A five-slide brand carousel from one chat message
The Versely Agent pitches carousel directions, writes every slide caption before a single image exists, then renders the set with the text baked in.
The message in the recording is six words: "Generate 5 slide slideshow for my brand." What comes back is not five images. It is a pitch, then a plan, then a rendered carousel whose captions were written before any pixels existed.
That order is the whole feature, and it is the opposite of how a slideshow generator usually works. Most of them take your sentence, make N pictures, and leave you to letter them afterwards. Here the text is the plan and the images are the render, which is why the placeholder tiles in the video are already captioned while the progress line still reads Generating - 0 of 5 slides.
The brand pill is doing more work than the prompt
Look at the composer before the message is sent. There is a Brand chip reading Abbie's, and the input placeholder has changed from "Message Versely AI..." to "What should we make for Abbie's?"
Nothing in the typed message says maple syrup, Canada, amber-gold, or Instagram. All of it arrives from the kit behind that chip. Six words plus a brand pill is a fuller brief than three paragraphs without one, because the paragraphs would still be your description of the brand rather than the brand's own products, palette and voice. Set the brand kit once and every later message inherits it — see the brand kit as the style guide for what actually lives in one.
Attach the chip. Then type the short thing.
Step one is a pitch, not a render
The agent does not start generating. It comes back with directions and asks you to choose:
Love it — a 5-slide carousel is a great format for Abbie's. Before I fire up the slideshow, let me pitch you a few directions so we make the right one (pick one word, or blend).
The answer in the recording is three words: "Let's go with 1st."
This step is cheap and it is the one most people skip by over-specifying. A carousel has exactly one job — a scroll-stopping through-line across five frames — and that job is decided by the angle, not by the adjectives. "Beyond Pancakes" (five ways to use pure maple syrup that aren't pancakes) is a different product from "Why Pure Matters" (additives, non-GMO, Canada). Same brand, same slide count, same cost. Choosing between them costs one word.
If you paste the five captions you already wrote, you have turned the planner into a renderer and thrown away the only step that could have found a better carousel than the one you walked in with.
The plan is the deliverable
What comes back after the pick is a production sheet, not a vibe:
Format: Instagram carousel, 4:5 portrait · warm amber-gold palette, cream background, red Abbie's accents · hero food-forward photography with clean text overlays
- Cover — "Beyond Pancakes" 🍁 — hero bottle shot, subtitle: "5 ways to use pure maple syrup (that aren't pancakes)"
- Coffee — "Sweeten your coffee, naturally" — maple drizzling...
Format, aspect, palette, accent colour, photographic treatment, then a numbered caption per slide. Read it before you approve it. This is the last point at which changing your mind is free — everything after it is a render.
The captions exist before the pixels
This is the part worth watching frame by frame. The agent fires run slideshow agent, says "On it — sending the 'Beyond Pancakes' brief to the slideshow studio now," and a card appears reading Generating - 0 of 5 slides.
Zero of five. And yet all five tiles are already lettered:
| # | Caption on the placeholder tile |
|---|---|
| 1 | beyond pancakes · 🍁 5 ways to use... |
| 2 | sweeten your coffee |
| 3 | upgrade your oats |
| 4 | yogurt's best friend |
| 5 | dinner deserves maple too |
The copy was locked at plan time and handed to the renderer as the brief for each frame. That ordering has a practical consequence: the image is composed for its caption — negative space where the words go, a food-forward hero that leaves the top third quiet — rather than a caption being dropped onto whatever came back. It is also why the deck reads as a sequence instead of five unrelated pretty photos. A slideshow is a sequence of claims; a slideshow is not a video model.
Then the confirmation:
Your "Beyond Pancakes" carousel is rendering — 5 slides, drawn with GPT Image 2 for that premium food-photography look, each with its text overlay baked in.
Baked in, not overlaid at display time. What lands in your library is five finished 4:5 JPEGs you can post anywhere, with no caption layer to lose.
