GPT Party 3.0. Alexey Koloskov: How neural networks will change our reality GPT Party 3.0, the largest Russian-language networking event dedicated to artificial intelligence, took place in Silicon Valley on March 9-10. More than 450 people gathered at Plug and Play to meet leading experts, entrepreneurs, and investors, discuss the latest trends in artificial intelligence, and gain practical knowledge. At GPT Party, Alexey Koloskov, founder of Photigy.com, spoke on the topic “How neural networks will change our reality.” Alexey shared his experience of using AI tools to create photo and video content, and also showed where AI is already becoming an indispensable assistant for creative people. Alexey Koloskov, founder of Photigy.com I am the head of one of the largest online schools for subject photographers Photigy.com. I am also an influencer and entrepreneur of the year according to Google+. At one time, I was one of the top ten most influential photographers teaching in the world, and I had two million followers, but they went away with Google+. Now I am actively interested in how artificial intelligence, in addition to business processes, can help creative people. We are entering an era where, watching a video, no one will be able to clearly say whether it was real or whether it was created by a neural network. Therefore, it will be critical for creators to learn how to use AI tools. Stable Diffusion is an open solution that we can use. It is developed by a group of enthusiasts who write the code and train the models. If you use it in business, you will have to pay, but for personal use it is free. It is a model that works independently of anyone, you can review the code, everything is open. It is an open platform, unlike enterprise solutions such as Midjourney or DALL-E, which are great and very effective in narrow solutions, but there are a number of problems. The first problem is censorship. Censorship and creativity do not mix, in my opinion. There are a number of queries that only Stable Diffusion can fulfill. The second problem is that it is impossible to know what rules corporate neural networks operate by. Let’s look at an example. A simple query: a bottle of wine with a girl in blood drawn on it. Below are the results that Midjourney and Stable Diffusion gave. Why is blood a problem? Nobody knows for sure. For example, this label looks normal to me, but in some countries it may be unacceptable. Although if you look at it in general, this is something that can help increase wine sales. If a person is brainstorming, then Stable Diffusion can help with this better than any other neural network. Let’s look at how you can use Stable Diffusion in your business. GPT Chat and Stable Diffusion can create content that cannot be recreated by any other tools. Let’s start with video advertising. We can adapt the same video to different audiences, making it more attractive to each of them. For example, one video can be aimed at young people, and another – at an older audience. We can create different characters and use different design styles. Stable Diffusion generates such videos very quickly, processing 6-second material in a couple of hours. And it’s completely free. You can create several videos per day, each of which will be unique in color schemes and characters for different target groups. I have seen how such videos reduce the cost of clicks and impressions several times, as they look unusual and attract attention. It really works. The next application is e-commerce photography. At our school, we teach product photographers who specialize in commercial photography – for example, jewelry photography. These photographers often worry that AI will soon take their jobs, but AI will not be able to create an image of a specific object in the near future. For example, if someone sells a product on Amazon or elsewhere, they need a photo of the real object. Here’s an example: I took a photo of a perfume bottle next to a window, holding a white sheet of paper in front of it as a background. Then the AI inserted this object into various scenes in less than 60 seconds. This is called outpainting, when you take a real object and paint the environment around it based on prompts. Another example of outpainting with wine. I took a picture at home and wrote the prompt “wine on a table in a restaurant with windows.” Look how it works: artificial intelligence even adds shadows and highlights. It’s not just cutting and pasting a picture in Photoshop. To make a person believe that this bottle is really on the table, you need an hour of work in Photoshop. In Stable Diffusion, this is done in one click. All this is absolutely free and can be quickly configured. Then I took a picture not only of wine, but also of a bouquet of flowers. I simply changed the prompt to “bouquet of flowers.” Artificial intelligence understood that it is necessary to work with a bouquet of flowers, and this is the whole charm for photographers. Influencers are people who have to create a lot of visual content. Take a video of a girl, with Stable Diffusion you can easily change her clothes or appearance. In the video below, I even inserted my face. Again, this is done very quickly and easily. A person who has to post a lot of content can choose characters, manage different accounts and generate a huge amount of content very quickly and for almost nothing. The same can be done with famous movies, and this 6-second clip will only take an hour. The next niche is prototyping. Stable Diffusion is a great tool when you don’t have a finished product yet. You have an idea and you create a sketch. For example, I drew a simple sketch just with a mouse, without even using a tablet. Then, you make a request, adding a base and additional details, such as the material of the vase – ceramic, shiny, blue. Artificial intelligence generates a vase based on this prompt. Prototypes and visualizations that previously required the use of an editor and took a lot of time and resources are now done almost instantly. What used to be done in special heavy and expensive editors can now be done in Stable Diffusion. You can specify the input data for the interior or exterior, define the front parameters, choose colors, textures and materials. This allows you to quickly create different options for clients without the need for 3D modeling. One of the main advantages of Open Source is the huge community, where everything is open and available for public use. The models you see are trained models for Stable Diffusion, taking up only a few gigabytes. You can download them and use them, most of them are even free for commercial use. In general, on the left we have corporate solutions that cost money and are quite narrow in their functionality. All the same and even more can be done in Stable Diffusion. Indeed, you can create things that I have never seen anyone do before. At the same time, the workflow can be created of any complexity. We are currently working on an application that will make the process even easier. Our main goal is to create a marketplace where on one side there will be guys who are able to solve serious problems using artificial intelligence, and on the other – clients who need it. If you combine them, it will be a huge market.
