AI Tools for Presentations: Generate the Atmosphere, Not the Evidence

Somebody in the third row has seen your stock photo before. Maybe on a competitor's homepage, maybe in the last deck they sat through this quarter. That is the small failure. The large one is the chart on slide nineteen that nobody in the room can source.

AI fixes the first problem cleanly and makes the second one considerably worse. We run the image models people use for this work — Nano Banana 2 Lite, Nano Banana 2 and Nano Banana Pro — which gives us a direct interest in you pointing them at the right half of the deck. Generate the atmosphere. Never generate the evidence.

The two halves of a slide deck

A deck is an argument and a surface at once. The argument is your numbers, your claims, the diagram of how the system actually works. The surface is everything the eye lands on before it reads a single word: the section divider, the texture behind the title, the illustration language that makes forty slides feel like one document rather than forty.

Image models are excellent at the surface. They are actively dangerous on the argument. Everything below is that sentence, in detail.

Where a generated image earns its place

Six jobs where we would reach for a model before a stock library:

• Section dividers. Six sections need six images that belong to each other and to nobody else. Highest-value use on the list, because it is where stock photography visibly fails.
• Backgrounds, textures and gradients. No subject, no focal point. Cheap to generate, trivial to reroll, and the model never has to be right about anything.
• The title-slide hero. Looked at longest, and often the only image that also has to survive being printed.
• A consistent illustration language. Forty slides drawn as though one illustrator did all of them.
• The localised deck. You built it in English, Madrid needs it in Spanish, and the artwork should survive the translation.
• The leave-behind. Cover art for the PDF that goes out afterwards, at a resolution that holds up in print.

None of that is content. All of it is the container content sits in — exactly where a model that invents things is safe.

Where it will embarrass you

Charts and data visualisation. A generated chart is invented data wearing the visual grammar of real data: gridlines, axis ticks, a trend that looks deliberate. It is the most authoritative-looking object on the slide and every number in it is fiction. That is the worst available failure in a business presentation, because the room believes it until somebody checks. Build charts in whatever holds the numbers.

Diagrams where the labels have to be correct. Nano Banana Pro renders legible text. That is not the same as knowing your architecture. It produces boxes and arrows with plausible words in them, arranged in a way that is plausibly wrong, and the person who owns that system spots it in four seconds. Draw it in a diagramming tool.

Anything geographic. Generated maps get coastlines, borders and city positions subtly wrong — hard to spot, impossible to defend once spotted. We wrote up why AI image generators fail at maps.

Your own product. A generated approximation of your UI is a lie about a thing the audience can go and look at. Take a screenshot.

The title slide is where the text question gets settled

Most slide images should contain no words at all; the words belong in the slide, in a real typeface, editable. But a title slide often wants one word baked into the artwork, and localisation wants the whole thing rebuilt in another language. That is the one place in a deck workflow where paying more is correct.

Nano Banana Pro is gemini-3-pro-image, and legible, styled, multilingual in-image text is the reason it exists. $0.075 a request on our API as of our 26 August 2026 price check. It also charges the same for 1K and 2K output, which makes 2K free money on that tier — take it, especially for anything destined for print.

Nano Banana 2 renders in-image text too. It is gemini-3.1-flash-image, the capability is there, and for a single short word it usually lands. We still would not trust it with the headline everyone stares at for ninety seconds, and we would not attempt a language swap on it at all. That is Pro's job, through the Pro edit endpoint: hand it the English title slide, ask for the Spanish one, the artwork survives while the words change.

The original Nano Banana cannot do this at all — 2.5-generation model, in-image text arrived with Gemini 3 — and it retires on 2 October 2026 regardless. Do not start a deck workflow on it.

One tier note unrelated to text: if you need a cut-out shape sitting over a slide background rather than a rectangle underneath it, GPT Image 2 is the only model we carry with a background: transparent parameter. $0.0525 at 1K.

Slides are 16:9 and the default is not

This is the detail that costs people their first afternoon. Our aspect_ratio default on the Google tiers — Nano Banana 2, Nano Banana Pro and the legacy model — is 9:16. Portrait. Ask for a slide image without pinning the ratio and you get a tall picture for a wide surface, crop it, and the composition you liked is gone.

Nano Banana 2 Lite defaults differently again, to 1:1. Also not a slide.

Pin it every time:

curl -X POST https://api.e2x.ai/v1/jobs/submit \
  -H "Authorization: Bearer $E2X_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/nano-banana-2-lite/text-to-image",
    "input": {
      "prompt": "Wide abstract texture of overlapping translucent paper layers in cool grey, soft raking light, no text, no logos",
      "aspect_ratio": "16:9"
    }
  }'

Two more ratios worth knowing. 21:9 exists on every Google tier and suits a full-bleed banner across the top of a content slide. Lite alone carries 4:1, 1:4, 8:1 and 1:8 — absurd for a photograph, perfect for a thin decorative rail down the side of a template.

