Google's newly released AI image generation model, Google Gemini 2.5 Flash Image (Nano Banana), relying on consistency and lightning-fast image generation speed, continues to dominate major platforms.
I also receive daily requests from group members to update: Sister K, hurry up and introduce Nano Banana gameplay, this is too awesome.
In this article, let's talk about Nano Banana together. The content mainly includes 3 parts: a technical interpretation by the Nano Banana core team, Nano Banana's gameplay, and 5 ways to use Nano Banana for free.
Everyone is welcome to add and exchange ideas in the comments section~
01. Nano Banana Technology Revealed
In a recent interview on Release Notes, the 4 core team members of Nano Banana were invited, helping us better understand the technology behind Nano Banana's key features:
Native Image Generation
Nano Banana's core breakthrough is native image generation, which can continuously reference context during the generation process and complete complex tasks step by step.
Unlike traditional text-to-image tools, Imagen is more like a single-point expert, while Nano Banana is an all-around player that can span modalities and support complex interactions.
This will also be the team's development direction: future models not only need to draw beautifully, but also need stronger understanding and reasoning abilities.
Consistency of Characters and Scenes
Consistency is another highlight of Nano Banana. In the past, models often failed when editing images. For example, I only wanted to change the curtains, but the bed and sofa changed too, or when changing the angle of a character image, the face changed.
This time the team demonstrated a case on site, using a close-up of the host's face to generate a full-body image of him wearing a giant banana suit.
prompt：zoom out and show him wearing a giant banana costume.keep his face visble.
Prompt: zoom out the lens, show him wearing a giant banana costume, and make sure the face is visible.
Nano Banana maintained consistency very well. The face was still that face, but the scene and clothing were completely changed, and the picture was still very natural. The entire generation process took just over ten seconds.
Another detail is thatGemini the collaboration between the team and the Imagen team makes the images more natural. Previous results sometimes looked like they were "pasted on," but now overall coherence can be achieved.
Text Rendering
Many people may think that "writing a few words in an image" is no big deal, but Nano Banana treats text rendering as a long-term core metric. They believe: text is structured content. If the model can learn how to handle text, it can also master more complex structures such as textures in images.
At present, Nano Banana performs very well on some simple text rendering, but there are also some shortcomings.
prompt：now write”Gemini nano”on the image.
Prompt: write "Gemini nano" on the image
A few years ago, almost no model could handle text well, and even very short prompts often failed. So the Nano Banana team decided to track this metric over the long term. No matter what experiment they did, as long as they kept observing, they could avoid performance regression. They found that many seemingly unrelated changes can also produce improvements in text rendering.
The team actually started from "finding the model's weaknesses" and gradually explored a path that can drive overall quality improvement.
Understanding and Creativity
Nano Banana has world knowledge, can understand vague instructions, and can also exercise some creativity.
The team mentioned a very interesting concept in the interview—reporting biases. For example, when you visit a friend's home and chat with others afterward, you will almost never mention the ordinary sofa in their home. But if you show someone a photo, the image contains the sofa.
So, if we want to truly understand the world, text may require more description, while visual signals are like a shortcut to understanding the world, directly presenting the environment, objects, and relationships without additional explanation.
Understanding and generation complement each other in this way. In the process of reading images and language, the model accumulates more solid world knowledge, and therefore can create more stably and naturally. Sometimes it can even generate content beyond the user's expectations
02. The Complete Guide to Nano Banana Gameplay
From popular gameplay, single-image, multi-image, and various ways to generate videos. Feel free to add more~
Popular Gameplay
1. Turn any image into a figurine
prompt：turn this photo into a character figure. Behind it, place a box with the character’s image printed on it, and a computer showing the Blender modeling process on its screen. In front of the box, add a round plastic base with the character figure standing on.
Turn this photo into a character image. Place a box behind it, with an image of the character printed on the box, and on the computer screen above the box displayBlenderthe modeling process. Add a circular plastic base in front of the box, with the character image standing on it.
