Nine image jobs that used to need Photoshop, Canva Pro or a freelance designer now run inside Gemini's chat window, and AI Central's New Nano Banana Prompting Guide reduces each one to a single sentence. The catch is setup. Two clicks decide whether any of it works, switch Gemini from Fast to Thinking, then open Create Images from the Tools menu. Everything after that is naming three things, the element, the change, and what must not change.
The two clicks that decide everything
AI Central starts at gemini.google.com, before a single image is uploaded. Gemini 3 puts a mode selector at the bottom of the chat bar, Fast answers quickly, Thinking solves complex problems, Pro handles advanced math and code. The first instruction is to move off the default.
Thinking mode solves complex problems instead of just answering quickly. Everything works better from here
The second click is the one AI Central says almost everyone misses. Image generation does not live in the main chat bar. It sits behind the plus icon, under Tools, alongside Canvas, Deep research, Create music and Guided learning.
This is where most people get stuck because they never look here
That is AI Central's diagnosis of why identical prompts work for one person and fail for another. None of what follows fires at all from the wrong mode with the wrong tool selected.
Repair is where the return is highest
The most immediately useful prompts AI Central includes do not create anything, they rescue what already exists. Old photos, low resolution screenshots and pixelated exports each get one line, enhance this image to 2K resolution, or the same sentence with 4K in place of 2K. No sliders, no export dialog, no plugin.
The blur fix is more interesting, because of what it forbids. AI Central's wording asks the model to turn a blurred photo into an ultra-sharp realistic editorial shot without changing the composition or objects. That trailing clause does the real work. A generative model handed a damaged photo will cheerfully redraw it, and the instruction to leave composition and objects alone is what stops a restoration becoming a reinvention.
The worked example runs it on a blurred photograph of Marilyn Monroe, and Gemini reports back that it sharpened her features and hair while preserving the exact original composition. The difference between that and a fabricated face is one clause of instruction.
Changing one thing without redrawing the picture
Three of AI Central's prompts edit rather than generate, and they escalate in scope. The narrowest is a slot template, replace the object at the position with the new object. Naming the position matters more than it looks. It stops the model guessing which of several similar objects you meant, the usual reason a targeted edit comes back having touched half the frame.
Colour is the middle step. The simple form changes the colour of a named element. The ambitious form AI Central demonstrates reads less like an edit and more like a note to a colourist.
Apply a cinematic blue tone across the whole image with a ray of moonlight on the main subject.
Widest of the three is a full restyle. AI Central's example asks for a marketing campaign with an Italian street vibe, and the model answers with a rustic campaign carrying cultural detail and a vintage screen printed look. The guide is explicit that this is not a colour change but a change of the whole aesthetic, which is why the instruction has to describe a world rather than a hue.
Turning a link into an infographic
The prompt with the broadest business use never touches a photograph. Paste a link or upload a document, then use AI Central's line, asking for an infographic of the source in a named ratio. The ratio is not decoration. It stops you receiving a default canvas that fits nowhere you intend to publish.
The example uses a Microsoft piece titled What's Next in AI: 7 Trends to Watch in 2026. Gemini reads it and builds a seven point graphic around the shift from AI as a tool to AI as a partner, taking in agent safeguards, the global health gap, AI as a lab assistant in physics and biology, superfactories rather than sheer data centre size, code that understands a repository's history, and the convergence of AI with quantum computing.
That is two jobs in one request, reading comprehension and visual composition. It is also the strongest argument for AI Central's insistence on Thinking mode, because condensing a seven part argument and then laying it out is exactly what a fast answer does badly.
Pictures of yourself, three ways
AI Central closes with three prompts that take your own photo as the input. One turns you into a custom figurine placed in a situation you name. One drops you or your group into a scene, surrounded by hundreds of objects doing something you specify. The third is worth studying line by line.
Generate a corporate headshot of my photo in the same style as the second image I uploaded. Do not copy the face from the second image.
That single sentence from AI Central carries the whole method. It uses two uploads and gives each a role, it points at the second image as a style reference rather than a subject, and it ends with an explicit prohibition. A model given a reference has no way of knowing which part of it you admired, and the face is the part you least want borrowed.
Why the pattern holds
Read the eleven steps in sequence and one skeleton shows through every prompt AI Central wrote. Each names an element, names a change, and wherever collateral damage is a real risk, names what must stay untouched.
- Name the element, not the picture. Replace the object on the left beats make this image better.
- Name the change concretely. A cinematic blue tone with moonlight on the subject beats moody.
- Name the constraint. Without changing the composition or objects, and do not copy the face, are the clauses that stop an edit becoming a regeneration.
- Name the output. A ratio for infographics, a resolution for upscales.
Vagueness is why people conclude image models cannot edit. What AI Central has assembled is a lesson in specificity wearing eleven costumes.
A note on the count and the limits
Reviewed August 2026. AI Central bills the set as eleven prompts. Counting the numbered steps, two are setup, opening Gemini in Thinking mode and finding Create Images, which leaves nine patterns that do actual work on an image. Two of the illustrated captions in the source read as half finished sentences, so the fill in the blank templates are the versions to trust.
The guide shows single results, not comparisons. Nothing establishes how often a prompt lands first time, and the repair prompts especially depend on how much detail survives in the original file. Treat all nine as starting points, not guarantees.
What to do with this
If you have never generated an image in Gemini, do the two setup clicks and run only the upscale line on a photo you already have. It is the fastest honest test of whether the workflow deserves more of your time.
If you already use image models elsewhere, steal the constrained prompts. The blur fix and the headshot are the two AI Central wrote with prohibitions inside them, and that habit transfers to every other tool you touch.
If you produce content for a living, the article to infographic prompt is the one with a business case. It collapses a research step and a design step into one request, and of everything here it is the most likely to replace a recurring job rather than a one off.
Do I need to pay for anything to run these prompts?
AI Central frames the whole thing as one free tool doing work that used to be paid for, no Photoshop subscription, no Canva Pro, no designer on Fiverr, just Gemini, a browser and the right way to ask. The mode selector offers Pro for advanced math and code, but none of the eleven steps call for it.
Why does my edit change the whole image instead of just one part?
Almost always because the prompt never said where to look or what to leave alone. AI Central's edit template names the object, the position and the replacement, and its restoration prompt ends with an instruction not to change the composition or objects. Add those two ingredients and edits stop spilling into the rest of the frame.
Can Gemini really turn a link into an infographic?
That is what AI Central's worked example shows, built from a Microsoft article on seven AI trends to watch in 2026. You paste a link or upload a document, name the ratio, and the model handles the summarising and the layout together. Feed it a source with clear structure, because the graphic can only be as organised as the argument it reads.
Will it copy the face from the photo I use as a style reference?
Only if you let it. AI Central's headshot prompt uploads two images, points at the second one for style, then instructs the model in plain words not to copy the face from it. That final sentence is the entire safeguard, and it is worth reusing anywhere you hand a model a reference image.