Forty slides that look like one deck

The failure here is not quality, it is drift. Forty independent prompts produce forty images that are each fine and collectively look like a mood board assembled by committee. Slide four is painterly, slide nine went photographic, slide twenty-two picked up a colour nobody chose.

The fix is one style reference reused for every image in the deck. Generate a single image you are happy with, then feed that image back as a reference on every subsequent request instead of trying to describe the style again in words. Nano Banana 2 accepts up to fourteen reference images. Nano Banana Pro splits its fourteen by role, including up to three dedicated style references, which is the mechanism that makes this reliable rather than lucky.

We wrote the method up properly in keeping Nano Banana output consistent across a set. If your problem is that individual images are not coming out right in the first place, that is a different fix and it is in the Nano Banana prompting guide.

What the imagery for a real deck costs

Take a forty-slide deck with an image on every section divider plus a title hero. Call it twelve images.

On Nano Banana 2 Lite at $0.0238, that is $0.29. On Nano Banana Pro at $0.075, it is $0.90. Reroll everything three times, because the first pass never is, and you are still under three dollars. A single stock photo licence costs more than that, and it comes with the property that other people are also using it.

Prices and product terms can change. Check each provider's current pricing before making a purchasing decision. Google Batch or Flex pricing is not directly equivalent to a standard on-demand API request because scheduling, availability, and processing conditions differ; it is therefore excluded from this comparison. This is a scoped comparison, not a claim that E2X is the world's cheapest option in every configuration.

The tier we would actually put you on

Not Pro. Not for the whole deck.

Pro for the title slide and for any image that carries a word or needs a language swap. Lite for every divider, texture and background, because those images contain no text, no product, no claim, and nothing that a cheaper model gets meaningfully worse.

The resolution argument goes the same way. A conference-room projector runs 1920×1080, so 4K on a projected slide is pixels you pay for and discard — Nano Banana 2 at 4K is $0.08 against $0.04 at 1K. Go to 4K for the printed programme. Not for the projection.

One clause to read before any of this becomes a pipeline. SynthID rides along inside every Google image output — Google's own provenance marking, invisible in the picture, and not exposed as a parameter to us or to anybody else reselling that API. Agency and enterprise deliverable terms sometimes prohibit handing over watermarked assets. Ask in week one of the engagement, not the week the deck ships.

The tier-by-tier breakdown is a separate piece, for anyone wanting the general answer rather than our deck-specific opinion: the Nano Banana model comparison. To pick by job instead of by name, start from text-to-image; the prices and defaults quoted above sit on the individual model pages in our catalog.

Frequently asked questions

What AI tools do you use for professional presentations?

For the visual surface of a deck — section dividers, backgrounds, textures and the title-slide hero — an image generation API is the right tool, and we use Nano Banana 2 Lite at $0.0238 a request for most of it, with Nano Banana Pro at $0.075 for anything carrying readable text. For the content of the deck — charts, architecture diagrams, maps, product screenshots — use the tool that actually holds the data or the real thing, because a generated version is invented information that looks authoritative.

Can AI generate the charts in my presentation?

It can produce something that looks exactly like a chart, and every value in it will be fabricated. That is the most dangerous object you can put on a business slide, because the audience will read it as evidence. Build charts in a tool connected to your real numbers and reserve image generation for the parts of the slide that carry no claims.

Which AI model renders readable text on a title slide?

Nano Banana Pro, gemini-3-pro-image, is the tier built for legible, styled, multilingual in-image text, at $0.075 per request on our API as of 26 August 2026. Nano Banana 2 renders in-image text as well and handles a single short word acceptably. The original Nano Banana cannot do it at all, because that capability arrived with the Gemini 3 generation.

What aspect ratio should I use for AI slide images?

Set aspect_ratio to 16:9 explicitly on every request. Our default on Nano Banana 2, Nano Banana Pro and the legacy model is 9:16 — portrait — and Nano Banana 2 Lite defaults to 1:1, so an unpinned request gives you the wrong shape for a slide every time. Use 21:9 for full-bleed banner strips.

How much does it cost to generate images for a whole deck?

A forty-slide deck with an image on each section divider and a title hero is roughly twelve images. At Nano Banana 2 Lite's $0.0238 that is about $0.29; at Nano Banana Pro's $0.075 it is $0.90. Even allowing three rerolls per image, the whole deck costs less than a single stock photo licence.

How do I keep AI images consistent across a whole presentation?

Generate one image you are satisfied with, then pass that image as a style reference on every subsequent request rather than re-describing the look in words. Nano Banana 2 accepts up to fourteen reference images and Nano Banana Pro reserves up to three of its fourteen slots specifically for style references. Describing a style in prose forty separate times produces forty slightly different styles.