Phase 4 · Gemini · Level 3 · Power User
Generating images and video: an honest look
By the end, you'll be able to…
- Describe what Gemini's image and video generation can and can't reliably do
- Judge the handful of workplace tasks where generated media helps
- Apply the rights, accuracy and disclosure cautions before using a generated image at work
Why it matters
Gemini doesn't only write. It can generate images and, on some plans, video from a text description. It's impressive and improving fast, and it's also the feature most likely to be oversold for office work. This lesson is a straight look: what it actually does well, the narrow set of workplace jobs where it earns its place, and the rights, accuracy and honesty questions you must settle before a generated image goes anywhere near real work.
What it does, and where it's headed
Gemini can create images from a text description, and can generate short video clips too, though video generation for a personal account usually needs a paid Google AI plan and is restricted by age. You describe what you want (a scene, a style, a mock-up) and it produces it; you then refine in conversation, asking for changes the way you'd iterate on any draft. Google's generate images and generate videos help pages carry the current, moving detail on models, limits and availability, so treat exact capabilities as a snapshot.
The trajectory is steep. Quality, control and the ability to edit within a conversation are improving release by release. That's the honest reason to be aware of this feature even if you rarely use it: it's becoming a normal part of the toolset. It's also the reason to be careful, because "impressive demo" and "reliable for work" are not the same thing, and this is the feature where that gap is widest.
The honest workplace relevance
Here's the part the marketing skips: for most office work, image and video generation is a minor tool. It does not help you tidy the Fernway meeting notes, answer a policy question, or clean the sales sheet, the jobs this course spends most of its time on. Being clear-eyed about that is itself a Power User skill; reaching for the flashiest feature when a boring one fits is a common way to waste time.
Where it does earn its place is a narrow, real set:
- Concept and mock-up visuals: a rough illustration to show an idea on an internal slide, a placeholder image while a design is commissioned, a quick "something like this" to align a discussion.
- Draft social or marketing imagery: for teams whose actual job is producing visuals, a fast first draft to react to and refine, not a finished asset dropped in unchecked.
- Storyboarding and internal explainers: a sequence of frames or a short clip to sketch how something might look before anyone commissions the real thing.
Notice the common thread: internal, provisional, clearly-a-draft. The moment an image is customer-facing, brand-critical, or presented as real, the bar jumps, and the caution below is what clears it. For Fernway, that might mean Tom generating a rough concept image for an internal pitch about the feedback process; it does not mean generating the company's actual marketing without a designer and a rights check.
Create a simple, friendly illustration for an internal slide about improving our customer-feedback process: a stylised flat-vector image of a support agent responding to a happy customer, calm blue-and-white palette, plenty of clear space for a title. This is an internal placeholder, so keep it clean and generic. No text in the image.
Why this works: A specific brief (subject, style, mood, and the intended use) gives the generator a target it can hit, and naming it as an internal placeholder sets the right expectations for what you'll accept back: a rough draft to react to, not a finished asset.
The three questions to settle before it's "for work"
A generated image is fine to play with; using one at work asks three questions first.
Rights and ownership. Who's allowed to use this, and for what? The rules on commercial use of AI-generated media, and on images that resemble real people, brands or copyrighted characters, are unsettled and vary by plan and region. For anything customer-facing or commercial, that's a question for whoever owns brand and legal at your organisation, not an assumption you make because the button was there.
Accuracy, the over-trust trap. A generated image is not a source of truth. Ask it for an infographic and the numbers and labels may be invented; ask for a diagram of a process and the arrows may point the wrong way; text rendered inside images is often garbled. This is a hallucination in visual form, and it's more dangerous than the written kind because a chart looks like data. Never let a generated image be the authority for a fact, a figure, or how a real process works. Build those from real data and label them yourself.
Design a clean, simple layout for an internal one-slide graphic showing a four-step feedback process: receive, acknowledge, action, close. Use placeholder icons and leave the step labels as clearly-marked blanks for me to fill in. Do not invent any statistics, percentages or figures.
Why this works: Explicitly telling the generator to leave the real numbers to you sidesteps its worst failure (inventing plausible-looking data) and keeps you as the source of truth for anything factual, using the generated image only for the layout and feel.
Disclosure and provenance. Passing off an AI-generated image as a real photo (of a person, a place, an event) is a trust and, increasingly, a compliance problem. Google attaches provenance signals (such as an invisible watermark) to media its tools generate, and detection is improving, so assume generated media can be identified as such. The safe workplace habit is to disclose it where it matters and never present a generated image as a genuine photograph of something that happened.
Good start. Three changes: make the palette warmer, remove the text you added along the bottom, and give the figures more space around them so there's room for a title. Keep the flat-vector style.
Why this works: Treating the first image as a draft and giving specific, concrete change requests (the same iterate-don't-restart habit as text) steers the result far faster than a brand-new prompt, and keeps whatever already worked.
Try it now
Common mistakes
- Trusting a generated chart or diagram as data. The numbers, labels and arrows can be invented while looking authoritative. A visual hallucination is still a hallucination. Build anything factual from real data yourself and use generation only for layout and feel.
- Skipping the rights question on commercial use. The button being available doesn't mean the image is cleared for customer-facing or commercial use. Rules on ownership and on resembling real people or brands are unsettled. Route those to whoever owns brand and legal.
- Passing generated media off as real. Presenting an AI image as a genuine photograph of a real person, place or event is a trust problem, and provenance signals mean it can increasingly be detected. Disclose where it matters.
- Reaching for it when a boring tool fits. Most office tasks don't need generated media at all. Using the flashiest feature where a text prompt or a spreadsheet would do is a way to waste time, not save it.
- Accepting the first image as finished. Like text, generation is a draft. One round of specific, concrete change requests gets you far closer than firing off a fresh prompt and hoping.
Keeping current
Image and video generation is one of the fastest-moving parts of Gemini. Models, editing controls, plan requirements, age limits and provenance features all change frequently. Trust your screen over this lesson and check Google's official pages: Generate videos with Gemini Apps, the video generation overview and the Gemini Apps release notes. Accurate as of 13 July 2026.