1. Workflow Overview

This workflow generates high-dynamic explosion effect videos using the "Wan2.1" video generation model, transforming static images (e.g., mecha characters) into dynamic clips with explosions, fire, and debris. Key features:

• Image-to-Video: Adds effects like explosions based on input images.

• Multimodal Control: Combines text prompts (e.g., "intense explosion + terrified expression") with image semantics.

• Professional Optimization: Supports tiled rendering and VRAM management for HD output.

Core Models:

• Wan2.1-I2V-14B: Main video model (14B params, 480P output).

• UMT5-XXL Text Encoder: Processes Chinese/English prompts.

• Explosion LoRA: WAN2.1 ZOEY Explosion I2V_Alpha, enhances physics details.

2. Key Components

Critical Nodes:

1. WanVideoModelLoader:

   • Loads Wan2.1 model (Wan2_1-I2V-14B-480P_fp8_e4m3fn.safetensors).

   • Supports bf16/fp8 precision and VRAM optimization.

2. WanVideoTextEncode:

   • Input bilingual prompts (e.g., "baozha, violent explosion behind the character...").

   • Uses UMT5-XXL for text embeddings.

3. WanVideoLoraSelect:

   • Loads explosion LoRA (default weight: 1.0).

4. WanVideoSampler:

   • Key sampler settings:

     • Steps: 20

     • Sampler: dpm++_sde (optimal for motion).

     • Seed: 1057359483639287 (fixed).

5. VHS_VideoCombine:

   • Final video synthesis (MP4/GIF, 16fps, CRF19).

Dependencies:

• Install ComfyUI-WanVideoWrapper and ComfyUI-VideoHelperSuite.

• Download models from LibLibAI to:

  • Main model: ComfyUI/models/wan_video

  • LoRA: ComfyUI/models/loras

3. Workflow Structure

1. Model Loading:

   • Load video model, text encoder, VAE, and LoRA.

2. Input Processing:

   • Resize input image (e.g., mecha_valkyrie.png to 832x832) → CLIP vision encode.

3. Video Generation:

   • Fuse text/image embeddings → Sampler → Latent frames.

4. Output:

   • Decode latent → Combine video → Save as MP4 (default: WanVideo2_1_T2V).

4. Input & Output

Input Parameters:

• Image: Recommended resolution ≥832x832 (e.g., mecha valkyrie image).

• Text Prompt: Must include explosion keywords (Chinese/English).

• Seed: Random or fixed (e.g., 1057359483639287).

Output:

• 480P MP4 video with explosion effects (saved to ComfyUI/output).

• Example output description:

  "Violent explosion behind the character, flying debris, terrified expression, vibrant colors."

5. Notes

• VRAM: ≥16GB required (24GB recommended for HD rendering).

• Compatibility:

  • Only works with Wan2.1 models (not compatible with Stable Diffusion).

  • LoRAs must match Wan2.1-I2V version.

• Troubleshooting:

  • Missing WanVideoWrapper causes node errors.

  • Adjust ImageResizeKJ if input isn't square.