ThunderHero LoRA for Wan 2.2 Image-to-Video: Lightning Effects on Any Image
A guide to using the ThunderHero LoRA with Wan 2.2 I2V — trigger words, ComfyUI setup for dual low/high-noise files, and how the cinematic lightning effect w…
On this page
ThunderHero LoRA for Wan 2.2 Image-to-Video
TL;DR:
- Drop-in LoRA for Wan 2.2 I2V A14B that adds cinematic lightning, electrical arcs, and superhero power effects to any input image.
- Trigger word is
Thunder Hero— add it to your prompt; the effect does the rest. - Two safetensors files (low-noise + high-noise) must both be loaded in ComfyUI — strength_model 0.8–1.0 is the sweet spot.
What ThunderHero Does
ThunderHero is a LoRA adapter, not a standalone model. Feed it a static image, write "Thunder Hero" in the prompt, and the Wan 2.2 pipeline turns the scene into a video where lightning, electrical bursts, and heroic power radiate from the subject. Portraits crackle with energy, cityscapes get struck by bolts, objects appear to charge with superpower.
The result is a short, stylized clip — not photorealistic action footage, but a dramatic, effects-heavy sequence that reads clearly as "superhero moment." The effect is consistent across subjects, which makes it useful for batch-applying a transformation look to a batch of images.
How It Was Built
ThunderHero was trained as a LoRA on top of Wan-AI's official Wan2.2-I2V-A14B model. It ships as two files because Wan 2.2 splits denoising into high-noise and low-noise stages — the LoRA has a variant for each stage. Both must be loaded for the effect to work. Training was GPU-sponsored by Redmond Ai.
License is CC0 1.0 — public domain, no strings attached. You can use it commercially, retrain on top of it, or republish without permission.
How to Use It in ComfyUI
- Install base model. Make sure your ComfyUI already runs Wan 2.2 I2V A14B. You'll need a workflow that starts with a Wan I2V checkpoint loader.
- Download the LoRA files. Get both:
[WAN2.2]ThunderHero_Redmond_low_noise.safetensors[WAN2.2]ThunderHero_Redmond_high_noise.safetensors
Place them in ComfyUI/models/loras/.
- Stack the LoRAs. Add a Load LoRA node after your base model loader. Route the model loader's MODEL output into the Load LoRA node's model input. Select both files.
- Route to KSampler. Connect the Load LoRA node's MODEL output to your KSampler.
- Set strength_model. Start at
0.8–1.0. Lower values (0.6–0.7) give a subtler effect; above 1.0 tends to oversaturate the lightning.
- Prompt with the trigger. The instance prompt is
Thunder Hero. Example prompts from the model card use just the trigger phrase — the visual effect is encoded in the LoRA, so you don't need elaborate descriptions.
Practical Notes
- VRAM. Wan 2.2 I2V 14B is a large base model. Expect similar VRAM requirements as stock Wan 2.2 — the LoRA adds negligible overhead (~146 MB for both files).
- Clip length. Wan I2V outputs short clips. ThunderHero doesn't extend duration; it adds visual effects to whatever length the base model produces.
- Subjects. Because the trigger is simple, test on a few images before batch-processing. Human subjects with visible hands/feet tend to show the most dramatic effect since lightning radiates across the body.
Common Errors and Fixes
Error | Cause | Fix
|---|---|---
No lightning effect visible | Only one LoRA file loaded (low or high) | Load both safetensors files in the Load LoRA node Model won't load | Wrong base model — not Wan 2.2 I2V A14B | Switch base model to Wan2.2-I2V-A14B Distorted output | strength_model too high | Reduce to 0.8–0.9 Trigger ignored | Prompt missing "Thunder Hero" exactly | Add Thunder Hero verbatim to prompt
FAQ
Can I use this with Automatic1111 or Forge? The model card only documents ComfyUI. Wan 2.2 I2V A14B is primarily a ComfyUI workflow; A1111 support for this specific base model is limited. ComfyUI is the reliable path.
Does it work with Wan 1.3 or other Wan versions? No. It's trained specifically for Wan 2.2 I2V A14B. Using it with other base versions will either fail or produce garbage.
Can I stack it with other LoRAs? You can, but lightning effects may conflict with style LoRAs. Test with the ThunderHero LoRA at full strength first, then layer additional adapters at reduced strength.
Is there a non-superhero variant? This is a single-purpose LoRA — cinematic lightning and electrical effects. For more general motion effects, pair the base Wan 2.2 I2V model with a different LoRA.
Sources
- ThunderHero on HuggingFace — model card, safetensors, CC0 license.
- ThunderHero on Civitai — trained words, stats, variant downloads.
- Wan2.2-I2V-A14B base model — required base checkpoint.