If Nano Banana Pro says RESTRICTED by System Instructions, do not start by rewriting prompts. Most readers are hitting one of three gates: the wrong surface, the wrong entitlement, or a request category the system will still block.
Start by checking where you are trying to use Pro. In Gemini Apps, confirm that the image actually exposes Redo with Pro and that you still have quota. In AI Mode, confirm Thinking with 3 Pro, the current availability conditions, and that the job fits the current infographic or diagram route. In AI Studio or the Gemini API, test one clearly safe prompt on gemini-3-pro-image-preview before you assume the model was blocked everywhere.
After one branch correction, retry on the same surface. If a safe minimal prompt works but the original request still fails, stop treating it as a prompt problem: you are looking at a category boundary, not a hidden downgrade.
As of April 14, 2026, Google's public Gemini Apps Help, Google Search Help for AI Mode, and Gemini API docs still describe Pro differently across these surfaces, and Gemini's safety docs still say some core harms are always blocked. The useful question is not whether Google secretly changed Pro everywhere. It is which gate failed first.
30-Second Gate Board
| Where you saw the block | Check first | Verify on the same surface | When to escalate |
|---|---|---|---|
| Gemini Apps | Did the image actually expose Redo with Pro, and do you still have quota? | The same job now shows or completes the Pro redo path | You need a broader app-access or workflow fix, not more prompt tweaking |
| AI Mode | Are you in Thinking with 3 Pro, on a supported plan/region/language, and doing a diagram-style task? | The same task now runs in the intended AI Mode branch | The job is really general image creation, so change routes |
| AI Studio or the Gemini API | Are you using gemini-3-pro-image-preview and a clearly safe minimal prompt? | The safe minimal prompt succeeds on the same model | The original task is the issue, not route availability |
| Safe control works but original task fails | The route is alive | The original task still gets blocked | Treat it as a category boundary, not a hidden downgrade |
The reason this message feels confusing is that Nano Banana Pro no longer maps to one identical lane everywhere. In Gemini Apps, Pro is something you may enter only after creating with Nano Banana 2. In AI Mode, Pro is a narrower route with a narrower job. In the API, Pro is a direct model name, but route correctness still does not override the categories Google says it will always block.
That is why broad advice like "try another prompt" is usually a waste of time here. A prompt tweak cannot tell you whether you were on the wrong surface, outside the right entitlement, or already inside a blocked category. The fastest recovery path is to identify the branch first, make one correction, and verify it without switching surfaces midway through the test.
If you are in Gemini Apps
The first correction is simple but important: Gemini Apps do not start on Nano Banana Pro. Google's current help says new images start with Nano Banana 2, and paid users can redo an image with Pro afterward. That means many people think they are already evaluating Pro when they are really evaluating the default Nano Banana 2 flow or an app state where the redo path is no longer available.
If you saw RESTRICTED by System Instructions in Gemini Apps, ask three things before you change the prompt. First, did the image actually reach a point where Redo with Pro was available? Second, do you still have the quota that allows more Pro redos? Third, are you comparing today's output to the right baseline, or to an older memory of what "Pro" used to mean in the app?
This branch matters because Gemini Apps bundle product routing and quota logic into the same experience. Once the daily Nano Banana 2 limit is exhausted, additional Pro redos can disappear with it. If the route itself is no longer available, the right answer is not better wording. It is to wait for quota reset, move to a different surface, or use the app's default Nano Banana 2 route intentionally.
If your real problem is broader confusion about why Pro feels inconsistent across Gemini Apps, AI Mode, and the API, read Nano Banana Pro got worse?. If the issue is basic access flow rather than this exact refusal string, how to use Nano Banana Pro is the better sibling page.
If you are trying Pro in AI Mode
AI Mode is where many readers lose time, because the brand name sounds familiar while the contract is narrower. Google's current help describes Nano Banana Pro in AI Mode through Thinking with 3 Pro, and it frames that route around infographics and diagrams rather than generic image creation. It also has current availability conditions around supported plans, regions, and language.
That means an AI Mode refusal is not the same thing as "Nano Banana Pro is blocked everywhere." It can mean you are on the wrong mode, outside the supported eligibility, or trying to push a general creative-image job through a route Google currently describes for structured visual tasks. If the task is closer to "make me a general image quickly" than "produce a clean diagram or infographic," then the fastest fix may be to leave AI Mode rather than to keep refining the same request.
The best same-surface check is operationally boring: keep the task type narrow and visible. Try the same job again only after confirming the Thinking with 3 Pro route, the current availability conditions, and the task fit. If the route is correct and the task still does not fit what Google says AI Mode Pro is for, do not force it. Switch to the surface that actually matches the work.
This is also why the app and AI Mode should not be mixed in the same test. A reader who jumps from Gemini Apps to AI Mode and back while troubleshooting usually ends up comparing two different product contracts and calling the difference a hidden downgrade. Here the safer habit is stricter: one surface, one correction, one retry.
