Electrified Flux

Trained on Flux1.DEV in the Civitai trainer.

Electrifies your prompt with glows, lights, wires, sparks, electrical arcs and the like.

There’s no particular trigger word but here are some of the words used in training, in deceasing order of frequency:

ethereal,intricate,white,image,detailed,glowing,futuristic,adorned

Example prompts:

glass man <lora:Electrified Flux v1:1>

digital artwork of an intricate glowing woman. Adorned with futuristic details. <lora:Electrified Flux v1:1>

LoRA strength of 1.0 seemed fine in testing. Anything between 0.4 and 1.2 should work.

In Forge, the generation paramaters I’ve been using are often:

Steps: 20, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Distilled CFG Scale: 3, Size: 864x1152

Hires CFG Scale: 1, Hires Distilled CFG Scale: 3, Hires upscale: 1.25, Hires steps: 10, Hires upscaler: 4x_foolhardy_Remacri

The Flux checkpoints I’ve been using with this LoRA are:

Flux.1-Dev BNB NF4

/model/638187?modelVersionId=721627

Fluxmania III

/model/778691?modelVersionId=1205317

Using Forge with either of these models I needed to set:

Diffusion in Low Bits: bnb-nf4 (fp16 LoRA)

or the LoRA had very little effect.

Showcase images were generated in Forge with either of the checkpoints above. The have full gen parameters and don’t use any other LoRAs. “Electrified” seems to work OK with other LoRAs.

Some people have reported that this setting also works, but I have not tried it myself:

Diffusion in Low Bits: Automatic (fp16 LoRA)

If you are using the Civitai generator, default settings should be OK.

Problems: using this LoRA by itself (without hand/leg fixing LoRAs) seemed to produce more mutations than without it, but sometimes I’d get a batch run of 10 or 20 images with no mutations at all. Go figure 🙄

[PUBLISHEDTOCIVITAIONLY]