Electro Surge is a LoRA for turning any character into a lightning-wielding powerhouse. Inspired by icons like Raiden, Thor, and Storm, this LoRA lets you generate images or videos of characters channeling electricity, shooting lightning bolts, and radiating power. Works best with weights between 1.0 and 1.6.
Prompt Examples
female samurai clenching her fists emitting lightning bolts, powerful pose, strong, electricity coming out of her body, glowing eyes, strong wind blowing in her hair, shaky handheld cameralightning being emitted from character, dramatic sky, electricity arcing from hands, cinematic lighting, intense expressionwarrior with electricity crackling around body, blue lightning, dynamic action pose, glowing effects, stormy backgroundUsage Tips
Recommended weights: 1.0–1.6 for best results
Works best with both image-to-video
Add details like "glowing eyes", "electric aura", or "stormy background" for more dramatic effects
Try different characters, poses, and settings for variety
Description
FAQ
Comments (10)
What do you mean by Recommended weights?
Are you referring to "Load LoRA" node params "stregth_model" and "strength_clip"?
yeah, the strength of the model
do you take requests for LTXV 13B LoRAs?
Is it just me or the dataset preparation for training is very complicated?
The github page doesn't mention using existing caption/image pairs, but instead something about using a json file for the captions and then my eyes glassed over...
Look at the bottom of the git, there is a "simple run" or something like that command that basically just do all the process for you, so you don't need to manually caption or prep the data, you just need to put a folder with a name like electricity_scenes in the folder of the trainer, then run this simple command and it'll run all the steps in sequence.
@PongoFlongo I see, but it doesn't say anything on how to use datasets which are already captioned.
In my experience manual captioning is better, not only in being more accurate, it also gives more flexibility because LLMs usually create rigid prompt structures, which is the reason I think some models are too picky about how to prompt them.
good job! May I ask how many videos for lora training
around 20
The Comfyui config can you share with us to reproduce this effect best.
This is such a great LoRA