We took a closer look at some of the new technologies for different disciplines at the agency: Text, image, video - and of course audio, because we will soon be launching our SPS podcast. What can the AI tools really do and to what extent are they suitable for use in a professional environment? Here is the SPS field report: Cool tools under a magnifying glass.
I am surprised at what Descript can do! The range of functions is very wide and the main function, namely transcribing videos and then editing a video based on the text, is amazing - I didn’t expect that. This makes editing super easy: you simply select the words you don't need, delete them as you would in Word and the tool cuts the video at the same time.
The overdub function is ingenious. If you have a slip of the tongue in the recording, you can record your own voice and use it again with text-to-speech. Unfortunately, it only works in English at the moment (which I only realized after I had already recorded a text in German for half an hour!). The principle is quite simple: you read an English text aloud for at least 30 minutes and speak it in, and after a few hours you get a voice model for text-to-speech with your voice.
Another ingenious feature: when you record a podcast, the AI automatically cuts out filler words like "um" and "ah". This completely eliminates the annoying audio editing work. There is a free version with some limits that allow only limited use. However, I can highly recommend the Pro version at 12 USD per month because it is very affordable compared to other editing tools.
"My conclusion: An affordable must-have for all podcast creators!"
Overall, I was convinced by the well-made videos and the great introductory tutorials in the software, the very simple recording, the wide range of functions and, in addition to the low price, the fact that the tool is very suitable for beginners. There are drawbacks for the unclear structure of the data created and for the fact that transcribing does not work that accurately if the pronunciation is not as perfect as it is with a trained speaker. At least it worked better in German (my mother tongue) than in English, which was unfortunately hardly usable. My conclusion: An affordable must-have for all podcast creators!
Rytr is a text tool that automatically generates content that you can adapt or further process. The tool is based on the GPT3 technology from OpenAI, an American company that researches artificial intelligence. Funded by big investors like Elon Musk and Microsoft, OpenAI has developed tools like the image software DALL-E or the chatbot ChatGPT. GPT-3, the third-generation Generative Pre-trained Transformer, is a machine learning model and relies on deep learning to create any kind of text.
"The target group is therefore anyone who needs texts, whether copywriters who use AIDA and PAS text tools for their copy, companies who need ads for Facebook, Google & LinkedIn or posts for social media, or HR departments who develop job descriptions and interview questions."
Basically, the tool is quite easy to use and it doesn't require a lot of expertise - so it's great for beginners too! The dashboard is in English, but the texts can be created in German or other known languages. In the application, Rytr offers a wide range for content generation: from emails and blog articles to social media posts and texts for landing pages or a CTA, everything is possible. In total, there are more than 30 different applications or templates to choose from, so-called "use cases". The target group is therefore anyone who needs texts, whether copywriters who use AIDA and PAS text tools for their copy, companies who need ads for Facebook, Google & LinkedIn or posts for social media, or HR departments who develop job descriptions and interview questions. But Rytr can also be used outside the business environment: the AI even creates song lyrics and stories, as well as video descriptions or ideas for new videos, and under "Magic Command" further applications such as poems or letters are possible.
Once you have decided on a use case, all you have to do is enter a short description or keywords that describe the specific topic in more detail, and the system spits out a text based on this information (and of course on the content learned through GPT-3!) in a matter of seconds. Here you can still choose from up to three variants of the text. You can then freely combine and edit the results. It’s very practical that Rytr has a clear interface with a rich text editor, so that editing and formatting the text is quick and easy. In addition, you can select the tone of the text to express certain emotions, such as appreciative, persuasive or critical. This gives the computer-generated texts a more human touch and formulates the texts more purposefully. The degree of creativity is also selectable, depending on whether a more factual writing style or something unusual is desired.
"The broad keyboard of texts on which the AI plays is quite impressive. However, in order to achieve convincing results in the end, you have to learn a bit in order to be able to give the AI optimal instructions..."
And how convincing is the tool as a whole? The broad keyboard of texts on which the AI plays is quite impressive. However, in order to achieve convincing results in the end, you have to learn a bit in order to be able to give the AI optimal instructions, despite the simple operation. Language and grammar are well implemented, but if you look at the finished texts in detail, there is still room for improvement in terms of context, especially with longer texts. Short texts such as social media posts were quite successful in the test. The integrated SEO analysis and the plagiarism check are real advantages, which makes Rytr a good tool for content for SEO purposes. The tool is excellent for an initial brainstorming session to get started with a topic and to generate ideas and formulations that can then be used to take off further. The very low monthly volume (5,000 characters) in the free package is used up quickly and is sufficient for a quick trial at most. So if you really want to work with it, you will have to opt for a paid subscription - for unlimited access you pay just under 28 euros a month and for monthly limit of 50,000 characters just under 9 euros.
„The tool puts a quickly taken selfie or a headshot of another person on pictures and photographed people whether actors, celebrities or a neighbor. And it does it to a frightening perfection!“
I first used Reface almost two years ago. The app came into my timeline via an advert on social media and, like all AI products, I had to try it because the promise was that I could "become any body I wanted": and it became my trademark.
