water_morping_wan

Ashmotv/water_morping_wan

water_morping_wan is a machine learning model from Ashmotv. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.

Downloads · 30 days

Access

Public

Updated Dec 11, 2025

Repo size

614 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors614 MB · 100%

At a glance

License: apache-2.0
Access: Public
Created: Dec 11, 2025
Updated: Dec 11, 2025
SHA: c86000f5

Model card

From the Hugging Face model README

wan2.2

Wat3r B3nding (Awan Effect) LoRA for WAN 2.2

This LoRA is fine-tuned for the WAN 2.2 video model to generate a specific, dynamic liquid transformation effect, transitioning a subject from one state or appearance to an entirely new one via a massive surge of hyper-realistic water.

🌊 Model Summary

Goal: To implement a stylized, fluid transition (the "Awan Effect") between two different subjects or scene compositions.
Base Model: Designed to be used with the WAN 2.2 video models (e.g., WAN2.2-T2V-A14B or I2V variants).
Unique Token (Trigger Word): wat3r b3nding

🚀 Usage

Recommended Prompt Structure

The LoRA works best with highly descriptive captions that specify both the starting subject/scene and the ending subject/scene, framing the transition with the trigger word.

Component	Example Detail
Starting Scene	A subject with long red curly hair and visible arm and chest tattoos, seated and smiling in a retro diner booth.
Transition Token	`wat3r b3nding`
Ending Scene	rising into a fluid human silhouette of a different subject wearing a high-collared, form-fitting blue sequined qipao-style dress with cutouts, surrounded by a swirling flock of black and white magpies.
Style/Descriptors	Dynamic, flowing, and elemental.

Example Prompt

A subject with long red curly hair and visible arm and chest tattoos, seated and smiling in a retro diner booth, Wat3r B3nding, then rising into a fluid human silhouette of a different subject wearing a high-collared, form-fitting blue sequined qipao-style dress with cutouts, surrounded by a swirling flock of black and white magpies. Dynamic, flowing, and elemental.

🎬 Video Examples

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/YeecI87TmKdbNnaZ-39Fz.mp4" controls width="320"></video> Example 1

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/uNcM3H09F2SW5HST1CVfu.mp4" controls width="320"></video> Example 2

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/lqoaaswl9P1sik1SbTRTz.mp4" controls width="320"></video> Example 3

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/us9IMq_v7jsJFcP4W49dg.mp4" controls width="320"></video> Example 4

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/1sYUHj7ryRblex5Y5OLBP.mp4" controls width="320"></video> Example 5

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/Q1_p1l_ZcQZI4hGSaXRJg.mp4" controls width="320"></video> Example 6

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/oLJFCz7TXV_G4U6yJyliN.mp4" controls width="320"></video> Example 7

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/vzraHbQ4MFoODSDLOtdzH.mp4" controls width="320"></video> Example 8

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/e97y20l0GEZuf_VeS42GW.mp4" controls width="320"></video> Example 9

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/hxdlopW4y-WtaWsr_mfUT.mp4" controls width="320"></video> Example 10

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/Z6s0bx3TN6-lKkkh3SDwY.mp4" controls width="320"></video> Example 11

<video src="https://cdn-uploads.huggingface.co/production/uploads/64d536b4bb9b5ddddc4e6d24/Q81jSnheugiivV90AHPV8.mp4" controls width="320"></video> Example 12

🛠️ Training Details

These settings were optimized for a small training dataset (3 videos/subjects) to prevent overfitting to the subjects and focus solely on the water-bending effect.

Setting	Value/Strategy	Notes
Trigger Word	`wat3r b3nding`	Use a unique token to minimize interference from the base model's knowledge.
Dataset Size	3 Video Clips
Target Steps	1100 Total Steps
Network Rank (Dimension)	32
Alpha	32
Learning Rate (LR)	$1e-4$	A very low LR is crucial for small datasets to ensure stable training.
Optimizer	AdamW	AdamW is the standard, reliable choice.
Precision	BF16	Required for memory efficiency on larger models.

License: apache-2.0
Created: Dec 11, 2025
Updated: Dec 11, 2025