Uwear.ai
p/uwear-ai
Generate on-model photos of real clothes in few clicks
Reda Mjahed
Uwear.ai β€” Generate on-model photos of real clothes in few clicks
3
β€’
Generate on-model photos of your clothing with one single flat-lay photo as input in less than a minute.
Replies
Reda Mjahed
Maker
πŸ“Œ
Hey Hunters! πŸ‘‹ After more than a year of iteration and exploration we are thrilled to announce the public launch of Uwear.ai and our core AI model Drape v1. We started this journey March 2023, dead set on building the best and most accessible virtual fashion camera for the industry. We tried many technical approaches, from fine-tuning to VTON models, but none really worked: - Fine-tuning was expensive and required multiple photos of the product, ideally worn, to deliver the best result. But businesses, big and small, only have a flat lay or ghost mannequin image to start with. - Most of the focus of AI research in this space has been around virtual try-on, the most famous being OOTDiffusion and IDM-VTON. But if you are a store who only cares about displaying your products on models, VTON approaches mean generating a first image, then inpainting the product on it. Inpainting clothing onto existing images often results in a loss of visual consistency between the clothing and the rest of the image (lighting, composition, wrinkles, etc.) - Although some papers like MagicClothing were close to what we were looking for, a straightforward text2img pipeline with a flat lay image as a condition, the quality wasn’t enough. So we had to train our own model. Drape is the result of this explorative journey. An SDXL adapter trained on a dataset of high quality public data we built and curated ourselves. πŸ‘˜ It is very good with text and prints πŸš΄β€β™€οΈ Can generate an image in 20 steps πŸ’» Run under 8G VRAM We have put a demo up for the curious folks on HugginFace But if you are a brand or manufacturer you can head directly to Uwear.ai Up next: V1 is far from perfect. Complex clothing, or not frequent in our dataset, may require up to 20 generations to get a good photo. Our focus is to get this number to 1 and reach a confidence level where you can publish visuals from our tool directly to your store with no human revision. We are early in our journey, and firmly committed to build the industry standard for generative visuals in the apparel industry. Looking forward for your feedback & features ideas! Cheers
John Wade
Insanely realistic. Can I make a profile and try clothes on myself?
Reda Mjahed
@john_wade thanks John, not yet but it is in the roadmap!