Artificial intelligence in photography

Main article: Generative artificial intelligence

2025

Developing a revolutionary method that dramatically speeds up and simplifies the generation of realistic images and texts

A team of Russian researchers from the Moscow Institute of Physics and Technology, Innopolis and Skolkovo has created an innovative generative modeling technology that dramatically speeds up the process of creating realistic images using artificial intelligence. A new algorithm called "optimal flow matching" allows you to obtain high-quality visual data in significantly less time compared to traditional methods of neural networks. The scientific breakthrough in the field of machine learning became known on September 22, 2025.

Russian experts have solved the fundamental problem of modern generative models - the inefficiency of data transformation trajectories. Scientific results were presented in the materials of the international conference NeurIPS 2024, which confirms the high level of Russian achievements in the field of artificial intelligence.

The technology is based on a flow matching method designed to smoothly transform one data distribution into another. The procedure resembles the creation of a virtual "river," along which elements of information move from the initial to the final state, gradually modifying their characteristics. Previous approaches formed complex winding routes, which significantly slowed down generation and required large computing power.

According to the publication, the fundamental difference between the Russian development lies in the use of specialized vector fields that determine the optimal straight-line trajectories of data movement. The mathematical basis of the method is associated with gradients of convex functions that provide the exact direction of information transformation without unnecessary deviations from the shortest path. ^[1]

Creating one image using AI requires as much energy as fully charging a smartphone

Creating one image using generative AI requires as much energy as you need to fully charge an average smartphone. This is stated in a joint study by the American company Hugging Face and Carnegie Mellon University, the results of which TAdviser got acquainted with in mid-May 2025.

The purpose of the work was to determine which AI tasks require the most energy, and therefore lead to the most significant carbon dioxide emissions into the atmosphere. The ten most common operations were considered, such as answering questions, generating text, classifying, creating images, etc. 88 different AI models were used to perform these tasks, and 1000 requests were processed for each of the operations. The researchers measured the energy consumed using a specially designed tool called Code Carbon. After that, SO2 emissions were calculated.

𝓲

Фото: Freepik

Experts came to the conclusion that image generation is the most energy and carbon-intensive AI task of those selected as part of the study. In particular, the creation of 1000 images using a powerful AI model such as the Stable Diffusion XL results in emissions of the same amount of carbon dioxide as a 6.56 km ride on an average gasoline-powered car.

On the other hand, text generation requires relatively little energy: 1000 requests require only 16% of the energy used to fully charge the smartphone. At the same time, the volume of CO2 emissions is comparable to their amount when overcoming only one meter of the way by car with ICE. In general, the authors of the work say, AI models optimized for certain tasks consume less energy than general-purpose neural networks when performing the same operations.^[2]

Top 10 Image Generation Tools

The most transformative influence on the economy is exerted by the GII segment in image and video generation. Wide segments will be affected here:

• video games,
• design and art,
• cinema (3-4 years in the future), marketing and advertising, content for the media industry and social networks, education and training (interactive courses).

In the future, technologies can be applied in architecture, industrial engineering, medicine, etc.

Current list (as of January 2025) of top image generation tools based on Spydell Finance tests:

1. FLUX

2. Midjourney

3. ImageFX от Google

4. Ideogram

5. Recraft

6. Playground

7. Dall-e

8. Artflow

9. Leonardo

10. Stable Diffusion

2024

Global Computing Photography Market Size Reaches $15.2 Billion

In 2024, spending on the global computational photography market amounted to $15.2 billion. More than 40% of this amount was in the Asia-Pacific region. Such data is contained in the Fortune Business Insights review published in early November 2025.

Computational photography is an approach to creating and processing images using digital computing. In particular, special algorithms automatically correct exposure and white balance, eliminate noise and correct distortions. Computational photography technologies allow you to apply various effects, including panoramas and HDR. In general, this approach makes it possible to use amateur cameras and smartphones to receive pictures comparable in detail to professional equipment.

One of the main drivers of the market is artificial intelligence. Neural networks are able to analyze the scene and objects in the frame: on the basis of the data obtained, optimal settings are used to achieve the best result. AI algorithms are used to eliminate noise, correct distortion and improve the quality of photographs obtained in low light conditions. Generative AI offers qualitatively new creative possibilities, such as changing stage lighting and background editing. AI technologies are widely used in image processing in professional fields such as medical imaging and machine vision systems.

The growing penetration of smartphones is having a positive impact on the industry. According to Exploding Topics, released in mid-June 2025, there are 7.21 billion smartphones globally (many people own several devices). Against this background, the popularity of mobile photo and video shooting is growing. At the same time, computational photography provides scene recognition, automatic exposure adjustment and correction of images based on AI, which allows you to obtain high-quality images with minimal effort.

The authors of the study cite as a deterrent significant costs associated with the creation and implementation of complex image processing technologies. The development of advanced hardware and software requires significant financial resources, but in the context of the prevailing geopolitical situation and macroeconomic situation, many companies are forced to reduce the size of investments.

Depending on application, the market is segmented into 3D visualization, virtual reality, augmented reality, mixed reality, standard digital visualization, etc. In 2024, the first of the listed sectors provided the largest share of revenue - 29%. At the same time, the direction of consumer electronics is leading, which is associated with the widespread use of smartphones and other equipment with built-in cameras. Geographically, Asia-Pacific dominates with 42.8% of revenue, or $6.51 billion. Globally, the major industry players are:

• Apple;
• Samsung;
• Google (Alphabet);
• Huawei;
• Xiaomi;
• BBK Electronics;
• Sony;
• Nokia;
• Lenovo;
• LG Electronics;
• Honor;
• Qualcomm;
• MediaTek;
• Pelican Imaging;
• Nikon;
• Micron Technology;
• Nvidia;
• Infineon Technologies;
• Renesas Electronics;
• Applied Materials.

