The LoRA creates contemporary clothes from a variety of designers. You can download images in the gallery as they contain the metadata, including the prompts (examples are also provided below). The LoRA can follow complex prompt instructions as it was trained on 154 images with detailed labels. The LoRA never saw human faces (I removed all of them) to limit any potential impact: it's focused on clothing, not supposed to alter the face of your subjects or any rendering style.
Some of the designers that inspired this work include:
Pauline Dujancourt
Rejina Pyo
Steven Stokey-Daley
Daniel Lee
Erdem Moralioglu (I am a major fan so the dataset reflects my preference)
Roksanda Ilincic
Oscar de la Renta
How to use
Describe what the person is wearing: what type of garment ('knee-length dress'), what color ('muted pink'), what pattern ('abstract floral pattern in deep maroon and white'), what material ('layered fabrics'), any additions or cut-out ('oversized sweater adorned with large, irregular cut-outs in the shape of flowers, which are cut from the fabric').
Don't worry if you are not a dress designer: just ask ChatGPT. Here is a ChatGPT prompt that you can copy/paste: "Your task is to describe a dress for a woman in one paragraph, including the type of cut, color, pattern, material, and any other alteration."
Sample prompts
DLSR photo of (Margot Robbie, 25 yo woman). She is walking on the French Riviera. She looks at the viewer with a natural smile. She wears pink sunglasses. She wears an elegant, stylish outfit consisting of a striking, knee-length dress that features a unique design. The dress is predominantly a soft, muted pink color adorned with a bold, abstract floral pattern in deep maroon and white. The pattern includes stylized, oversized flowers and leaves. Style of Erdem Moralioglu. cl0th1ng
Portrait Photo of (20 yo asian male, DylanWang) holding a book in a bookstore. He is looking at the viewer with a natural smile. His outfit is a blend of intricate, layered fabrics and accessories that create a visually stunning, futuristic look. The top of the outfit consists of a dramatic, asymmetrical, black and white kimono-style garment with a bold, dark floral pattern. cl0th1ng", "populated_text": "Portrait Photo of (20 yo asian male, DylanWang) holding a book in a bookstore. He is looking at the viewer with a natural smile. His outfit is a blend of intricate, layered fabrics and accessories that create a visually stunning, futuristic look. The top of the outfit consists of a dramatic, asymmetrical, black and white kimono-style garment with a bold, dark floral pattern. cl0th1ng
DLSR photo of (ana de armas, 25 yo woman) at the beach, smiling at viewer. The gentle waves roll in the background with sailboats. She leans against the trunk of a coconut tree. She is is wearing a stunning, strapless gown that features a unique, mosaic-like design. The dress is constructed from a material that appears to be a combination of fabric and glass, creating a visually striking, stained glass effect. Style of oscar de la renta. cl0th1ng
Description
FAQ
Comments (4)
See update in comments.
I created some wildcard files to use with this LoRA. Be sure to still add the trigger word to your prompts.
color.txt (631 various colors)
clothingFemaleAllLoraDesigner.txt
clothingFemaleDressLoraDesigner.txt (Dresses only)
clothingMaleAllLoraDesigner.txt
@BICEPS It's a cool idea. You made a good use of the example prompts I gave. Some of your keywords may work (or not) 'by chance' though, because they're not necessarily in the tags of the training set. If you'd like, I can ZIP all of the training labels and send them your way. That way you can create wildcards that will trigger, including on combinations that you may not have envisioned.
@Ranoer Go for it. I had used Gemini Pro with Antigravity to build the wildcards from your description here and had some pretty good results (at least imo). If you send me the training info I'll go ahead and refine these and post an update. Thanks.
I looked at the training data you sent me and did some tests and you're right that the LoRA won't activate unless you're prompting pretty specifically for what you trained it on. As it turns out Z-Image Turbo isn't so bad with some of this designer stuff on it's own which had me fooled. I refined the the wildcard file to use all of the descriptions you trained on. Here it is:












