ArtificialGuyBR

Home / Blog / AI video

Atomic Explosion LoRA for WAN I2V 14B

LoRA for WAN I2V 14B generating explosion effects from static images, with low and high noise variants for ComfyUI.

2 sources cited AI video

On this page

  1. TL;DR
  2. What This Model Does
  3. Training and Base Model
  4. How to Use the Model
  5. License
  6. Common Errors and Fixes
  7. FAQ
  8. Sources
  9. Related Models

Atomic Explosion LoRA for WAN I2V 14B

Sample explosion generation from the LoRA

TL;DR

What This Model Does

The Atomic Explosion LoRA for WAN I2V 14B is a Low-Rank Adaptation (LoRA) model that specializes in generating cinematic explosion effects from static images. It works with the Wan-AI Wan2.2-I2V-A14B image-to-video model, transforming a single input frame into a dynamic video sequence showing a dramatic explosion.

The model is trained on two noise variants, providing flexibility for different generation scenarios. The low_noise variant is suited for cleaner, more controlled explosions, while the high_noise variant can produce more chaotic and intense effects. Both files should be installed and used depending on the desired outcome.

Training and Base Model

This LoRA is built on top of Wan-AI/Wan2.2-I2V-A14B, a 14 billion parameter image-to-video model. The base model handles the core video generation, while the LoRA adds the specific explosion effect style.

The project acknowledges special thanks to Redmond Ai for generous sponsorship and GPU time provided during training. This support enabled the creation of this free, open-source LoRA model.

How to Use the Model

Installation in ComfyUI

  1. Download both [WAN2.2-I2V]EXPLOSION_REDMOND_low_noise.safetensors and [WAN2.2-I2V]EXPLOSION_REDMOND_high_noise.safetensors from the releases
  2. Place both files in ComfyUI/models/loras/
  3. Ensure your ComfyUI environment is configured to run Wan I2V models

Workflow Setup

  1. Add a Load LoRA node after your base model loader in ComfyUI
  2. Load your base Wan I2V model
  3. Connect the output of the model loader to the model input of the Load LoRA node
  4. Select either the low_noise or high_noise variant in the Load LoRA node
  5. Connect the MODEL output of the LoRA node to your KSampler
  6. Adjust the strength_model parameter, starting at 0.8-1.0 for best results

Prompting

Use the trigger phrase "EXPLOSION" (from Hugging Face) or "ATOMIC EXPLOSION" (from Civitai) in your prompt. Examples include:

License

The model is released under the Creative Commons CC0 1.0 Universal license, allowing free use for both commercial and non-commercial projects. The Civitai listing also mentions Apache 2.0, though the Hugging Face card specifies CC0 1.0.

Second sample from the LoRA

Common Errors and Fixes

Error | Cause | Fix | Source

|---|---|---|---

Model not loading in ComfyUI | Files not placed in correct directory | Ensure both .safetensors files are in ComfyUI/models/loras/ | HuggingFace card Weak explosion effects | Low strength value | Set strength_model to at least 0.8, ideally 1.0 | Civitai workflow guide Inconsistent results | Wrong base model | Confirm using Wan2.2-I2V-A14B as the base model | HuggingFace card

FAQ

Q: Do I need both low_noise and high_noise files?

A: Both files are recommended for full flexibility. Install them both and choose the variant in your Load LoRA node depending on whether you want a cleaner or more chaotic explosion effect.

Q: Can I use this with other diffusion UIs?

A: The LoRA format is compatible with any tool that supports LoRA loading with Wan I2V models. The setup instructions provided are ComfyUI-specific, but the same principle of loading the LoRA and applying it to the base model applies to other environments like diffusers.

Q: What kind of trigger words should I use?

A: The HuggingFace card suggests "EXPLOSION" while the Civitai page lists "ATOMIC EXPLOSION" as the trained word. Either should work, but using the exact trained word may produce stronger results.

Sources