AI in Digital Marketing. Mobio Group’s Research
Neural networks have gradually, but densely entered our daily lives. And now it is difficult to name a field of human activity that does not use AI technologies. Naturally, marketing and advertising have not been left out of the technical progress. Both actively use all the possibilities of neural networks. In Internet marketing AI is used to improve the results of social networks and search engines, to show the most relevant content and advertising messages to users, to monitor and analyze the market, to track the behavior of the target audience and so on. Neural networks have helped marketing teams get rid of complex multidimensional or monotonous chores and make them faster and more efficient.
One of the most popular topics today in the application of AI in marketing is the discussion of GANs. Generative Adversarial Networks are a class of deep learning frameworks with a generative model structure. In simple terms, generative AI is a class of machine learning algorithms designed to create new original content based on a set of input data. GANs can perform creative tasks previously thought to be unique to humans, such as creating text, images, music, code, Web sites, and videos.
In this article Mobio Group systematizes the applications of GAN in the advertising business that we have identified as promising.
Visual Implementation
1. Image generation on the basis of the text
Artificial Intelligence makes it possible to generate text-based images. Based on given parameters, theme, style, or location, words can be used to create the desired visual material, both animated images and realistic pictures, indistinguishable from photographs. Networks are capable of creating completely realistic images of non-existent people or animals, abstract and surrealistic landscapes and pictures that cannot be made with a camera, and even virtual worlds.
2. Semantic image translation and 3D shape generation
Based on a semantic image or sketch, a realistic version of an image can be created. By analyzing millions of photos, a neural network recreates scenes and landscapes that mimic real landscapes. AI also generates text, numbers, and 2D images into 3D shapes with topology, rich geometric details, and textures. To build a photorealistic world, the 3D artist needs to create the bare minimum, and GAN will continue the process of 3D modeling complex landscape scenes.
3. Transforming an image into an image
Neural networks have an infinite number of transformation variations, and the user regulates the desired result by his requests. What can be done with the use of different theses (queries):
1. Transfer the style of one image to another or prescribe a certain style in the query with text (realism, anime, fantasy, cyberpunk, sketching, steampunk, etc.). GANs also make it possible to get images in the style of famous artists (like Van Gogh or Dali) or photographers.
2. Transform image elements such as color, medium or shape while preserving the constituent elements (e.g. turn a daytime image into a nighttime one).
3. Restore or reconstruct photos and low-quality images merge images.
4. Create convincing fakes of celebrities which are hard to distinguish from real photos.
Video Generation
AI allows you to convert text or image into video. Suitable for generating simple “speaker+phone” videos. The quality of more complex videos is still questionable. Google, for example, does not yet provide access to using its video creation system, but demonstrates the results of its service capable of generating video with a resolution of 1280×768 pixels and a frequency of 24 frames per second based on verbal queries.
The ability of AI algorithms to superimpose one person’s face on another person’s real video leads to deepfake. There are a lot of funny deepfake videos online, but the number of cybercriminals and deepfake crimes is also growing. Services are already being created that automatically recognize fakes, but it is worth noting that new machine learning technologies are constantly emerging that make fakes more and more realistic.
The main image generators are Midjourney, DALL-E, Stable Diffusion, Deep Dream Generator, NightCafe, AI Playground, Let’s Enhance. Almost all services are not free, except for a small trial period, when beginners are given free minutes, energy, attempts, etc.
For details on testing the potential of AI to create graphic and photorealistic images for various agency tasks in practice, see our series of articles (here and here).
The main video generators are Pictory.ai, Phenaki, InVideo, Imagen Video, Veed.io, Lumen5, Designs.ai, Elai, Synthesia. As with image generators, video generators are not free either. At least the ones that deserve close attention.
Simple videos don’t require much professionalism and are recorded in a few minutes. With Synthesia we made this simple clip for free. We made this kind of video with Pictory.ai, but the potential of this service is much wider if you pay attention to it.
The main design generators are Uizard, Brandmark, Khroma, VanceAI to help you create designs for mobile apps, websites or weblogs, logos, color palettes and more.
Audio Implementation
1. Text-to-speech generator
Against the backdrop of the development of Text-to-Speech (TTS) technology, speech synthesizers that recognize context-aware text using artificial intelligence have become publicly available. This technology has many applications, including audiobooks, movie voiceovers, podcast recording, “speech navigation,” voice chat, etc. Synthesized speech can be recorded in audio format or saved in OGG format. However, so far all these methods cannot surpass the quality of the natural human voice.