In 2025, the computational photography market is expected to reach $17.4 billion. Fortune Business Insights analysts believe that in the future, the CAGR will be 15.7%. Thus, by 2032, costs may increase to $48.38 billion.^[3]

Google introduced a free neural network Gemini 2.0 for creating AI assistants

On December 11, 2024, Google introduced the next generation artificial intelligence model - Gemini 2.0. This universal neural network is capable of generating images, text and sound. Read more here

Free AI has been released to create 3D models from photos. Game development has become even easier

In early December 2024, Microsoft introduced a new artificial intelligence-based service, the Trellis platform. This open source neural network is designed to generate 3D models of objects from a photo or text description. Read more here.

Wildberries launched a generative AI service for creating photos of clothes on virtual models

On November 1, 2024, the Wildberries marketplace introduced the new Virtual Photo Studio service, which uses generative neural networks to create photographs of clothes on virtual models. The service is available to sellers with a Jam subscription for 133 categories of men's and women's clothing. Read more here.

Russian scientists have created a "doubting" neural network capable of recognizing unknown objects in the photo

Students of MISIS University and MIPT, together with scientists from the non-profit artificial intelligence research laboratory T-Bank AI Research, proposed the SDDE ensemble neural network (Salience Diversified Deep Ensembles), which more accurately identifies objects in images not uploaded to databases. MISIS announced this on October 31, 2024. Read more here.

Free photo background replacement service launched in 2 minutes

At the end of October 2024, the free online service IC-Light V2 was launched, with which you can quickly replace the background with photos. In addition, this tool allows you to control the lighting in images. Read more here.

Marketplace "Sberbank" launched an AI-generator of photos of clothes on a virtual model

On October 10, 2024, it became known that the Megamarket marketplace, owned by the Sber ecosystem, presented an innovative service that uses artificial intelligence to generate photographs of clothes on virtual models. This technology is designed to simplify the process of creating content for merchant cards in the Fashion segment. Read more here.

Free neural network for Midjourney image generation launched

At the end of August 2024, a browser version of the Midjourney neural network, designed to create images by text description, became available. A certain number of pictures can be generated free of charge. Read more here

"Vkusville" began to generate package design using neural networks to speed up the release of products

Vkusville began using neural networks to generate package designs to speed up the release of products. This was reported on February 26, 2024 in the retailer's official Telegram channel. Read more here.

2023

Adobe sells AI-generated photos of the Israel-Hamas war. The media use them and pass them off as real

In early November 2023, it became known that Adobe it was selling generated artificial intelligence images depicting the conflict between Israel and the group HAMAS in the Gaza Strip. These illustrations have varying degrees of realism; moreover, sometimes the media pass them off as real, without explicitly indicating that they were created by. neuronets More. here

Yandex has released a neural network capable of creating images according to the description

Yandex has released a generative neural network Masterpiece, which can create images according to the description. The company announced this on April 5, 2023. Read more here.

2022: Russian scientists presented a method for classifying photographs based on a quantum neural network

Russian physicists of the Laboratory of Quantum Information Technologies of the University of MISIS, the Russian Quantum Center and Moscow State University named after M.V. Lomonosov first presented a method of classification of photographs with high accuracy for 4 classes of images, based on the architecture of the quantum convolutional neural network (QCNN). Representatives of NUST MISIS reported this to TAdviser on November 24, 2022. Read more here.

2021: Application of AI in photographic environments

The modern world can no longer be imagined without neural networks and increasingly impressive developments in the field of artificial intelligence. These entities, 20 years ago, which seem to us to be something from the field of science fiction, today penetrated literally every house. And no, we are not talking about robotics and artificial intelligence, artificial intelligence is a much broader concept.

AI algorithms help us start the day with a weather forecast, then build the best way to work in Google Maps, taking into account traffic on the roads, and in the evening our favorite media service will offer the film to our liking. Rest assured, even ads in the browser will be selected by "them" specifically for you, given previous searches.

Even the sphere arts of AI was not spared. For example, among professionals and photography lovers, techniques based on neural networks are becoming increasingly widespread. This article discusses the practical application of artificial intelligence tools in a photographic environment.

2017: Real-time photo retouching algorithm

On August 4, 2017, it became known that Google engineers, during a joint study with scientists from the Massachusetts Institute of Technology (MIT), created an algorithm that retouches photos in real time.^[4]

Developers are increasingly resorting to the so-called "computational" photography: various algorithms and AI technologies designed to improve images taken with smartphones. According to representatives of Google and MIT, their technology will not only allow high-quality processing of photos, but also obtain a result comparable to that of a professional photographer.

During neural network testing, five photos created by Adobe and MIT were selected. Each image was retouched by five different photographers. The resulting images were then used to determine exactly how each image can be improved by adjusting brightness, saturation and other parameters.

On the left is the original 12MP photo, on the right is the same photo processed by the Google-MIT algorithm

Images captured by modern cameras are often perceived as raw material for photography. Before uploading a photo to social networks, even those who shoot on the phone spend a minute or two aligning color and contrast using popular applications, says the MIT blog.

The Google-MIT algorithm is based on a "convolutional neural network." The processing power of modern mobile devices is not high enough for the full operation of the system, but the researchers were able to bypass this limitation - the system performs most of the calculations on a reduced copy of the original image, and then transfers the results to a high-resolution photo.

The researchers tested the program on a regular smartphone (the model is not indicated), and the algorithm was able to display a processed image with a 1920×1080 resolution and a refresh rate of 40-50 Hz in real time. The size of all software does not exceed the size of one digital photo and can be used to process images in various styles. According to the researchers, the neural network can be "trained" on a new set of images to mimic the style of certain photographers.

Notes