|
|
--- |
|
|
license: openrail++ |
|
|
tags: |
|
|
- image-to-video |
|
|
- text-to-video |
|
|
- wan2.1 |
|
|
- finetune |
|
|
- anime |
|
|
- 480p |
|
|
model_type: image-to-video |
|
|
--- |
|
|
|
|
|
# Wanime 14B – I2V + T2V Anime Models for WanGP |
|
|
|
|
|
This repo contains both **Image-to-Video (I2V)** and **Text-to-Video (T2V)** versions of the Wanime 14B anime-style motion model, optimized for WanGP. These models are based on the original release by [Zazc](https://civitai.com/models/1626197?modelVersionId=1840561) and converted to `.safetensors` format for secure local use. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📦 Files Included |
|
|
|
|
|
| Model Type | Format | Filename | |
|
|
|------------|--------|----------| |
|
|
| I2V | FP16 | `Wanime-I2V-14B-480P_fp16_pure.safetensors` | |
|
|
| I2V | INT8 | `Wanime-I2V-14B-480P_quanto_fp16_int8.safetensors` | |
|
|
| T2V | FP16 | `Wanime-T2V-14B_fp16_pure.safetensors` | |
|
|
| T2V | INT8 | `Wanime-T2V-14B_quanto_fp16_int8.safetensors` | |
|
|
|
|
|
--- |
|
|
|
|
|
## 🛠 How to Use in WanGP |
|
|
|
|
|
> Place all `.safetensors` files in: |
|
|
``` |
|
|
D:/pinokio/api/wan.git/app/ckpts/ |
|
|
``` |
|
|
|
|
|
> Then use these JSON finetune definitions in: |
|
|
``` |
|
|
D:/pinokio/api/wan.git/app/finetunes/ |
|
|
``` |
|
|
|
|
|
### 🔹 I2V Finetune JSON (`wan_i2v_wanime14B_480p.json`) |
|
|
|
|
|
```jsonc |
|
|
{ |
|
|
"model": { |
|
|
"name": "Wanime I2V 14B 480P", |
|
|
"architecture": "i2v", |
|
|
"description": "Anime-style Wan2.1 14B image-to-video model (480p baseline).", |
|
|
"URLs": [ |
|
|
"Wanime-I2V-14B-480P_fp16_pure.safetensors", |
|
|
"Wanime-I2V-14B-480P_quanto_fp16_int8.safetensors" |
|
|
], |
|
|
"modules": [], |
|
|
"auto_quantize": false |
|
|
}, |
|
|
"prompt": "", |
|
|
"negative_prompt": "out of frame, cropped, error, low quality, watermark", |
|
|
"resolution": "832x480", |
|
|
"video_length": 81, |
|
|
"guidance_scale": 1.0, |
|
|
"num_inference_steps": 8 |
|
|
} |
|
|
``` |
|
|
|
|
|
### 🔹 T2V Finetune JSON (`wan_t2v_wanime14B.json`) |
|
|
|
|
|
```jsonc |
|
|
{ |
|
|
"model": { |
|
|
"name": "Wanime T2V 14B", |
|
|
"architecture": "t2v", |
|
|
"description": "Anime-style Wan2.1 14B text-to-video model.", |
|
|
"URLs": [ |
|
|
"Wanime-T2V-14B_fp16_pure.safetensors", |
|
|
"Wanime-T2V-14B_quanto_fp16_int8.safetensors" |
|
|
], |
|
|
"modules": [], |
|
|
"auto_quantize": false |
|
|
}, |
|
|
"prompt": "", |
|
|
"negative_prompt": "out of frame, cropped, error, low quality, watermark", |
|
|
"resolution": "832x480", |
|
|
"video_length": 81, |
|
|
"guidance_scale": 7.5, |
|
|
"num_inference_steps": 10 |
|
|
} |
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## 💡 Prompt Ideas |
|
|
|
|
|
- “A mysterious anime girl walking across a glowing bridge at night, dramatic camera pan” |
|
|
- “A side-scrolling mecha battle in a ruined city, 80s anime style” |
|
|
- “A child running through falling sakura petals, slow motion, cinematic” |
|
|
|
|
|
--- |
|
|
|
|
|
## ✨ Notes |
|
|
|
|
|
- Use INT8 models for faster performance and lower VRAM use |
|
|
- Compatible with WanGP 2.1+ local UI via `--multiple-images` or `--t2v` launch mode |
|
|
- FP16 models recommended for 3090/4090-class GPUs |
|
|
- **This is a newly prepped finetune for WanGP and may require experimentation with guidance scale, step count, and prompt format to achieve optimal results.** |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧩 Credits |
|
|
|
|
|
- Original model: [Zazc on Civitai](https://civitai.com/models/1626197) |
|
|
- Base backbone: **Wan 2.1 14B** |
|
|
- Conversion & formatting: **hdtrnk** |
|
|
|
