Why Most Crypto Banners Fail (and How AI Creative Changes the Workflow)

Imagine four banners landing in a network's approval queue on the same afternoon, all from the same advertiser, a hypothetical staking dashboard. The 728×90 is a wall of eight-point text listing supported chains. The 300×250 is the company logo on a gradient, nothing else. The 320×50 is the 728×90 scaled down until the text is unreadable. The 480×320 has three buttons: "Learn more", "Start now" and "Join Discord".

All four will be approved. All four will serve. All four will underperform, and the advertiser will conclude that banners do not work in crypto. They do; these ones were never going to.

Having looked at a lot of crypto display creative, I think the failures cluster into a handful of mistakes, each of which is cheap to fix. The second half of this piece is about how in-network AI tooling changes the economics of fixing them.

The five mistakes

1. Treating the banner as a brochure

The leaderboard with eight lines of text is the most common failure and the most understandable. The team has a lot to say and the banner is the only space they bought. But a banner is seen for a fraction of a second in peripheral vision while someone reads something else. It can carry one idea. Eight chains, four features and a fee comparison are seven ideas too many.

2. No single call to action

A banner with no CTA leaves the viewer to guess what clicking will do. A banner with three CTAs splits an already tiny amount of attention three ways. One verb, one button, one destination.

3. The logo-only banner

Brand-only creative works for companies whose name already means something to the viewer. For almost every crypto project, it does not. A logo on a gradient communicates "a company exists". The viewer has no reason to find out which one.

4. Shrinking desktop creative for mobile

A 728×90 scaled to 320×50 is unreadable at the pixel level, and on a phone it is also competing with a thumb. Mobile needs its own composition, not a smaller copy of the desktop one.

5. Ignoring the page it will sit on

Crypto sites skew dark. A dark banner on a dark page disappears; a banner that mimics the site's chrome looks like navigation and gets ignored, or gets clicked by accident, which is worse because it costs money on CPC and converts nobody. Good creative is distinct from its surroundings without shouting.

What good looks like, by format

Affizy serves five formats. Four of them cover the vast majority of placements, and each has its own logic.

728×90 leaderboard

A horizontal reading line. Logo mark on the left, one short headline in the middle, a CTA button on the right. Headline of about five to seven words, set large enough to read at arm's length. No body copy. The visual, if any, is a cropped product shot or an abstract that supports the colour scheme, not a scene.

300×250 medium rectangle

The most versatile size and the one most often placed inside article content, which means it gets the most genuine attention. Think in three stacked zones: a visual that occupies the top half, a headline of one or two lines, a CTA button with the logo mark nearby. This is the one format where a sub-line of supporting text can earn its place.

320×50 mobile banner

The hardest format in advertising. There is room for a logo mark (not a wordmark), a phrase of three or four words and a CTA, and nothing else. The CTA needs to be a comfortable thumb target. If you cannot say it in four words, say less.

480×320 spotlight rectangle

The largest of the set and closer to a small landing panel than a classic banner. Here a real visual can breathe: a product screenshot, a stylised illustration, a scene. Headline, one supporting line, one CTA. The temptation is to fill the extra space with more text; the better use is more negative space around a stronger image.

The 160×600 skyscraper also exists for sidebar placements and follows the leaderboard's rules rotated vertically: logo at top, headline in the middle, CTA at the bottom.

Across all formats

Static images and HTML5 animated banners both serve on Affizy. If you animate, resolve to a static end frame that works on its own, because many viewers only see the final state. Keep the motion gentle and the loop finite.

How AI creative changes the loop

The reason the four bad banners at the top exist is not that the advertiser does not know better. It is that the loop for making better ones is slow and expensive. Brief a designer, wait, review, request changes, wait, export five sizes, discover the mobile version does not work, go round again. By the time a decent set exists the campaign has been live for a week on the first draft.

In-network AI creative tools shorten the loop at three points. The example here is Studio Affizy, which lives inside the advertiser dashboard on the platform and is paid in the $AFZY utility token rather than cash.

Brand auto-fill from a URL

Paste the project's URL and the tool reads the site: logo, colour palette, tagline. The designer's first twenty minutes, collecting assets and matching colours, disappear. More importantly, the output starts on-brand rather than being corrected towards brand later.

Prompt-based generation

Describe the visual you want and the tool generates it using Flux Pro image models, sized to the format. For an advertiser, the practical effect is that producing a visual for the 480×320 and a different, mobile-appropriate composition for the 320×50 is a matter of two prompts rather than two design tickets. An AI Creative banner costs 15 AFZY; a template-based banner where you supply the assets costs 1 AFZY.

Remix

Take an existing creative and produce a variant: different visual, same brand frame, or the same concept in another format. Remix costs 5 AFZY. This is the feature that changes behaviour, because it makes A/B testing creative cheap enough to actually do. When a second headline costs a few minutes, advertisers test headlines.

For campaigns that need motion, the same studio has a four-step wizard that produces a 70-second branded promo video in 1920×1080 MP4, priced at 750 AFZY.

Where the AI stops

It would be easy to oversell this, so here is what the tools do not do.

AI images should not carry text. Image generation models, Flux Pro included, are unreliable at rendering legible, correctly spelled words. The right workflow keeps all copy, the headline, the sub-line and the CTA, in an overlay layer on top of the generated image, where it is real text that can be edited and read. Studio is built around this split; if you prompt an image model to include your slogan, you are asking for trouble.

AI does not fix the offer. A banner for a product nobody wants will fail beautifully. The creative loop being faster means you find that out sooner, which is valuable, but it is a diagnosis rather than a cure.

Brand consistency needs a human. Auto-fill gets colours and logo right. Whether the generated visual fits the brand's tone, serious or playful, institutional or degen, is still a judgement call.

Animated banners are not yet AI-generated. Studio produces static creatives today. AI-generated HTML5 animated banners are on the roadmap, and uploaded HTML5 banners from your own designer already work.

A workflow that avoids the four bad banners

1. Write one sentence that states the single thing the banner must say. If you cannot, the problem is upstream.
2. Paste your URL into the creative tool and let it pull logo, colours and tagline.
3. Generate a visual for the 480×320 or 300×250 first, with no text in the prompt. Judge it as an image.
4. Add headline and CTA in the overlay. Five to seven words for the headline, one verb for the button.
5. Remix into the other sizes, composing fresh for 320×50 rather than scaling.
6. Launch two headline variants, check which clicks and which converts, and keep the one that converts.

The whole sequence takes an afternoon rather than a week, and the advertiser ends up with creative that respects the format, the page and the viewer's attention.

Crypto banners do not fail because the medium is broken. They fail because making good ones used to be expensive, so most people shipped the first draft. That constraint has mostly gone. Studio Affizy is one way to see the new loop in practice; the wider product is described at https://affizy.com.