I don’t know if anyone has, has had or knows someone who Have an online storebut if so, a big hug. Mounting a store is not precisely easy and much less when your products have to look. If you sell, say, sweatshirts, the only way to teach how they look is photographing a model that has been put. And each model is a world: they have an eye color, a hair, skin color, etc. You have to take into account locations, make content for networks, in short. It is a really complex and, of course, an issue that can solve.
That is what he proposes Dreamshota generative artificial intelligence tool Made in Spain who won the hackaton of AI2030 (event that Xataka was Media Partner) And that we will know today better. And attention, because perhaps it can be useful to some Xatakero who wants to undertake with his own fashion brand and does not know where to start.
Fashion AI
Image | Tamara Bellis
How does a generative artificial intelligence work? Grosso (Very Grosso) Modeif we want an AI to learn to generate images of a horse you have to give a lot of images of a horse. Thousands, tens of thousands. If we do that with fashion images with different styles, funds, contexts, postures and models we will obtain an AI capable of generating images of models that integrate the elements we want.
That is exactly what they did in Dreamshot, train a personalized model (Dreamshot v2) with More than 10,000 fashion images. Javier Jiménez, founder and CEO of the company, explains in conversation with Xataka that Dreamshot is based on Flux. Dreamshot is not a founding model as such, but a Fine Tunning Flux for which fashion images have been used both licensed and own. The first model, Dreamshot V1 Aka Lookbook is Open Source and can be used from Prompthero.
Behind each product photo there are hours and hours of work and a huge investment | Image: Khaled Ghareeb
Javier states that the photos are both own and ceded by the brands, ergo with permission. “We have not scrape anything from the Internet (…) It is an ethical training, we have all the rights of the photos we have trained with,” he says.
How does it work? Although it does not completely eliminate the need to make a photo shoot, it does significantly reduce the amount of photos we have to do and the work and time invested. Because? Because if we want to squeeze the tool to the maximum we will have to upload photos of our productsideally varied photos, from different angles, with different perspectives and in different places. Minimum five. The thing is that once these photos have been made, the possibilities are huge.
Dreamshot interface | Image: Xataka
On the left the reference image of the green jacket. To the right, the result generated by the AI using the prompt “60 Years Old Man Wearing Bottle Green Overshirt. He is in A Forest Full of Trees During Sunny Day” | Image: Xataka
Great here to better appreciate the detail | Image: Xataka
Once we have the photos, we can upload them to Dreamshot and ask the AI to generate images of said product in other contexts or in models that do not exist. Or take stock photos that in another context would require mounting a complete set. It is a way to reuse photos to infinity. You don’t have to take photos of the product again, but ask the AI to generate them for us.
On the left the original image, on the right the one generated with the changing the model for a person with dark skin, with beard and dark green eyes | Image: Xataka
Another option that works surprisingly well is that of change models. If we have a photo of a light skin model and light eyes and we want one of a model with dark skin and dark eyes, in another context two models should be counted. With Dreamshot we just have to upload the photo and ask the AI to change it by entering a prompt. The result is quite striking, although a trained eye will quickly distinguish that the image is generated with AI. However, things as they are, the substitution of faces is spectacular.
On the left the original image, on the right the generated with the changing the model for a blonde woman with blue eyes | Image: Xataka
So is the Background changesomething that in normal conditions would mean displacements and time. This, more than for the store, it is interesting to use it to generate content for social networks. If we have a timeless garment or accessory, we may be interested that the networks in networks go in tune with the season of the year, for example. Well, it will be enough for us to select the photo of our model in the mountain and ask you to put it in the city, on the beach, in a forest or in front of the Roman Colosseum.
On the left the original image, on the right the generated with the bottom of the Roman Colosseum | Image: Xataka
It should be noted that Dreamshot divides the photo into three parts: model, clothing and background. We can change any of all three, but not alter the image itself. It is not possible to take the photo of a person and undress her. We could change the shirt or change the model, but not wear a photo of a dress dressed and undress or put it in swimsuit or interior. It is a dangerous matter, but Dreamshot has covered himself well and with success in this sense.
Why Dreamshot?
How does this idea arise? Javier Jiménez tells us that his first company It was a jewelry ecommerce“I was 10 years and every month we had photo sessions and there were thousands of euros, a month of preparation for each photo shoot … Then the AI came out and I turned and made a community that was one of the largest worldwide, Prompthero“. Surely that name is familiar to those who have been curious about AI.
After this, Jiménez decided to set up a startup that combines his two previous experiences: “I want to help take photos with generative for eCommerce.” This is something slow, very slow. I have mounted a couple of online stores, one of them dedicated to jewelry, and I know what it is to take photos, prepare setups, locations, etc. When I tell Javier, he laughs and confesses that with that type of business “you can complicate everything you want and you can spend all the money you want.” Dreamshot aspires to solve this.
The company is recent and small, it was created just half a year ago and consists of three people, but Javier confirms that they already have “enough customers” whose names cannot be revealed, but he does tell us that most are fashion groups, shoes and sunglasses. Jiménez slides that they are now “starting with car customers.” Nor can we talk about brands, but it is a movement with all the meaning of the world. The AI understands pixels, so for practical purposes it does not matter changing the bottom of a photo of a shirt, some shoes or a car.
Javier believes that they have started “before the market is ready and we have the technology already developed. I think that in the coming months there will be a boom. The brands are beginning to adopt this 2025 and as we are already positioned there they are even coming to us to look for those solutions and integrate this technology in their day to day.”
The future
Imgen | Stefano Intintoli
A question that could arise when seeing this in operation is How will it impact models. Javier explains that his AI “does not train with the models, but with the shirt. You are never using the face of the model for anything, it is not replicating the identity of the model and taking photos of the model.” The example that Jiménez puts is good: “If you take Rafa Nadal, you have to ask permission to replicate his image.”
The other option would be at the opposite end: models or influencers that license their image so that companies can use them in tools with AI. “That will come for sure,” says Javier. “There will be brands that can be allowed to take influencers that could not be allowed otherwise and there will be influencers that, without leaving home, could earn money.”
There have not yet been approaches in this regard, but until it happens, if it happens, from Dreamshot they have other plans. The first, A line of virtual models. Right now you can’t choose which model appears in the photos generated with the tool, but soon yes. That will allow consistency in images and campaigns.
Image | Bruce Mars
The second, to achieve the best image and video quality of the market, something that also depends on the progress of large technological ones such as OpenAI. In that sense, Jiménez believes that “we are going to spend the game in one or two years. Within a few years the image quality will be brutal.” Javier understands that the generative AI opens the door to an ultrapersonalized internet in which, when entering a website, the image is different for each user.
The third, open segments. Today is clothes, tomorrow they will be cars and Javier believes that the following could be the furniture. Again, it makes a lot of sense anymore will escape that it is a huge market. So is that of beauty and makeup, but it is more complex for the amount of colors, textures and options that should be taken into account. The AI is not yet prepared for that. Still.
Hackaton winners in AI2030
As we said before, Dreamshot was the hackaton winner who took place in AI2030event dedicated to generative artificial intelligence and its application in the professional world of which Xataka was Media Partner.
They won with a most interesting proposal: An artificial intelligence system focused on programmatic advertising. Basically, they developed a proposal to analyze the performance of the ads issued so far and generate, depending on the conclusions extracted, ads that work better with new images and specific customizations.
Cover image | Atikh Bana