October 22, 2026 · Triox Informatika · 6 min read
Image, voice, video—the spectacular side of AI at the company
AI does not only write; it draws, speaks and makes films. We look at which tool is for what—images in ChatGPT and Gemini, Midjourney, Adobe Firefly, voice cloning, video generation—what a company can use it for, and where the legal and trust boundaries are.
In brief
- Image generation is now built into the major assistants: ChatGPT and Gemini create an image from a single sentence and modify existing ones. For professional work, Midjourney and Adobe Firefly are the two main directions.
- Voice is two separate areas: text-to-speech (natural speech from text, in any language) and transcription (text from speech). Both are mature and immediately useful in daily work.
- Video generation is spectacular, but at companies it is for now suited to short clips and internal material. The tools from Google and OpenAI are the best known.
- The limits are not technical: copyright, brand identity, deception and the risk of voice cloning. A simple internal rule is enough, but there has to be one.
The parts of this series so far have been about text: emails, documents, analyses. But it is the spectacular side of AI that captures most people: a sentence becomes an image, a paragraph becomes a voice, a description becomes a short film. At a company, this area is at once the most fun and the easiest to get wrong. In this part we go through what is for what, and where to stop.
Images
The built-in image generators
Most people do not need a separate image generation tool. ChatGPT and Gemini create an image from a single sentence: “create an illustration for the internal newsletter that represents the summer maintenance shutdown, in a simple, bright style.” They also modify existing images: “remove the background,” “make it brighter,” “redraw it in a line-art style.” Quality has jumped so much in the past year that for most internal purposes—presentations, newsletters, intranet, social media posts—it is more than enough. Text in the image also comes out legibly, which was not the case for a long time.
Midjourney—the artistic direction
Midjourney is the tool whose images look most like a “photo” or a “painting.” For marketing material, mood images and campaign ideas, this is the professional choice. It is handled a little differently—a web interface, many settings—and it is paid. For those who regularly produce creative material it is worth it; for occasional use, the built-in tools are enough.
Adobe Firefly—the safe direction
Adobe’s image generator differs in that the model was trained on rights-cleared images: Adobe’s own stock material and freely usable content. This matters at companies: Adobe undertakes that the generated image can be used for commercial purposes. Built into Photoshop and the other Adobe programs, it complements the existing designer’s workflow—it does not replace the designer; it speeds them up.
What is it good for at a company?
- Illustrating internal material: newsletter, intranet, training material, presentations.
- Brainstorming: ten drafts for a campaign before the designer gets to work.
- Quick image edits: removing the background, resizing, variations.
- Social media: images for regular posts when there is no photo of your own.
What is it not for?
- Logos, brand identity. A generated image cannot be protected the way a designed identity can, and a brand should not be built on a random style.
- Real people, real products. If the customer believes the photo shows the real product and it does not, that is deception.
- Official documents, certificates. Obviously.
Voice
In the area of voice, AI has quietly but completely changed the game, and this is the part that is most directly useful.
Text-to-speech
Today’s text-to-speech tools speak naturally, with intonation and emotion, in Hungarian too. ElevenLabs is the best known, but OpenAI, Google and Microsoft also offer excellent voices. What is it good for?
- Narration for training material and internal videos, without a studio or an actor.
- Customer service menus and automatic responses that do not sound like a machine.
- “Listening” to long material in the car—we wrote about NotebookLM’s audio summary in the Gemini part.
- Multilingual material: the same text in ten languages, in the same voice.
Transcription—speech to text
The reverse direction is also mature: a recording turns into accurate, punctuated text broken down by speaker in minutes. Teams, Meet and Zoom have this built in, but a voice memo from a phone or a dictaphone interview can be transcribed too. For anyone who records client meetings, interviews or training sessions, this is the biggest time saver.
Voice cloning—the delicate area
A few seconds of voice sample is enough for the AI to speak in anyone’s voice. That is good when the managing director wants the text for an internal video read in their own voice and has no time to record it. And it is bad when a fraudster calls the accountant in the boss’s voice—we described that case in our article from early summer. The lesson is twofold: your own voice may be cloned only with explicit consent and documented, and the company should have a password or call-back rule for every financial request that arrives by voice only.
Video
Video generation is the newest and most spectacular area. The tools from Google (Veo) and OpenAI (Sora) create surprisingly realistic clips of a few seconds from a single sentence, and they are available in the Gemini and ChatGPT interfaces too. Other players—Runway, for example—offer more detailed control for the creative profession.
What is it good for at a company—today?
- Short clips for social media, presentations, internal communication.
- Mood and concept videos for discussing a campaign before the production company gets to work.
- Avatar training videos: a “presenter” reads out the internal policy, in several languages, without a studio. There are separate tools for this (HeyGen or Synthesia, for example), and many large companies already use them for training.
What is it not for—yet?
For longer videos with a storyline that require accurate product depiction, the technology is not yet consistent enough. The clips are short, details can change from one frame to the next, and the computing cost is high. This may change within a year or two, but today video is more of a supplementary tool than a main one.
The limits—not technical, but legal and trust questions
Copyright
The copyright status of images, voice or video made with AI is still unsettled in many countries. Two practical rules: for commercial use, use a tool whose provider undertakes that the output can be used (Adobe does this explicitly; the others’ terms are worth reading); and do not ask for the style of a known artist, brand or protected character.
Labeling and deception
Under the EU AI Act, artificially generated or modified content must be labeled in certain cases, especially if it imitates a real person or event. Regardless of the legal obligation, a simple rule protects you from losing trust: if the customer could believe they are seeing something real and it is not, say so. The illustration in the internal newsletter does not need a label; a “customer testimonial” with an AI avatar does.
Personal data
Colleagues’ faces, voices and names may be put into an AI tool only with consent. That is not an AI question but a data protection one.
Brand identity
An easily made image tempts everyone to make their own, and the company’s materials quickly drift apart. A short internal guide—what style, what colors, what we use it for and what not—solves this, and the designer will be grateful too.
How to try it
- Ask for an image for an internal newsletter in ChatGPT or Gemini. Three versions, in different styles. See which one fits the company.
- Have a page of text read aloud by one of the text-to-speech tools in Hungarian. Compare it with what you heard two years ago.
- Have a meeting transcribed—if the Teams or Meet note-taker is switched on, it is already there.
- Write a half-page internal rule about what you use, what you label, and what you do not put in front of AI.
Who is it for?
Marketing people, internal communications, trainers, and anyone who regularly produces material that until now required a designer, a studio or a videographer—and for whom speed and volume matter more than a perfect end result. It does not replace professional creative work, but it drastically shortens the road to it.
In the next part we turn to the most practical topic: how to ask AI so that it gives a good answer—because most disappointment comes not from the tool but from the question.
Would you like a simple internal rule for material made with AI?
If you have a question, a request or a specific need related to this topic, write to us. We will help you write a half-page guide that covers the legal and the brand considerations at the same time.
Triox Informatika Kft.