hdtrnk/VACEPhantomlightx2v

📖 Wan2.1 VACE + Phantom Lightx2V (Finetune)

Author / Creator: Ciro_Negrogni
Based on merge by: Inner_Reflections_AI
Original Guide: Wan VACE + Phantom Merge – Inner Reflections

---

🔹 About This Finetune

This is the Lightx2V variant of the VACE + Phantom merge, created by Ciro_Negrogni and derived from Inner_Reflections_AI’s original merge.
Purified to FP16 for WanGP compatibility and supplied with optional INT8 quantization.

• Architecture: vace_14B
• Mode: Image/Video conditioning with multi‑image reference support (2–4 refs in WanGP UI)
• Variants: FP16 (pure) and quanto INT8

---

🔹 Files

• Wan2.1_VACE_Phantom_Lightx2V_fp16_pure.safetensors
• Wan2.1_VACE_Phantom_Lightx2V_quanto_fp16_int8.safetensors (or _quanto_bf16_int8 depending on dtype)

---

🔹 Usage in WanGP

Place the JSON definition in:

app/finetunes/vace_phantom_lightx2v.json

Example JSON snippet:

{
  "model": {
    "name": "VACE Phantom Lightx2V 14B",
    "architecture": "vace_14B",
    "description": "VACE + Phantom Lightx2V merge by Ciro_Negrogni, based on Inner_Reflections_AI. Multi-image support enabled. FP16 + INT8 versions.",
    "URLs": [
      "ckpts/Wan2.1_VACE_Phantom_Lightx2V_fp16_pure.safetensors",
      "ckpts/Wan2.1_VACE_Phantom_Lightx2V_quanto_fp16_int8.safetensors"
    ],
    "modules": [],
    "auto_quantize": false
  }
}

---

🔹 Notes

• Experimental merge — results may vary. Adjust steps, guidance, and refs for best quality.
• Optimized for multi-image setups with clear subject separation.
• Refresh the UI if you see dropdown errors (Value: on is not in the list...).

---

🔹 Credits

• Original Merge & Guide: Inner_Reflections_AI
• Lightx2V Variant: Ciro_Negrogni
• WanGP Packaging: Conversion to FP16/INT8 and JSON prep for WanGP finetune system.