How each slide is explained
A slide is not running text: it is loose blocks on a page, and often a chart, a table or a diagram. Filmina works with that, and it is where the two things that set it apart from a read-aloud tool come from: it explains instead of reading, and it points at what it is talking about.
What is read from the slide
Two things are kept from every page: the image, exactly as it looks, and the text blocks with their place on the page. That place is what later lets the frame surround the exact block the voice is talking about. The image goes to the model along with the text, so it also reads what is not written in blocks. To keep the thread, it also gets a short summary of what was explained on the slide before.
Not everything written on the page is content. Before explaining, running headers, footers, page numbers, dates and stray addresses are dropped: numbers and dates are recognised by their shape, and headers and footers because they repeat page after page along the top or bottom edge. If they got through, the full coverage rule would force an explanation of the footer.
A scanned presentation is a photo of each slide, with no text inside. It's explained from the image: Filmina recognises the text on each one, with its place on the page, and the model looks at the image as it does with any other. Whatever can be read, handwriting included, is explained; whatever cannot is left out rather than guessed. If no text is recognised, there are no blocks and the frame surrounds the whole slide. If you have the original presentation, upload that instead: it reads better.
Segments
The explanation does not come back as one block of text: it comes back split into segments of one idea each, between three and seven per slide, ten to thirty spoken seconds each. Every segment knows which blocks of the page it covers, and only groups blocks that sit together: a heading with its paragraph, a box with its list. A segment explains its own blocks and nothing else: it does not jump ahead to what comes further down, so the voice never talks about one thing while the frame points at another.
That is what lets the frame move: the segment being spoken says where to look. It is also the unit you navigate by, because the up and down arrows jump from one idea to the next, not from one second to the next. In the script, each segment is a paragraph.
You don't pick a level of detail: every slide is explained in full. The only shorter version is the one the Skim step of the plan uses, for something you've already seen, as explained in Study plan.
The rules of the explanation
The explanation is written to be heard, not read, and that comes down to specific rules:
- It does not read it out. It rephrases, connects and says what each thing means and why it matters: what a definition states, what a formula computes, why a step follows from the one before. The slide is already in front of you.
- Absolute precision. Figures, names, formulas, dates and technical terms are kept exactly as they appear. Nothing gets rounded and no term gets swapped for a casual synonym.
- No preamble and no wrap up. No «let's look at», «on this slide we see» or «we are on slide 22 of 62». Those are the seconds you spend waiting for the part that matters, so the first sentence is already saying something about the subject.
- No formatting. Plain flowing prose: no asterisks, bullets, numbering, tables or emoji. None of that can be heard.
- Notation, in words. «P of A given B», not «P bracket A slash B». Symbols and formulas are written the way they are said before they reach the voice, which is where read-aloud tools break, and the symbol said is the one on the slide, not the one usually used in the subject. A long formula is not dictated: it says what shape it has and what each part does.
- What was already said is not repeated. If the slide keeps the same heading as the previous one, or repeats blocks that were already explained, they are not defined again, not even in passing: only what this slide adds gets told.
- The reading list is not read out. Authors, editions, publishers and links break up the explanation and are not something you study. A reading list slide is dealt with in one sentence, and a link or a long code is never dictated: it says what it is.
Coverage is checked in code
Asking a model not to skip anything is not enough: it has to be checked. When the answer comes back, every block is checked to make sure none was left unexplained, and if one is missing the model is asked again for that block alone. The slide's images count too: each one gets explained, or the model has to say it is only decoration. And if a text block is still missing after the second request, it is read exactly as written: less elegant, but nothing is left out.
Coverage does not depend on the model doing as it is told. It is checked, and whatever is missing is asked for again.
Charts, tables and diagrams
The model gets the slide as an image, the same one you are looking at, as well as its text. That way it reads what is not in the blocks: the data in a chart or a table, the steps of a diagram, a screenshot, the small note in a corner. A table is explained by its pattern, with the values that show it, not row by row. And what it cannot read with confidence, it does not guess.
When a segment is only about an image, the frame goes to it. A slide that is nothing but an image still gets explained, if the image says something. If it is only decoration (a logo, a background, a photo that comes along), the voice asks you to take a look and moves on, instead of describing an ornament to have something to say.
The loose labels of a diagram are a case of their own. They are explained together, in a single segment, describing what the diagram shows and how its parts relate according to the slide. They are labels on a drawing, not claims made by the material: giving each loose word its own segment would mean inventing a definition for each. Formulas, on the other hand, are not labels: for each one it explains what it computes and what each term stands for.
Your notes count too
If the presentation has notes, Mina takes all of them into account: each slide takes the lines that share its least common words in the file, and the model gets them along with the blocks. What a note adds gets told; what only underlines something already on the slide gives that part more weight. Apart from those notes, nothing else in the space goes into the explanation: Mina uses your other files, books or audio to answer you in the chat, not to explain. How to write notes is in The notebook.
Changing a note redoes only the slides those lines talk about; the rest stay as they were.
Which language it explains in
The explanation is in the language the presentation is written in, English or Spanish, not the app's: a presentation in English is explained in English even if Filmina is set to Spanish. It is detected once, on upload, and the voices on offer are that language's. If there is not enough text to tell, the app's language is used.
If something slips through
The rules make the model invent far less, not make it unable to ever invent. That is why every paragraph of the script has a This is wrong: tap it and Filmina looks at the slide again, with its image and your notes, and if you were right it rewrites that paragraph and voices it again. What is confirmed is learned for that presentation, and if the same kind of error repeats across several files, for all of them. How to use it is in Listening to explanations.
It gets ready ahead of you
Nothing is prepared until you tap Explain. From then on, while you listen to a slide, the next two are being prepared in the background, so you hardly ever wait when you move on. The whole file isn't prepared at once: only what you're about to hear.
The slide you are looking at always goes first: if you jump to another, it moves ahead of the ones being prepared, and Start with this one moves it up on the spot. Closing the file stops whatever was still queued, and when you come back it picks up what is missing.
What gets generated is saved in your account. The script is saved per language and the audio per voice: changing voice does not think the explanation through again, it just says it again. And listening again to what was already explained does not generate it again or use up minutes from your plan.