---
base_model: Wan-AI/Wan2.2-T2V-A14B
pipeline_tag: image-to-video
license: apache-2.0
tags:
- wan
- wan2.2
- lora
- text-to-video
- image-to-video
- uncensored
- hyper-realistic
- open-source
- cyberpunk
- anime
- character
- tetsuo
widget:
- text: "tetsuo_character, cyberpunk woman walking through neon city, rain, volumetric lighting, cinematic"
---

# Tetsuo AI - WAN 2.2 Video Generation LoRA

> **Status: Awaiting training images** - See [issue #1](https://github.com/tetsuo-ai/tetsuo-model/issues/1)

An uncensored, hyper-realistic LoRA adapter for [WAN 2.2 T2V 14B](https://github.com/Wan-Video/Wan2.2). Built for high-definition video generation with no content restrictions, aligned with Tetsuo AI's open-source ethos.

## Uncensored and Open Source

This model is trained without safety filters or content restrictions. We believe generative AI should be open, transparent, and free from arbitrary censorship. Released under Apache 2.0 for personal and commercial use.

## Hyper-Realistic Output

Fine-tuned with a focus on high-definition, photorealistic video rendering. WAN 2.2 supports up to 720p/1080p resolution with strong temporal consistency and motion quality. The model produces cinematic video output with realistic lighting, textures, and camera motion.

## Character Training

The initial training run focuses on the Tetsuo AI character, a cyberpunk anime female rendered in two modes:

1. **High-contrast B&W manga** - Heavy ink work, aggressive linework, seinen manga aesthetic
2. **3D cyberpunk portraiture** - Volumetric lighting, neon blue highlights, rain, dark metallic surfaces

Trigger word: `tetsuo_character`

Trained using ComfyUI's built-in LoRA training nodes, the same unified pipeline as the Flux image model.

## Usage

### ComfyUI

1. Download the LoRA `.safetensors` file
2. Place it in `ComfyUI/models/loras/`
3. Load WAN 2.2 T2V model with the **LoRA Loader** node
4. Include `tetsuo_character` in your prompt

## Example Prompts

```
tetsuo_character, cyberpunk woman walking through neon city, rain, volumetric lighting, cinematic
```

```
tetsuo_character, sitting at holographic terminal, typing, dark room, blue screen glow, focused expression
```

```
tetsuo_character, performing on stage, electric guitar, neon-lit venue, detached intensity, camera slowly zooming in
```

## Training Details

| Parameter | Value |
|-----------|-------|
| Base Model | WAN 2.2 T2V 14B (Q8_0 GGUF) |
| Method | LoRA (Low-Rank Adaptation) |
| Rank | 32 |
| Steps | 1000 |
| Learning Rate | 1e-4 |
| Optimizer | AdamW |
| Trigger Word | `tetsuo_character` |
| Training Dtype | bfloat16 |
| Bypass Mode | Yes (quantized base) |
| Framework | ComfyUI built-in training |

## Links

- [GitHub - Training Pipeline](https://github.com/tetsuo-ai/tetsuo-model)
- [Flux Image LoRA](https://huggingface.co/piccassol/tetsuo-ai-flux-lora)
- [Tetsuo AI](https://www.tetsuocorp.com)
- [Twitter/X](https://x.com/tetsuoai)

## License

Apache 2.0