1. Workflow Overview

• Function: Generates animated videos that pack input objects (e.g., dolls) into gift boxes with LoRA effects.

• Goal: Combine WAN video models and LoRAs for creative video generation.

• Key Models:

  • WAN2.1-I2V-14B (Base video model, FP8 precision).

  • "Packed into Box" LoRA (Animation effect).

  • UMT5-XXL (Text encoder for prompts).

2. Node Breakdown

Critical Nodes:

1. WanVideoModelLoader: Loads WAN model (Wan2_1-I2V-14B-480P_fp8_e4m3fn.safetensors).

2. WanVideoLoraSelect: Applies effect LoRA (e.g., "Packed into Box").

3. LoadImage: Input static image (e.g., "POPMART-style doll").

4. WanVideoTextEncode: Encodes prompts (e.g., "Bounce into a gift box").

5. WanVideoSampler: Generates video frames (sampler: dpm++_sde, 20 steps).

6. VHS_VideoCombine: Exports MP4 (16fps, H.264).

Dependencies:

• WAN Models: Download Wan2_1-I2V-14B and Wan2_1_VAE to models/wan_video.

• LoRA: Place Packed into Box LoRA in models/loras.

3. Workflow Structure

Group Logic:

1. Model Loading:

   • Input: WAN model, LoRA, VAE.

   • Output: Video generation pipeline.

2. Input Processing:

   • Input: Image + text prompts.

   • Output: Image/text embeddings.

3. Video Generation:

   • Input: Embeddings, seed.

   • Output: Latent video frames.

4. Video Synthesis:

   • Input: Decoded frames.

   • Output: MP4 (480P by default).

4. Inputs & Outputs

Inputs:

• Required:

  • Static image (e.g., doll photo).

  • Positive prompt (e.g., "Bounce into a gift box").

• Optional: Seed (randomized), LoRA strength (default 1.0).

Outputs:

• MP4 video (e.g., WanVideo2_1_T2V.mp4) with gift box animation.

5. Notes

1. Hardware:

   • ≥16GB VRAM (FP8 requires high-end GPUs).

   • NVIDIA 30/40-series recommended.

2. Common Issues:

   • Missing models: Check models/wan_video.

   • Choppy video: Reduce FPS (e.g., 12) or resolution.

3. Optimization:

   • Enable fp8_e4m3fn if supported.

   • Use offload_device to save VRAM.