1. Workflow Overview

This workflow transforms static images into cinematic videos with dynamic camera motions, featuring:

1. 3D Orbital Effects: Convert 2D images to 3D perspectives via LoRA.

2. Smart Camera Movement: Simulate professional shots (dolly/pan/rotate).

3. Multi-Style Support: Adapt to cyberpunk, ancient styles, etc.

4. HD Output: Generate 480P videos (16FPS) via Wan2.1 model.

Use Cases: Product demos, architectural visualization, game trailers, short videos.

2. Core Models

3. Key Nodes

4. Workflow Groups

1. Model Loading Group

   • Load Wan2.1 base model + Orbital LoRA (weight=1).

   • Initialize UMT5 text and CLIP vision encoders.

2. Input Processing Group

   • Image Input: Static scene (e.g., cyberpunk_city.jpeg).

   • Text Prompts:

     • Positive: "smooth camera rotation around"

     • Negative: Exclude static/low-quality terms.

3. Video Generation Group

   • Sampling:

     • Sampler: dpm++_sde (20 steps, CFG=5).

     • Fixed seed (1057359483639287) for reproducibility.

   • Resolution: 480x832 (adaptive to landscape/portrait).

4. Output Group

   • 16FPS MP4 video (CRF=19 for quality/size balance).

5. Inputs & Outputs

• Required Inputs:

  • Image: ≥1024x1024, clear subject (e.g., building/product).

  • Motion Prompt: English instructions (e.g., "pan left slowly").

• Output:

  • Video file (default output.mp4) with metadata.

6. Notes

1. Hardware:

   • ≥12GB VRAM (RTX 3090+), UMT5-XXL requires BF16 support.

2. LoRA Tuning:

   • Weight >1 enhances 3D effects but may reduce stability.

3. Troubleshooting:

   • Artifacts: Lower CFG (current=5) or reduce motion range.

   • Flickering: Enable teacache_args VRAM optimization.

4. Extensibility:

   • Swap LoRAs (e.g., 360_rotation/time_lapse) for different styles.