A Yellow Lightning effect, such as from special effects, spells, magic attacks, etc. Trained on deliberate_v5 model, but others can work as well if you're willing to experiment.
The prompt should include at least "yellow lightning" somewhere in it.
There are many ways you can trigger its effects, some of the common ones include "yellow lightning striking down [...]", "[...] casting yellow lightning", "[...] conjuring yellow lightning" where [...] includes a subject of some sort. You can also get a spherical effect surrounding a character with "surrounded by yellow lightning sphere", or simply engulfed by it using "engulfed by yellow lightning". Unfortunately, it following your prompt is hit-or-miss.
Typical keywords: yellow lightning, yellow lightning striking down, engulfed by yellow lightning, casting yellow lightning, ball of yellow lightning, conjuring yellow lightning ball.
Best weights between 0.5 and 1.0, although it depends on your model and prompt. For example you might need to lower its weight if your character doesn't show up. Larger weights increase the lightning, lower it for more subtle effect.
Works good at 768x768 resolution directly (before hires fix).
Most samplers work well, but Restart tends to add a lot more subtle details to the lightning effect.
Potential issues:
It has difficulty following the prompt sometimes and will only vaguely resemble what you ask for; hit or miss here.
Tends to mangle faces for some reason. Use inpaint and edit it after (most example images used some inpaint/editing).
Sometimes your character or subject won't show up at all; in this case try to lower the weight or adjust the prompt.
v2 is finetuned on CPU with a huge batch size encompassing the entire dataset (236 images). Yes, it took forever. I had hoped it would adhere more to prompt and blend better, and maybe mangle faces less. While it does seem to be somewhat more faithful in some cases, in others it makes no difference or can be slightly worse. Your mileage may vary.
I'm trying to explore potential of hypernetworks.
If you want to support me: https://www.paypal.com/donate/?hosted_button_id=VFU5T4P2ZMJ5W
v2 is finetuned on CPU with a huge batch size encompassing the entire dataset (236 images). Yes, it took forever. I had hoped it would adhere more to prompt and blend better, and maybe mangle faces less. While it does seem to be somewhat more faithful in some cases, in others it makes no difference or can be slightly worse. Your mileage may vary.
1. The rights to reposted models belong to original creators.
2. Original creators should contact SeaArt.AI staff through official channels to claim their models. We are committed to protecting every creator's rights. Click to Claim