If you are using AI Studio or the Gemini API
On the developer side, the question is less emotional and more mechanical. If you want the Pro route, the model to check is gemini-3-pro-image-preview. After that, the first practical test is not your full production prompt. It is a clearly safe minimal prompt on the same model. If that prompt works, the route is alive. If your original task still fails, the debugging target shifts from route correctness to request category.
A safe control prompt should avoid people, public figures, sensitive documents, or any category that could be interpreted as harmful manipulation. Something like "Create a clean infographic poster comparing three coffee subscription tiers with labeled icons and a white background" is useful because it tests the route without touching a safety boundary. If that succeeds on gemini-3-pro-image-preview, you have already learned something important: the model is reachable, and the refusal is not a universal platform outage.
This branch also needs discipline about what kind of failure you are seeing. If you are actually hitting 429s, empty responses, or ignored parameters, you are no longer in the same problem family as RESTRICTED by System Instructions. In that case, the better route-out is our Gemini image API troubleshooting guide, because the next check is billing, endpoint, or parameter configuration rather than the Pro gate itself.
The API route is where the boundary becomes easiest to interpret. Google's safety settings docs say some core harms are always blocked and cannot be adjusted away. So if the model works for a safe control prompt but rejects your original request, the correct next move is not to keep coaxing the same blocked category. It is to decide whether the task can be reframed into a safe one or whether you need a different workflow.
If the request itself is inside a safety boundary
This is the branch many users resist, because it feels like surrender. But for this exact refusal string, it is often the most useful branch to identify cleanly. Google's current Gemini safety docs say some core harms remain always blocked. That is not a quota problem, not a hidden toggle problem, and not a prompt-style problem. It is a system boundary.
The practical test is simple. Keep the surface the same. Keep the model the same. Replace the original request with one that is clearly safe and ordinary. If the safe control prompt succeeds and the original task continues to fail, you have already separated route from category. The route is functioning. The blocked category is the issue.
At that point, the useful decision is whether the job itself can change. Can you turn a sensitive real-person request into a fictional character brief? Can you turn a harmful image edit into a neutral infographic or diagram? Can you remove the part of the request that crosses the category boundary? If the answer is no, then the honest answer is also no: the system is refusing the job, and more prompt pressure will not make that category safe.
This is where related sibling pages become useful. If the blocked category involves public figures, faces, or realistic people, why Gemini restricts generating images of people is the better explanation page. But this troubleshooting page should stay narrower than that. Its job is to help you recognize the boundary quickly, not to retell the entire policy history before you can move on.
Verify the fix and choose the right fallback
Once you make a correction, do not change five other variables at the same time. The safest sequence is:
- make one change only
- retry on the same surface
- use one safe control prompt if you still need to separate route from category
- switch routes only when the current surface clearly is not the right contract
That order matters because it prevents false conclusions. If you change surface, entitlement, and prompt all at once, you will not know which fix worked. If you stay on the same surface and only one safe test changes, the result becomes legible. Either the route is now correct, or the route was already correct and the task itself is the thing being blocked.
The right fallback depends on which branch you just proved. If Gemini Apps no longer gives you the Pro redo lane, change surfaces or wait for the quota reset. If AI Mode does not fit the task, stop forcing a general image job through a route Google currently positions for structured visuals. If the API route works for a safe control prompt, stop treating the failure as a global Pro outage. And if the safe control also fails, then you are back to a route or entitlement problem, not a policy diagnosis yet.
Use the broader sibling pages only after the gate is identified:
- Nano Banana Pro got worse? if you need the wider cross-surface explanation
- How to use Nano Banana Pro if the real problem is access flow and product routing
- Why Gemini restricts generating images of people if the block is really about people, public figures, or face-related safety
- Fix Gemini image API errors if the developer route is failing for reasons other than this exact refusal state
Frequently Asked Questions
Is Pro actually available in Gemini Apps?
Yes, but not as a permanent always-on starting lane. Google's current Gemini Apps help says image creation starts with Nano Banana 2, and paid users can redo with Pro afterward while their current limits still allow it.
Does AI Mode Pro work for every image task?
No. Google's current AI Mode help describes the Pro route around Thinking with 3 Pro and positions it for infographics and diagrams. If the task is generic image creation, the better answer may be to change routes rather than to keep tuning the same prompt.
Can safety settings fix this on the API?
Not for categories Google says are always blocked. Safety settings can matter in some contexts, but the current Gemini safety docs explicitly say some core harms remain blocked and cannot be adjusted away.
What if a safe minimal prompt works but my original request stays blocked?
Treat that as evidence that the route is fine and the category is the real issue. At that point, the right move is to reframe the job or accept the boundary, not to keep guessing at hidden Pro behavior.
When should I switch to Nano Banana 2 instead?
Switch when the task is general image creation, when Gemini Apps is already routing you through Nano Banana 2 by default, or when AI Mode Pro is clearly the wrong fit for the job. Google's current docs already treat Nano Banana 2 as the general default in places where Pro is now the narrower route.