Google’s Gemini video maker should be one of the most useful creative tools on the market, promising to turn a simple prompt into a short video with motion, sound and production polish.

In practice, too many outputs arrive with obvious mistakes, forcing users to spend more credits fixing problems the tool should not have created in the first place.

It can feel like paying to supervise a machine that keeps making the same basic mistake.

The most frustrating problem is text.

Ask Gemini to produce a scene with a sign, product label, news graphic, shopfront or on-screen words, and the result can quickly collapse into warped lettering, nonsense fragments and broken words that look like they were written by someone dreaming in another alphabet.

That might be funny once. It is less funny when each attempt eats into a limited credit balance.

Google’s own documentation makes clear that its video tools are tied to paid plans and AI credits. Google says AI plans provide access to video generation in Gemini and Google Flow, while Flow credit costs are charged per generation, not necessarily per request.

Some requests can create multiple generations, meaning users can burn through credits faster than expected. Veo 3.1 Lite, Fast and Quality generations cost different amounts, with Veo 3.1 Quality listed at 100 credits per generation and Gemini Omni Flash edits listed at 40 credits

That is where the consumer problem begins. If a generation technically completes, but the words are unusable, the user is left with the bill.

Google says that if an AI tool fails, credits should not be affected, although they may take time to reappear.

But a bad video is not the same as a failed video. A clip full of mangled text may still count as a successful generation in the system, even if it is useless to the person who paid for it.

This is not just one user being impatient with new technology. Research on text-to-video systems has found that these models still struggle to generate legible and coherent text, including short words and phrases.

That matters because text is not a decorative extra in many real-world videos. It is often the whole point: a headline, a brand name, a call-to-action, a price, a warning sign, or a location marke

The issue is not that AI video is imperfect. Everyone understands the technology is still developing. The issue is the pricing model.

When a tool repeatedly produces unusable output and then charges users again to fix it, the product starts to feel less like a creative assistant and more like a poker machine with better branding.

Independent testing has also raised similar concerns. The Verge described Google’s newer Omni video model as a “mixed bag”, noting that some results were strong while others produced strange errors.

The review also pointed out the cost of repeated edits, with one round of edits costing credits and a user on a paid plan burning through most of a monthly allowance after roughly 20 clips and a few edits

That is the part Google needs to fix.

A fairer system would treat obvious text corruption as a quality failure, not a completed job. If a user asks for a sign that says “Tech Business News” and Gemini returns something closer to “Teech Bzsnuss Nuws”, that should not cost the same as a usable video.

At the very least many say Google should offer automatic low-cost regeneration for broken text, clearer warnings before text-heavy prompts are generated, or a separate text-rendering layer that lets users edit words after the video is created.

Because right now, Gemini’s video maker can be visually impressive and commercially irritating at the same time.

It can produce cinematic movement, believable lighting and decent sound, then ruin the entire clip with one mangled word in the background.

For casual users, that may be tolerable. For publishers, advertisers, small businesses and creators trying to produce clean work, it is a serious flaw.

Google does not need to pretend AI video is perfect. But if it is going to charge people every time they ask the machine to try again, it should not make them pay full price for fixing mistakes the model should never have made in the first place.