File size: 2,770 Bytes
beb31fe | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 | ---
license: apache-2.0
library_name: diffusers
pipeline_tag: image-to-video
tags:
- wan
- video-generation
- image-to-video
- diffusers
base_model: alibaba-pai/Wan2.1-Fun-V1.1-1.3B-InP
---
# Wan2.1-Fun-V1.1-1.3B-InP (Diffusers)
This is a diffusers-format conversion of [alibaba-pai/Wan2.1-Fun-V1.1-1.3B-InP](https://huggingface.co/alibaba-pai/Wan2.1-Fun-V1.1-1.3B-InP) (Wan-Fun Inpaint V1.1 1.3B) from VideoX-Fun format.
## Model Details
- **Architecture**: WanTransformer3DModel with `in_channels=36` (16 noise + 4 mask + 16 image latent)
- **Parameters**: 1.3B
- **Pipeline**: `WanImageToVideoPipeline` (standard diffusers, no patching required)
- **Resolution**: 480x832 (480p) recommended
- **Frames**: 49 frames at 16fps (~3 seconds)
This model has the same I2V architecture as the official [Wan2.1-I2V-14B-480P](https://huggingface.co/Wan-AI/Wan2.1-I2V-14B-480P-Diffusers) (`in_channels=36`), but at 1.3B scale.
## Usage
```python
import torch
from diffusers import WanImageToVideoPipeline
from PIL import Image
pipe = WanImageToVideoPipeline.from_pretrained(
"engineerA314/Wan2.1-Fun-V1.1-1.3B-InP-Diffusers",
torch_dtype=torch.bfloat16,
)
pipe.enable_sequential_cpu_offload()
image = Image.open("first_frame.png").convert("RGB")
output = pipe(
image=image,
prompt="A person is talking naturally",
negative_prompt="static, blurred, low quality",
height=480,
width=832,
num_frames=49,
num_inference_steps=50,
guidance_scale=5.0,
)
from diffusers.utils import export_to_video
export_to_video(output.frames[0], "output.mp4", fps=16)
```
## Conversion Details
Converted from VideoX-Fun format using 1:1 weight key mapping (983 keys). No architectural modifications were needed -- the standard `WanImageToVideoPipeline` handles `in_channels=36` natively.
### Components
| Component | Source |
|-----------|--------|
| Transformer | Converted from `alibaba-pai/Wan2.1-Fun-V1.1-1.3B-InP` |
| VAE | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers` |
| Text Encoder | `Wan-AI/Wan2.1-T2V-1.3B-Diffusers` (UMT5-XXL) |
| Image Encoder | `Wan-AI/Wan2.1-I2V-14B-480P-Diffusers` (CLIP ViT-H-14) |
| Scheduler | UniPCMultistepScheduler (`flow_shift=3.0`) |
### Comparison with TI2V variant
| | This model (InP) | [TI2V](https://huggingface.co/engineerA314/Wan2.1-Fun-V1.1-1.3B-TI2V-Diffusers) |
|---|---|---|
| `in_channels` | 36 (noise + mask + image) | 32 (noise + image) |
| Pipeline patches | None needed | `prepare_latents` override required |
| Origin | Wan-Fun Inpaint | Wan-Fun Camera Control (adapter removed) |
## Acknowledgements
- [Alibaba PAI / VideoX-Fun](https://github.com/alibaba-pai/VideoX-Fun) for the original Wan-Fun models
- [Wan-Video](https://github.com/Wan-Video/Wan2.1) for the Wan 2.1 architecture
|