It is free to use and offers unlimited use in a pro version. The tool puts a quickly taken selfie or a headshot of another person on pictures and photographed people whether actors, celebrities or a neighbor. And it does it to a frightening perfection! This is how I first got to know the tool. But quickly the range of functions was expanded and I could also use my head in videos of various kinds - which was also implemented very well.
In the meantime, I can animate photos, speak texts and the AI manages to change even poorly made source material so that my head sits perfectly in the picture. In addition, the latest feature, the creation of an avatar via a multitude of pictures that you upload of yourself. Completely generated for free. I myself have always given my profile pictures on various social media platforms a new look, just for fun. Harvey Specter or the Godfather then became an "Godfather Thomas", which I made my profile picture. In the background, the AI processes all the images to replace the heads better and better and more clearly.
I can very well imagine that this tool can and will be used for professional purposes. The downside, of course, is that this technology opens the door to abuse. Reface does try to put a stop to this by displaying watermarks on the generated material. But resourceful people will know how to remove them, I fear. All in all, Reface is a fun tool, with enormous AI power. There is hardly a tool that I personally use as often as this one.
Gigapixel AI is an image processing software from Topaz Labs that improves the resolution and details of images. Using "deep learning", the tool can display low-resolution images in higher quality and with greater detail. This is a task that is fundamentally difficult and is solved by most AIs by simply using more pixels. The result is then often an equally flawed image, but in larger dimensions. Not so with Gigapixel AI: with its database filled with portraits, landscapes, architecture and environmental influences, the neural network has an excellent idea of photorealistic shots and thus creates a natural extrapolation of the images.
"When we are faced with the challenge of having only poor-resolution images as source material and also no stock material in sufficient resolution, this tool is a real savior, especially for large print productions!"
Perhaps the most impressive technology behind this program is "Face Recovery", introduced in version 6.1, which generates low-resolution portraits with incredible detail that can be enlarged by up to 600 per cent. Actually, Gigapixel AI was developed for photo-realistic still lifes. Nevertheless, great results are already being achieved in the motion domain. In the future, we can benefit from a wide range of applications, be it in print through incredible quality and detail generation or, for example, in game design through higher resolution structures and finer texture generation. The tool has already been used to create a "remastered" version of Final Fantasy VII, to upscale the textures in Counter-Strike 1.6 and to turn Deep Space 9 into a 4K experience. In agency practice, Gigapixel is a very useful and a comparatively inexpensive tool that produces brilliant results: When we are faced with the challenge of having only poor-resolution images as source material and also no stock material in sufficient resolution, this tool is a real savior, especially for large print productions!
The beginnings of artificial intelligence as image generators go back a long way, but the massive rise in their popularity and widespread availability is a more recent phenomenon. One of the first known generators was Google's DeepDream, a neural network that could recognize patterns in images and "process" them until dreamlike or nightmarish creations emerged with eyes, faces and limbs barely recognizable as the original input image.
A next step were text-to-image generators, most notably DALL-E by OpenAI, which could generate images with simple text input, often with absolutely amazing results. Today, neural networks and algorithms are so advanced that AI-generated image content is making headlines all the time. In practice, however, it is still the case that one can and must understand how the systems react to the text inputs. Illustrative, abstract or artistic interpretations of textual input allow far more leeway than photorealistic text prompts.
"Exciting, perhaps, but nothing that the human brain wouldn't immediately recognize as a mistake."
So if you expect to have found the perfect alternative to stock image pages, you will be disappointed when the generated model has a few too many fingers and the hairstyle connects to a tree in impossible twists in the background. Exciting, perhaps, but nothing that the human brain wouldn't immediately recognize as a mistake. So we are left with abstract, dream-like, colorful scenes that give the feeling that the AI itself can dream. And all this only with a series of highly complex mathematical systems that make something out of nothing.
Fun-Fact: The cover images of our AI blog posts were all created with DALL-E. For our AI glossary, for example, I entered the following text "an expressionistic oil painting of an artificial intelligence glossary". Click here to see the result and our glossary.
AI has evolved incredibly in recent years and is currently getting a huge boost. When AI-based tools are used as tools in the creation process, they can take over time-consuming (partial) work and efficiently deliver immediate results with very good output. With impressive speed, they are well suited for quick inspiration and idea generation and accelerate processes, especially at the start of a project. However, they cannot completely take over the work in all areas. In any case, they are powerful technologies that can support creatives in their work, but not replace them. At least not yet.
At SPS MARKETING, we see ourselves as a B2B Experience Hub and have mastered the entire spectrum in industrial marketing for over 25 years - classic, digital and global. As ReFORMers, TransFORMers and PerFORMers, we work with our clients to implement exceptional concepts at high speed. You too can benefit from our wealth of experience and contact us for a no-obligation consultation!
