README.md

metadata

license: apache-2.0
tags:
  - image-to-video
  - video-generation
  - quantized
  - wan-2.1
  - fusionix
library_name: pytorch
model_type: phantom_14B
pipeline_tag: image-to-video

🌀 Wan2.1 Phantom-FusioniX 14B (Quantized)

A high‑performance image-to-video model based on WAN 2.1 Phantom 14B, merged with the FusionX research LoRAs—and now fully quantized (FP16 + INT8) for faster iteration. This version is configured primarily for use with WanGP (the Gradio-based GUI).

---

📦 Repository Contents

.
├── ckpts/
│   ├── Wan14BI2VFusioniX_phantom_14B_fp16.safetensors
│   └── Wan14BI2VFusioniX_phantom_14B_quanto_bf16_int8.safetensors
└── README.md

Note: Other example configs, workflows, and prompts are not included in this repo.

---

🚀 Quickstart

1. Clone this repo

   git clone https://huggingface.co/hdtrnk/Wan2.1_Phantom_FusioniX
   cd Wan2.1_Phantom_FusioniX
   

2. Install dependencies

   pip install torch torchvision --index-url https://download.pytorch.org/whl/cu124
   pip install -r requirements.txt
   

3. Place the checkpoints under ckpts/

   • FP16 model: ckpts/Wan14BI2VFusioniX_phantom_14B_fp16.safetensors
   • Quantized model: ckpts/Wan14BI2VFusioniX_phantom_14B_quanto_bf16_int8.safetensors

4. Configure your WanGP preset
   Create or update a JSON preset (e.g. wan_phantom_14B_fusionix.json) with:

   {
     "model": {
       "name":         "Phantom-FusioniX 14B (Quantized)",
       "architecture": "phantom_14B",
       "signature":    "phantom_14B",
       "description":  "WAN2.1 Phantom 14B merged with FusionX, quantized fp16+int8",
       "URLs": [
         "https://huggingface.co/hdtrnk/Wan2.1_Phantom_FusioniX/resolve/main/Wan14BI2VFusioniX_phantom_14B_fp16.safetensors",
         "https://huggingface.co/hdtrnk/Wan2.1_Phantom_FusioniX/resolve/main/Wan14BI2VFusioniX_phantom_14B_quanto_bf16_int8.safetensors"
       ],
       "auto_quantize": true
     },
     "prompt":               "",
     "negative_prompt":      "",
     "resolution":           "832x480",
     "video_length":         81,
     "seed":                 -1,
     "num_inference_steps":  8,
     "guidance_scale":       1.0,
     "repeat_generation":    1,
     "multi_images_gen_type": 0,
     "prompt_enhancer":      "",
     "activated_loras":      []
   }
   

5. Run WanGP

   python wgp.py --preset wan_phantom_14B_fusionix.json
   

6. Refresh your browser after each restart to clear Gradio’s cache.

---

🙏 Acknowledgments

• WAN Team – baseline text & image-to-video code
• FusionX authors – CausVid, AccVideo, MoviiGen1.1, MPS
• vrgamedevgirl84 – providing the FusioniX safetensors
• Community – ComfyUI, HuggingFace Hub, Gradio