It offers you the next step while it waits
While that finishes — want me to draft the post caption + hashtags for when you're ready to publish?
The render takes a minute or two. The post caption does not depend on the render. Saying yes there costs you nothing and removes the step that usually kills a carousel two days later, when the images are sitting in a folder and nobody has written the copy.
The Post button skips the download
When the card flips to 5 slides ready, it grows two buttons: Open and Post.
Post lands you in the Auto Post composer with the work already loaded — 5 selected · carousel in the media tray, a caption box with a Write with AI row (Casual, Hype, Professional, Funny), and the schedule panel below. The only thing standing between that screen and a live post in the recording is the line "Connect an account first."
That is the whole distribution path: no download, no re-upload, no reordering five files by filename in a phone gallery. Post or schedule to social costs 1 credit per connected account.
Open goes the other way, into the slideshow editor, where individual frames can be swapped without re-rendering the set — change a slide, not the whole slideshow.
Round two: Pinterest is a source, not a style
The second half of the recording is a different message against the same brand:
Generate slides from pinterest for my brand
The agent posts Fetching memories and user profile... and then does something more useful than obeying. It remembers the first deck and refuses to repeat it:
You've already got the "Beyond Pancakes" Pinterest set — so for this round, which way do you want to go?
- Why Pure Matters (my pick) — pure maple vs. "pancake syrup", zero additives, Non-GMO & gluten-free, straight-from-Canada story.
- Fresh pins, same theme — a new photo batch of "Beyond Pancakes" so you can pick the best shots.
- Name a topic — breakfast meal-prep, maple cocktails, gifting season — your call.
Those three options are genuinely different purchases: a new angle, a new batch of the same angle, or your angle. Option 2 exists because sometimes the deck was right and the photos were not, and you should not have to re-argue the concept to get a second roll of film.
The set that comes back carries the same treatment — claim lines per frame, ending on a CTA slide ("From Canadian forests to your table — link in bio"). Open any frame and the Details panel names its provenance: Model: Pinterest · Type: Image · Item 3 of 5, with a Copy prompt button. Searching Pinterest for inspiration is a named job; here it is feeding a slideshow rather than a moodboard.
What this costs you in attention
| Step | Your input | What it decides |
|---|---|---|
| The message | 6 words + brand chip | Format and count |
| The pitch | 1 word | The angle — the only irreversible choice |
| The plan | Read it | Palette, aspect, five captions |
| The render | Nothing | Images composed for the copy |
| Post | Pick accounts | Distribution, 1 credit each |
Two decisions. Everything else is reading.
The failure mode is not "the AI made bad slides." It is approving the plan without reading it, then blaming the render for a caption you never looked at. The plan is printed in the chat precisely so that does not have to happen.
FAQ
Do I have to pick one of the pitched directions?
No — the pitch says "pick one word, or blend." Naming your own topic is option three in the second round for the same reason. The pitch exists to stop you defaulting to the first angle you thought of, not to limit you to the agent's three.
Can I control the aspect ratio and palette?
They are stated in the plan before the render — 4:5 portrait, warm amber-gold, cream background, red accents in this run — and the plan is the checkpoint. Say what you want changed there and it is free. Say it after the render and it is a re-render.
Why are the captions burned into the images?
Because the deliverable is a finished carousel, not a project file. Burned text posts to any platform, survives a download and re-upload, and cannot drift out of position on a different renderer. If you need the frames without text, ask for them at plan time.
What if only one slide is wrong?
Open the deck and change that slide. The editor treats a slide change as a delta, so you are not paying to re-render four frames you were happy with.
Does the brand chip change the images or just the words?
Both. Palette, accent colour and photographic treatment in the plan came from the kit, not from the typed message — which is why nothing in the six-word prompt mentions amber-gold or cream.
Is the Pinterest route a different product?
It is a different source for the frames, not a different format. You still get a planned, captioned, numbered set; the images are sourced through Pinterest search rather than drawn by an image model, and each frame's Details panel says so.