
+
<!-- Version: 20260718_upload -->
|