2. Conversion of speech to speech
Generative AI applications related to sound involve generating voice using existing voice sources. Using STS, you can create voiceovers for advertisements or games without hiring a voiceover specialist.
3. Music Transformation
With generative AI, it is now possible to create music based on your preferences for style, genre, etc. These tools can be used in many areas of creativity, including games and advertisements.
The main voice generators are iSpeech, Text-to-Speech, Voicemaker, Play.ht, and Lovo.ai.
The main music generators are AIVA, Soundraw, Amadeus Code.
In marketing, when using these services it became possible to create background music for advertisements without having specific musical skills. We tried to record a melody for a commercial with Christmas discounts theme. This tune is our first experience in this direction. The result is quite modest, but the prospects are promising — neural networks were able to complete the unfinished Ludwig van Beethoven’s 10th Symphony (BeethovANN Symphony 10.1).
Textual Implementation
In marketing and games, generative AI is used to create dialogs, headlines, advertising slogans and texts, blog posts, chat rooms for real-time communication with customers, or for creating product descriptions, articles and content in social networks.
There are many neural networks for text generation — Frase IO, Peppertype, Outranking, Writesonic, and GPT-3 is one of the largest (up to 4.5 billion words generated per day).
Code Generation
Another application of generative AI is software development because of its ability to generate code without the need for manual coding. Because of this quality, code development is possible not only for professionals, but also for people who are not involved in programming. Thus, ChatGPT produces full-fledged code (macros, scripts) for Google Sheets and programming languages (Python, etc.) with further application and automation of routine work.
In marketing, the code generator can be used to write the right bots. For example, a bot that will send links to competitors’ articles which have got more than 100 likes. This is how the interest of the audience in a particular subject will be monitored.
The main code generators are CodeWhisperer, Copilot, Tabnine, AlphaCode.
Creating content (text, audio or video) with GAN is the future that has already arrived. And this means that it’s time to start using the opportunities neural networks give us.
So, for example, ChatGPT can be used to get:
information (request for a list of the app’s 10 main competitors)
a short text in the form of a summary containing the main information from the article or text (text translation with an extract as an example of an article from AppsFlyer)
semantic kernel (request for keywords for the application with subsequent advertising)
You can use ChatGPT to generate naming, headlines, texts, descriptions, terms of reference, and media planning. For a broad overview of ChatGPT features (including the potential in ad campaign analytics).
The most promising image generators by far seem to us to be Midjourney and DALL-E. To compare the two neural networks and investigate how applicable they are to the agency’s routine tasks, we took three real-world tasks facing the Mobio Group team for testing.
The first task was to create a large figure of 5% in gentle pastel colors from the attributes of spring for an advertising banner. Midjourney was unequivocally better at this task. True, the percent sign did not lend itself to any network.
Request:
number 5 assembled from flowers and butterflies, spring pastel colors
Midjourney
DALL-E
The second task was to create recognizable meme images but without 100% copying them to avoid legal implications of use.
For the test, we tried to get the images of Dwayne Rock Johnson and a famous meme with a smiling cat.
Both networks handled this task quite well, but DALL-E made it a lot easier to get the right result thanks to its tool for generating variants similar to the original image.
