Your AI image can look sharp at first glance and still fall apart when you zoom in. Dark gradients become grainy, fog breaks into fragments, and tiny lights start looking like debris.
I run into this most often with complex AI images: night scenes, detailed armor, wet reflections, crowds and distant buildings. My cyberpunk mech image had all of them, so it became a useful stress test.
Preserve the main subject and reduce image noise spots
That was the entire instruction I used for the final Nano Banana Pro cleanup pass. It produced the smoothest result in this test while keeping the main mech and overall city composition recognizable. It is not a pixel lock, so I still checked the small details.
Below, you can inspect the original GPT Image 2.5 result, the pass for reducing image noise spots and the final Nano Banana Pro cleanup side by side. I also cropped the exact same areas so the difference is easier to judge.
This test builds on our broader GPT Image and Nano Banana Pro editing comparison. For the product workflow and model selector, see how to use Nano Banana in Ima Studio.
The three-stage test
| Stage | Model | Parameters | Purpose |
|---|---|---|---|
| A | GPT Image 2.5 | 2048×1152, medium | Establish the composition |
| B | GPT Image 2.5 | 2048×1152, medium | Reduce image noise spots with a short instruction |
| C | Nano Banana Pro | 16:9, 2K | Focused image-noise-spot cleanup |
Step 1: Generate a detailed scene that exposes visual artifacts
The Canvas prompts were written in Chinese. The article shows faithful English translations so readers can reuse them without mixed-language instructions.
A giant bipedal mech warrior stands in a rain-soaked cyberpunk metropolis. Positioned slightly right of center, it towers over the scene. Its heavy segmented armor is dark gray and black, with complex mechanical tubing and hydraulic joints. Magenta glowing sensor strips run across its chest and head.
The baseline established one towering mech, a wet city, layered traffic routes and high-contrast dark armor. Its rain, fog and dense reflections created surfaces where grain and fragmented detail were easy to inspect.
Step 2: Add the prompt for reducing image noise spots
High-definition output with minimal image noise spots. Do not render fine details as fragmented debris-like artifacts.
The second pass kept the composition close to the baseline while making the mech surfaces, fog and city reflections more continuous. The short instruction worked as a useful constraint, not as a guarantee of identical details.
Step 3: Use Nano Banana Pro for the focused cleanup
Preserve the main subject and reduce image noise spots
Stage C kept the same main mech and overall city composition while producing the smoothest reviewed gradients and reflections. The matched crops also show local redraws in armor panels, signs and distant structures, so the honest claim is “main subject and composition preserved,” not “identical pixels.”
Matched crops at the same working scale
I normalized all three stages to a 2048×1152 working canvas, then took the same 640×360 crop coordinates. These crops make the cleanup easier to inspect, but they also reveal local redraws that are hidden in the full image.
Mech armor and joints
Rainy sky and haze
City screens and fine lines
Wet street and reflections
The matched crops support a narrower conclusion than the full image alone: Stage B reduces fragmented detail while keeping the scene close to the baseline; Stage C cleans gradients further, but it also redraws small armor, screen and city details. The workflow preserves the main subject and overall composition, not identical pixels.
What changed in this run?
| Check | Stage A | Stage B | Stage C |
|---|---|---|---|
| Mech pose and geometry | Baseline | Preserved | Preserved |
| Image noise spots on dark armor | Dense shadow detail | Clear subject; no obvious debris artifacts | Cleanest reviewed result |
| Sky and haze | Busy | Smoother | Smoothest while retaining rain |
| Screen marks | Present | Present | Reduced; minor shopfront marks remain |
What can go wrong
A smoother image may simply be smaller
Nano Banana Pro used 2K and landed at 2752×1536, so the final image did not rely on a lower-resolution output to look smoother.
“Preserve” is not a pixel-lock guarantee
Both tools regenerate images. The mech pose and city layout survived, but small sign details still changed.
Neon signs can produce fake text
Cyberpunk scenes encourage sign-like patterns. If exact text is required, remove generated marks and typeset the final copy outside the image model.
Run the two-model cleanup in IMA Studio
Every stage of this test ran inside IMA Studio: one account, one credit balance, and one place where the results are kept. Here is how to repeat it, what it costs, and where your images end up.
How to run it
- Open Image in the left sidebar. The button below opens the same page with Nano Banana Pro already selected.
- Check the model. Use Nano Banana Pro for the cleanup pass. To rebuild the baseline scene first, pick GPT Image 2.5.
- Add the image you want to clean in the reference slot (up to 14 images; type
@in the prompt to point at one), then pastePreserve the main subject and reduce image noise spots. - Set ratio and resolution. We used 16:9 · 2K for the Nano Banana Pro pass.
- Read the number on the Generate button, then click it. That number is what the run will cost.
- Collect the result from History on the right of the page, or from Assets in the left sidebar.
What this test costs
Credits are the one currency shared by every model in IMA Studio. The rates below are what IMA Studio listed on 8 October 2026; the Generate button always shows the live number.
| Model and setting | Credits per image |
|---|---|
| Nano Banana Pro · 1K | 12 |
| Nano Banana Pro · 2K (Stage C) | 17 |
| Nano Banana Pro · 4K | 22 |
| GPT Image 2.5 (Sunburst) · 2048×1152 · medium (Stages A and B) | 17 |
| GPT Image 2.5 (Sunburst) · 2048×1152 · high | 29 |
One pass through all three stages costs 51 credits (17 + 17 + 17) at these rates. Every retry or extra variation is a new generation and is charged again.
When you hold more than one type, IMA Studio spends the credits that expire first. If a generation fails because of server load, network trouble or scheduled maintenance, the credits are refunded automatically. Content blocked by moderation is not refunded. New accounts get one-time welcome credits. Source: the IMA Studio pricing FAQ.
Where your results live
- History is the panel on the right of the Image page. Each entry keeps its task type, age, model, resolution and prompt, with Remake, Add to reference, Continue in Canvas and Download underneath.
- Assets in the left sidebar holds everything you generated, grouped by date. It has Generated, Collections and Asset Packs tabs, plus image, video and audio filters and a search box.
Short answer: your results are stored in your account, but IMA Studio does not promise to keep them forever. What we could verify:
- These files are still there. The three test images were generated on 30 September 2026 and were still in Assets on 8 October 2026.
- Storage has a quota. The plans dialog lists cloud storage for assets: 5 GB on Basic, 10 GB on Plus and 20 GB on Pro. The account we captured shows a 1 GB meter at the bottom of Assets.
- No retention period is published. The Terms set no minimum storage time, and they say content may be deleted when an account is terminated.
Practical rule: treat Assets as your working library. Download the finals you cannot afford to regenerate.
Try the two-model workflow
I used GPT Image 2.5 for the creative decision and Nano Banana Pro for the focused cleanup. That division made the comparison easy to inspect: one model built the scene, and the other had one narrow job.
About The Author
