kaiwen-wang's picture
Update model card and add converted checkpoint
22f29d9
|
Raw
History Blame Contribute Delete
1.26 kB
---
license: cc-by-nc-4.0
---
# Wavelet Phase Diffusion (WPD)
Model weights for *Wavelet Phase Diffusion for Structurally and Semantically Consistent Sim-to-Real Translation*.
[Project Page](https://kit-mrt.github.io/Wavelet-Phase-Diffusion/) · [Paper](https://arxiv.org/abs/2607.21628)
## Files
| File | Stage | Description |
|---|---|---|
| `flux.safetensors` | Image | FLUX WPD LoRA for image sim-to-real translation. |
| `flux_wo_lrf.safetensors` | Image | FLUX baseline trained without Low-Frequency Randomization (LFR). |
| `wan_high.safetensors` | Video | Wan2.2-14B WPD LoRA for the high-noise stage of video translation. |
| `wan_low.safetensors` | Video | Wan2.2-14B WPD LoRA for the low-noise stage of video translation. |
| `high_noise_model_converted.safetensors` | Video | 4-step high-noise checkpoint for Wan2.2-14B, converted from Wan2.2-Lightning. |
| `low_noise_model_converted.safetensors` | Video | 4-step low-noise checkpoint for Wan2.2-14B, converted from Wan2.2-Lightning. |
## Notes
- The image models are intended for FLUX-based translation workflows.
- The video models are split across high-noise and low-noise stages.
- The baseline `flux_wo_lrf.safetensors` is useful for ablation comparisons against the full WPD model.