Dwayne Johnson Midjourney
Request:
https://s.mj.run/YcIg_hysdcI dwayne the rock johnson smile, looking at camera, poster, reflections, dynamic pose, Cinematic, Color Grading, Depth of Field, hyper-detailed, insane details, intricate details, beautifully color graded, Unreal Engine, Editorial Photography, Depth of Field, DOF, White Balance, 32k, Super-Resolution, Megapixel, ProPhoto RGB, VR, Massive, Halfrear Lighting, Backlight, Natural Lighting, Incandescent, Optical Fiber, Moody Lighting, Cinematic Lighting, Studio Lighting, Soft Lighting, Volumetric, Contre-Jour, Beautiful Lighting, Accent Lighting, Global Illumination, Screen Space Global Illumination, Ray Tracing Global Illumination, Optics, Scattering, Glowing, Shadows, Rough, Ray Tracing Reflections, Lumen Reflections, Screen Space Reflections, GB Displacement, Scan Lines, Ray Traced, Ray Tracing Ambient Occlusion, Anti-Aliasing, FKAA, TXAA, RTX, SSAO, Shaders, OpenGL-Shaders, GLSL-Shaders, Post Processing, Post-Production, Cel Shading, Tone Mapping, CGI, VFX, SFX, insanely detailed and intricate, hypermaximalist, elegant, hyper realistic, super detailed, dynamic pose
Dwayne Johnson DALL-E
Cat meme. Original:
Midjourney
Request:
https://s.mj.run/pe-FYjlI8q4 white cat, ears back, look at camera, close , cute face, scottish lop-eared, looking at camera, poster, reflections, dynamic pose, Cinematic, Color Grading, Depth of Field, hyper-detailed, insane details, intricate details, beautifully color graded, Unreal Engine, Editorial Photography, Depth of Field, DOF, White Balance, 32k, Super-Resolution, Megapixel, ProPhoto RGB, VR, Massive, Halfrear Lighting, Backlight, Natural Lighting, Incandescent, Optical Fiber, Moody Lighting, Cinematic Lighting, Studio Lighting, Soft Lighting, Volumetric, Contre-Jour, Beautiful Lighting, Accent Lighting, Global Illumination, Screen Space Global Illumination, Ray Tracing Global Illumination, Optics, Scattering, Glowing, Shadows, Rough, Ray Tracing Reflections, Lumen Reflections, Screen Space Reflections, GB Displacement, Scan Lines, Ray Traced, Ray Tracing Ambient Occlusion, Anti-Aliasing, FKAA, TXAA, RTX, SSAO, Shaders, OpenGL-Shaders, GLSL-Shaders, Post Processing, Post-Production, Cel Shading, Tone Mapping, CGI, VFX, SFX, insanely detailed and intricate, hypermaximalist, elegant, hyper realistic, super detailed, dynamic pose
DALL-E
The third task was to obtain a photorealistic image based on the following description: a tired 40+ woman sitting in the kitchen, drinking tea and eating cookies, gloomy tones.
We tried both generating from various text queries and generating from a base photo, but after many attempts we were unable to get an image with the scene we wanted.
Midjourney
Request:
This photo shows a middle-aged woman sitting alone in her kitchen on a cold and rainy evening. She is sad and looks downcast, her eyes fixed on the cup of coffee in her hands. The room is dimly lit, with only a small lamp providing the only light. The woman is wearing a simple grey dress and her hair is tied back in a low ponytail, her face expressionless. She looks lost in thought, her mind miles away from the gloomy kitchen. It is a poignant image, one that speaks to the loneliness and isolation that many of us feel in our daily lives.
Request:
This photo depicts a middle-aged woman sitting in a dimly-lit kitchen and biting into a cookie. She looks sad and downcast, with her eyes cast downward and her shoulders slumped. Her expression suggests she is in a state of deep contemplation and her posture suggests she is feeling isolated and alone. The kitchen is sparsely decorated and has a somewhat gloomy atmosphere. The woman is wearing casual clothing and her hair is tied back in a low ponytail. The cookie she is holding in her left hand is still whole, but her right hand is already starting to break it into small pieces. The scene conveys a sense of sorrow and loneliness.
DALL-E
Request:
“This photo shows a tired woman in her 40s sitting in her kitchen in the twilight, taking a bite from a cookie. She looks exhausted and overwhelmed, but still manages to take a break for a snack”.
Evaluation of Results and Conclusions Mobio Group
Midjourney
A great tool for a generation of AI artists. Beautiful and quality results. For our work will be suitable as an art generator and for creating graphic elements on a given theme.
Weaknesses:
- Working in discord and lack of API for easy integration into business tools makes it very difficult to use in the daily life of the agency.
- The engine is quite slow, even compared to DALL-E, but the price is five times lower “per try”.
DALL-E
Good tool for the designer. There are great tools for editing, replacing, finishing — very useful.
There’s an API for integration with Python business tools.
As a “masterpiece generator” relative to Midjourney, it’s average. Although maybe it needs to be studied better before drawing final conclusions. Maybe DALL-E needs a different way to build queries.
The two tools can work well together: Midjourney to generate art and DALL-E to edit it.
In any case, at this stage, AI can not replace the work of the designer, because his work is not only to draw images, and involves primarily thinking through the concept that can solve the problem.
Software updates or new plug-ins for familiar programs allow us to create more original images and optimize our work. The specifics of the work of the creative department of a performance agency is that we often need many iterations and variations of creatives before we find the most suitable ones to promote a particular product of our client. Speed and the ability to create content quickly in this environment becomes crucial to success.
Olga Mazur — Head of Creative, Mobio Group
MidJourney is a new round of development, and while we think the current version of artificial intelligence does not allow for accurate predictive results, it is definitely applicable.
