1. Workflow Overview

This architecture animation workflow is optimized for low-VRAM devices (6GB+), generating up to 60s videos. Key features:

• Input: Static building image (e.g., work-04.jpg) → Output: Camera motion animation (push-in, orbit, etc.)

• Uses tiled decoding and model quantization (FramePack BF16)

• Exports 30FPS MP4 (H.264, 512x512 default)

2. Core Models

3. Key Nodes

4. Workflow Groups

1. Model Loading Group

   • Nodes: DownloadAndLoadFramePackModel, DualCLIPLoader

   • Input: Model name (e.g., FramePackI2V_HY)

   • Output: Initialized models + dual-CLIP encoders

2. Image Preprocessing

   • Nodes: ImageResize+ → GetImageSizeAndCount

   • Input: Original image (608x640 → resized to 512x512)

   • Output: Normalized image + dimensions

3. Animation Generation

   • Node: FramePackSampler

   • Params:

     • Steps 30, CFG 10

     • Motion prompt: "Camera slowly pushes forward"

   • Output: Latent video sequence

4. Video Export

   • Nodes: VAEDecodeTiled → VHS_VideoCombine

   • Params: FPS 30, CRF 19

   • Output: FramePack_00001.mp4

5. Inputs & Outputs

• Inputs:

  • Image: 608x640 JPG/PNG (auto-resized to 512x512)

  • Motion prompt: Verb-centric phrases (e.g., "Drone flyover of buildings")

  • Seed: Fixed at 47 for reproducibility

• Output:

  • 512x512 MP4 (YUV420P with metadata)

6. Notes

1. VRAM Optimization:

   • Use tile_size=128 for 6GB VRAM (60s video max)

   • Do NOT modify temporal_overlap=8 to avoid crashes

2. Prompt Tips:

   • Use action-focused short phrases (e.g., "Pan across the skyline")

   • Negative prompts are pre-configured

3. Troubleshooting:

   • Model Download: Manually place FramePackI2V_HY in ComfyUI/models/framepack/

   • Choppy Video: Reduce frame_rate to 24 or shorten total_second_length