Bringing Plant Growth to Life: An In-Depth Look at the WAN2.1 I2V Workflow

Workflow Overview

This is a WAN2.1 model-based image-to-video workflow specifically designed to generate timelapse videos of plant growth processes. Starting from a single input image, the workflow uses AI models to create videos showing plants gradually growing and covering object surfaces.

Core Models

1. WAN2.1-I2V Model: Main model "Wan2_1-I2V-14B-480P_fp8_e4m3fn.safetensors" for image-to-video conversion

2. Plant Growth LoRA: "Wan2.1-植物生长-FDGrowth_v4" specialized for plant growth effects

3. T5 Text Encoder: "umt5-xxl-enc-bf16.safetensors" for text prompt processing

4. CLIP Vision Model: "open-clip-xlm-roberta-large-vit-huge-14_visual_fp32.safetensors" for image embeddings

Key Components

1. WanVideoModelLoader: Loads WAN video model

   • Install via ComfyUI-WanVideoWrapper

   • Requires model files in specified directory

2. WanVideoLoraSelect: Selects plant growth LoRA

   • Weight set to 0.9 for balanced effect

3. WanVideoTextEncode: Text encoding node

   • Uses specific trigger phrase starting with "fdgr0w"

   • Includes detailed negative prompt list

4. WanVideoSampler: Video sampler

   • Uses unipc sampling method

   • 30 sampling steps, CFG scale 6

5. VHS_VideoCombine: Video composition

   • Outputs MP4 format at 16fps

   • Uses yuv420p pixel format

Workflow Structure

1. Model Loading Group:

   • Loads WAN main model, VAE, T5 encoder and CLIP vision

   • Selects plant growth LoRA

2. Input Processing Group:

   • Loads input image (e.g. "A (13).jpeg")

   • Adjusts image resolution (recommended 480×832)

3. Prompt Group:

   • Processes plant growth description text

   • Example prompts include detailed growth process

4. Video Generation Group:

   • Controls frame count (recommended 33/41/49)

   • Sets sampling parameters

5. Output Group:

   • Decodes video frames

   • Composes final MP4 video

Inputs and Outputs

Input Parameters:

• Input image (recommended 480×832)

• Plant growth description prompt

• Video frame count (33/41/49)

• Random seed (593617810391093)

Output Results:

• MP4 format video (16fps)

• Filename prefix "GrowA"

Notes

1. High VRAM requirement (recommend ≥12GB GPU)

2. Adjust "blocks to swap" (default 5, can increase to 20-30) if VRAM insufficient

3. Must use specific trigger phrase starting with "fdgr0w"

4. Recommended resolutions: 512×512, 576×576, 480×832 or 832×480

5. Includes GPU cleaner node (easy cleanGpuUsed) to prevent VRAM leaks