We plan to:
1. Use MJ as an alternative to stock images
2. Create some elements of banners or marketing materials
3. Use as a substitute for the work of illustrators
Artificial intelligence is becoming a great auxiliary tool for analytics, creative ideas, text and art generation, and for successful advertising campaigns in general. Neural networks are constantly self-learning, and we need to learn how to work with them.
Full list of all available neural networks by category: images, texts, sound + catalogs/samples. Almost all of them are available without registration or at least free.
IMAGES AND VIDEOS:
1. https://creator.nightcafe.studio/ (generates images and videos);
2. https://openai.com/dall-e-2/ (create, finish drawing, add objects to the image);
3. https://deepdreamgenerator.com/ (creates realistic images in specified styles);
5. https://www.starryai.com/ (nft generator);
6. https://www.fotor.com/ (nft art);
7. https://runwayml.com/ (video editing and editor, animation and 3D models);
8. https://www.wombo.art/ (surrealistic designs, turns photos into cartoons);
9. https://thispersondoesnotexist.com/ (generates a portrait of a nonexistent person);
10. https://looka.com and https://www.logoai.com/ (draws logos);
11. https://artsexperiments.withgoogle.com/poemportraits (superimposes text on a portrait);
12. https://dream.ai/create and https://hotpot.ai/art-generator (pictures by description, without registration);
13. https://stablediffusionweb.com/ (generates pictures by text description, no registration);
14. https://www.midjourney.com/home/ (avatars in different styles, combines pictures into one, generates by text description);
15. https://www.ostagram.me/ (combines styles of different pictures into one);
16. https://sketch.metademolab.com/ (animates sketches and children’s drawings);
17. https://imglarger.com/, https://vanceai.com/ (improves image quality);
18. https://lexica.art/ (all ai graphics);
19. https://beta.tome.app/, https://www.beautiful.ai/ and https://piggy.to/ (create presentations with colorful slides on any topic by text request);
20. https://www.creatosaurus.io/ (generally everything in one place);
21. https://www.craiyon.com/ (generates 9 pictures from 1 image in one click, without registration);
22. https://prisma-ai.com/lensa (forgotten classic);
23. https://h5.tu.qq.com/web/ai-2d/cartoon/index (turns photos into anime);
24. https://uizard.io/ (makes site layout by hand sketch);
25. http://goart.fotor.com/ (stylise a photo to look like famous portraits or paintings);
26. http://deepangel.media.mit.edu/ (removes people or objects from an image);
27. https://experiments.withgoogle.com/cartoonify (turns a photo into a cartoon);
28. https://dezgo.com/ (generates pictures without censorship);
29. https://mage.space/ (regular clear convenient generation of pictures, without restrictions);
30. https://wonderdynamics.com/ (creates movies, approved by Spielberg);
31. http://photoai.io/ (very advanced photo-editor, changes models’ backgrounds and clothes);
32. https://www.tattoosai.com/ (generates tattoo designs);
33. https://github.com/microsoft/visual-chatgpt (Visual ChatGPT — neuronics allows you to edit images right in chat. For example, it removes any objects, changes their color, knows what objects are in the picture. It remembers the context and edits well. Of course, you can generate pics by any request);
34. https://pebblely.com/ (generates content (pictures). Ideal for smm).
VOICE AND SOUNDS:
1. https://voice.headliner.app/ — voices of famous people;
2. https://imaginarysoundscape.net/ — photo voiceover;
3. https://www.riffusion.com/ — generates music by text description;
4. https://podcast.adobe.com/enhance — increases audio quality to studio level;
5. https://otter.ai/ — turns call recording into text, convenient for telephone conversations;
6. https://experiments.withgoogle.com/sound-maker — creates your own sounds;
7. https://experiments.withgoogle.com/scribbling-speech — turns speech into animation in real time;
8. https://www.figma.com/community/plugin/1159123024924461424 — converting any website to editable Figma design. This plugin will allow you to use the design of any site as a ready template for your own projects. It can also be useful in redesigning the old site when the layouts were lost. All you need to do is install the html.to.design plugin and import the HTML code of the desired site into Figma by simply copying the link. The best part is — all objects will be editable, which means you can fully customize the design.
roomGPT.io is a neural network that generates interior design. If you have been planning to renovate your apartment for a long time, this service will definitely help you. You just need to upload a photo of the room, and the neuronics will come up with a couple of seconds a new design in the style of your choice. Everything is free.