2. Draw a real scene based on a map
prompt：draw what the red arrow sees.
Prompt: Draw what the red arrow sees.
Draw a DEM with contour lines.
draw the real world view from the red circle in the direction of the arrow.
Draw with contour linesdigital elevation model。
Draw the real-world view from the red circle along the direction of the arrow.
The above case comes from X blogger @Simon
3. Cartoon to reality
prompt：Depict as a live big budget costume test on set, shot on film.
Variant Prompt: For easier additional editing.Depict as a live big budget costume test on set, shot on film against green screen.
Depict it as a big-budget costume fitting on a film set, shot on film.
Variation prompt: For easier additional editing, depict it as an on-site big-budget costume test, shot on set, using film against a green screen.
The above case comes from X blogger @Brent Lynch
4. 360-degree product display
prompt：This exact car in this exact environment.
Change Perspective: Perfect side angle view.
This car and its exact environment.
Change the perspective: a perfect side-angle view
After generating images from different perspectives, use Kling 2.1 to generate a video through the first and last frames.
The above case comes from X blogger @Rory Flynn
5. Restore old photos
prompt：Restore and colorize the picture without altering, removing, or adding any detail or element.
Prompt: Restore the image coloring, but do not change, delete, or add any details or elements.
The above case comes from X blogger @Rodrigo Bressane
6. Isometric 3D view
Turn the design drawing into a 3D view.
The above case comes from X blogger @levelsio
From 2D drawings to 3D models, it looks quite stunning. However, current feedback shows that the generated images are not precise enough, for example, they may be blurry and the window positions may be inaccurate.
7. Switch perspectives
prompt：aerial perspective of a camera behind the blurry ceiling fan looking down at the girl sitting in an hospital waiting room.
An aerial perspective of a camera behind a blurry ceiling fan looking down from above at a girl sitting in a hospital waiting room.
Single-image editing
1. Change text
Prompt: Change the text in the image to: Why don't you ask the god gemini?
2. Change clothing
Prompt: Change the person's clothes into a down jacket.
3. Generate a scene with reference to a person
Prompt: Have the person in the image have dinner with SpongeBob SquarePants.
4. Remove
Prompt: Remove the person on the left side of the image.
5. Change elements
Prompt: Replace the old Twitter logo in the image background with the current X-shaped logo.
6. Understanding vague instructions
Prompt: Make the person in the image look like a Native American.
7. Change the background
In the first step, replace the background with a green screen; subsequent background replacements will work better.
prompt:Replace the background with a solid color green screen
Prompt: Replace the background with a solid green screen
prompt：replace the background with the attached image. Make sure [subject] is lit to match the image；
replace the background with [describe your scene]. Make sure [subject] is lit to match the scene.
Prompt: Replace the background with the attached image. Make sure the lighting on [the subject] matches the image.
Replace the background with [describe your scene]. Make sure the lighting on [the subject] matches the scene.
8. Interior design with Nano Banana
Original image:
After Nano Banana "renovation":
9. Turn real scenes into game assets
prompt：Concisely name the key entity in this image (e.g. person, object, building). Create 3d pixel art of the isolated key entity in isometric perspective, 8-bit sprite on a white background. No drop shadow.
Prompt: Concisely name the key entities in the image (for example, people, objects, buildings). Create separate 3D pixel elements from an isometric perspective, 8-bit transparent images, with no shadow effects.
10. Turn city buildings into 3D
The content in 【】 can be modified according to the actual city.
prompt：Turn this photograph of a 【Parisian building】 into a isometric tile, in the style of the five other 3D.
Prompt: Convert the Paris buildings in this image into 3D isometric models.
The above case is from X blogger @Emm | scenario.com
Multi-image editing
1. Pose reference
Prompt: take the anime man and woman in the first image and put them in the poses of the stick man in blue and stick woman in red. erase the stick figures.
