If you are using ChatGPT, Gemini, Claude and so on for your prompt generation, you can feed it with this paste so the model knows directly how to use the LoRA - example prompts included.

Aetherium Couture: Dark Angel is a fashion LoRA focused on dark gothic couture, sculptural bat-wing structures, thorn-like silhouettes, sharp ornamental detailing, sheer lace, black mesh, glossy leather, patent leather, dark metallic accents, dramatic collars, shoulder constructions, gothic crowns, and elegant fallen-angel inspired fashion.

The LoRA is designed to transfer a recognizable dark angel design language onto different garments rather than forcing one fixed outfit. It works especially well with gowns, bodysuits, corsets, lingerie, fitted dresses, jumpsuits, tailored jackets, lace garments, sheer mesh, leather, patent leather, and other body-framing couture silhouettes.

Trigger

aetherium_couture_darkangel

What it can create

• sculptural black leather and patent leather bodices

• sheer black lace and nude illusion mesh

• thorn-like and spiked garment structures

• small bat-wing shoulder accents

• large bat-wing backpieces

• winged collars and sculptural neckpieces

• gothic corsets and fitted bodysuits

• dark angel lingerie

• dramatic sheer gowns and mermaid silhouettes

• cape-like wing drapes

• ornamental gothic filigree

• blackened metallic accents

• gothic crowns, tiaras, and spiked headpieces

• dark runway couture

• fitted jackets and structured fashion pieces with gothic detailing

• elegant, seductive, aggressive, or regal dark fashion

The LoRA responds well to detailed descriptions of the garment type, material, wing size, shoulder construction, amount of sheer fabric, and whether the result should feel elegant, gothic, sensual, aggressive, regal, or theatrical.

Recommended prompting

Describe:

• the base garment

• whether the material is lace, mesh, leather, patent leather, or a combination

• whether you want small shoulder wings or large back wings

• the shape and placement of spikes, thorns, and pointed structures

• the amount of sheer or nude illusion fabric

• whether metallic details should remain subtle or become more prominent

• whether a crown, collar, headpiece, or winged backpiece should be included

Useful wording includes:

small bat-wing shoulder accents, sculptural winged collar, thorn-like leather detailing, sheer black lace over nude illusion mesh, glossy black leather corset bodice, dark metallic filigree, gothic crown with spiked filigree, cape-like bat-wing drapes, large sheer lace bat wings, sharp architectural shoulders, black lace lingerie with winged details

It can also transfer its dark angel design language into garments that were not a major focus of the dataset, such as tailored jackets, pencil skirts, fitted jumpsuits, mini dresses, and other structured fashion pieces.

Notes

Large wings, small shoulder wings, gothic crowns, spiked collars, and dramatic backpieces are available, but they are not forced when the prompt is specific.

For a more restrained look, describe small shoulder accents, subtle thorn-like detailing, or a single gothic focus area such as the collar, shoulders, or bodice.

For larger theatrical results, explicitly describe large bat wings, dramatic winged backpieces, elaborate gothic crowns, cape-like wing drapes, or extensive spiked structures.

The LoRA has a natural tendency toward fitted, sensual, body-framing silhouettes with lace, mesh, cutouts, and dark glossy materials. If you want a more controlled or covered design, describe the garment clearly as a structured gown, tailored dress, covered bodice, long sleeves, opaque skirt, or restrained couture silhouette.

This is, just like my other clothing LoRAs, not a “single outfit LoRA”. Instead, I trained the design language, giving you the freedom to create similar looking outfits and garments.

Would love to see what you create with it.

The LoRA was trained on the default flux1-dev model for maximum compatibility.

Dataset and training data:

• 98 images

• 5 steps per image

• 10 epochs

• bucks with 1152px max size, no auto scaling

• cosine scheduler with 0.2 warmup and 0.7 decay

• Captioning was done with OpenAI

• alpha == dim: 32

If you use it with other loras, for example a character lora, try to lower the weight - I had good results at around 0.4 - 0.5.

As always: I did not polish any of the images in the showcase, they all came out like this. I did 2 images per prompt and picked the better looking one.