Lesson 5 of 6 · 15 min · Checked on 10 October 2026
Creating with AI: images, voice, video and slides
By the end of this lesson you will be able to:
- Tell apart what AI already creates well from what still comes out badly
- Ask for an image with subject, style, light, framing and what you do not want
- Turn a text of your own into a presentation or a voice-over
- Respect consent, trademarks and the duty to label generated content
The six-fingered malassadas
Rita has a pastry shop and wants a poster for her Carnival malassadas (Azorean fried dough). She asks the AI for an image: "malassadas on a platter, with a pastry cook sprinkling sugar, and the text MALASSADAS DE CARNAVAL".
The image is lovely. But the pastry cook's hand has six fingers, and the text says "MALASADAS DE CARNAVL", with letters missing. Rita asks for another image, this time with no people and no text, with free space at the top. Then she writes the title herself in a poster program.
This is what this lesson teaches: AI helps a lot with creating, as long as you know what it still does badly.
What already works well, and what still comes out badly
| Already works well | Still often comes out badly |
|---|---|
| An image for a poster or a post | Hands and fingers: too many, too few or crooked |
| A temporary logo, until a designer makes one | Text inside the image: swapped letters and invented words. It has improved a lot, but check letter by letter |
| A voice-over for a short video, with a synthetic voice | The same character across several images: the shop's mascot changes face from image to image |
| A presentation from a text of your own | Real places and people: your parish's church will not come out the same |
The logo is "temporary" for a reason: an image made only by the machine is, in general, not protected as your work, and it may look like a brand that already exists (Level 1, lesson 5).
How to ask for an image: five pieces
- Subject: what appears. "A platter of malassadas sprinkled with sugar."
- Style: realistic photograph, watercolour, child's drawing, vintage poster.
- Light: morning light, late afternoon, studio light.
- Framing: from above, from the side, close up, with free space for text; vertical or horizontal.
- What you do NOT want: no people, no hands, no text, no logos.
Realistic photograph of a platter of malassadas sprinkled with sugar, on an old wooden table. Natural morning light, coming in through a window on the left. Seen from above, slightly tilted, with free space at the top for me to put text. Vertical format. No people, no hands, no text, no logos.
Tip
To fine-tune, change one piece at a time: "Same, but with late afternoon light." If you change everything at once, you do not know what improved.
Tip
Always write the poster text yourself, in a poster or presentation program, on top of the image. That way there are no swapped letters and you can correct the date without generating everything again.
Voice, video and presentations
Voice. You write the text, choose a synthetic voice and listen. Write numbers and abbreviations the way they are said ("three euros fifty", "C T T" for CTT, the Portuguese post office). Local place names (Calheta, Urzelina, Topo) are sometimes pronounced badly: try writing them another way until they sound right.
Video. There are already tools that make short clips from a request, but the movements still come out strange. For the shop or the association, it often works better to film with your phone and use the AI for the text, the subtitles or the voice-over.
Presentations. You give it your text and ask for the structure:
Turn this text into a presentation of 8 slides for the parents' meeting. On each slide: a short title, at most 3 one-line bullet points and a note with what I should say. Text: [paste here]
Some assistants create the presentation file straight away; others give you the text to paste into the program you use. Some of these features only exist in paid versions or have daily limits: see the Current state page.
Rights: people, trademarks and labelling
Real people. A real person's face and voice are only used with their consent. In Portugal, the right to one's own image is in article 79 of the Civil Code. Never use children, neighbours or colleagues in a fake image or video without permission. Not even as a joke.
Trademarks. Do not ask for "a logo similar to the one from [well-known brand]" or for cartoon characters for your products. Even when made by the AI, it may infringe the rights of whoever owns the brand.
Labelling what was generated. The AI Act (the European regulation on AI) has transparency rules in article 50. Since 2 August 2026, anyone who uses AI at work, in a company or in an organisation, to create or alter images, audio or video that resemble real people, places or events and could pass as genuine (a deepfake), must say that this content was generated or manipulated. For a clearly artistic, satirical or fictional work, it is enough to state it in a way that does not spoil the work.
Official sources: Regulation (EU) 2024/1689, on EUR-Lex and article 50 on the European Commission's AI Act Service Desk. This is not legal advice.
Watch out
If in doubt, label. A sentence like "Image created with AI", legible and visible at first sight (in the caption or on the image, not tucked in a corner), is what Article 50 asks for: clear and perceptible at first exposure. And never present as real what was invented: customers who do not exist, a product more beautiful than it is, a person saying what they did not say. Even in personal use, where this AI Act rule does not apply, it is a matter of good faith.
Exercise
How I know it worked an image with no letters or strange hands, with free space, and a final poster with the text written by you and a note that the image was created with AI.
If it didn't work
- Does your tool not create images? Not all of them do. Try another one from Level 1, lesson 4, or ask: "Write me the image request so that I can use it in another tool."
- Does it insist on adding text? Repeat at the end of the request "no text, no letters, no words", or crop that part out of the image.
- Strange hands? Ask for no people, or for a framing in which the hands do not appear.
- Did it refuse the request? If you asked for a real person or a well-known brand, the refusal is on purpose.
Videos
In the specified section: builds a slide deck from documents, with the sources in NotebookLM, a prompt for the slide script, Gemini with Canva and the export. The rest of the video is a course advert.
ChatGPT Images 2.5 tutorial: generate images, edit them, remove backgrounds, transform photos and combine reference images.
Test what you learned
Corrections to this lesson (1)
- : Second external review: the «Image created with AI» label must be clear and visible at first sight, not tucked in a corner; the warning and the exercise were adjusted.