Prompt: Place the anime male and female from the first image into the poses of the blue stick figure man and the red stick figure woman, and erase the stick figures.
The above case is from X blogger @Justine Moore
prompt：Model pose like the sketch.
Prompt: Make the model's pose look like the sketch.
2. Image position reference
Mark the positions directly in the image, and Nano Banana can generate an accurate image.
3. Comics
Prepare reference images of the character; all three views are indispensable.
Mark the sections
Add prompts to the image to generate a comic:
The above case is from X blogger @けいすけ / AIマンガ＆開発
4. Element fusion Use Nano Banana to merge 13 images into one image.prompt：A model is posing and leaning against a pink bmw. She is wearing the following items, the scene is against a light grey background. The green alien is a keychain and it’s attached to the pink handbag. The model also has a pink parrot on her shoulder. There is a pug sitting next to her wearing a pink collar and gold headphones.
Prompt: A model leaning against a pink BMW, light gray background, wearing the following items: a green alien keychain hanging on a pink handbag, and a pink parrot standing on the model's shoulder. Beside her is a poodle wearing a pink collar and gold headphones.
This case is from X blogger @Travis Davids
Series IP/anime characters
For example, we have a character image like this.
prompt：First, please set up the basic color palette and the shadows and saturation.
Prompt: First, please set the basic color palette and shadows and saturation.
prompt：Next, please do the character model sheet.
Next, please create a character model sheet.
prompt：Next, please provide the [basic action set].
Next, please provide a basic action set.
prompt：Please give me the costume design set.
Prompt: Please give me a costume design set.
prompt：Please make an expression sheet.
Prompt: Please create an expression sheet.
Convert the image to line art, then color it with brand colors
Steps:
– Prepare the original image
– Convert to line art
– Color the line art with a color palette
– Change the character to brand colors
– Prepare the original image
– Convert to line art
– Color the line art using the palette
– Change the character to the brand colors
Video (storyboard) creation
1. First-person horseback ride through the twentieth century.
First use Nano Banana to generate various scenes:
prompt：dashcam google street view shot | Hobbiton streets | hobbits carrying out daily tasks like gardening and smoking pipes | sunny day.
Prompt: Dashcam Google Street View shot | Hobbit town street | Hobbits doing daily tasks, such as gardening and smoking | Sunny day
prompt：dashcam google street view shot | Seat of Seeing on Amon Hen | Ruined pavilion atop the hill, a hobbit-like figure from behind climbing the steps, the winding path down visible overlooking the river and lands beyond | panoramic view under emerging stars at dusk.
Prompt: Google Street View shot from a dashcam丨The ruined pavilion at the summit of Amon Hen's Seat of Seeing, a hobbit-like figure climbing the steps from behind, a winding downward path visible, overlooking the river and the distant lands, panoramic view at evening with the first stars appearing.
Use the prompt to create a first-person horseback riding view image:
prompt：dashcam google street view shot
Prompt: First-person view riding a horse on grassland. 12th century
Use Kling 2.1's first-and-last frame animation to generate a video clip.
prompt：”scene_description”: “The rider gallops out from the ruins of the ivy-covered statues, leaving the storm-lit plains behind. The path winds through rugged terrain as the pace remains fast. Ahead, a towering dark castle glows with eerie green light atop jagged cliffs, its spires piercing the stormy sky. Cloaked figures march steadily toward the fortress across a massive stone bridge.”, “visual_style”: “dark epic fantasy, cinematic, continuous POV”, “camera_movement”: “smooth forward gallop, first-person view without cuts, transitioning naturally from the ruined statues across the plains to the castle bridge”, “main_subject”: “the white horse’s head and rider’s gloved hands, centered as they race toward the looming fortress”, “background_setting”: “storm-darkened mountains and cliffs, a vast stone bridge spanning a deep chasm, leading to the glowing green-lit castle”, “lighting_mood”: “ominous twilight with green highlights from the fortress and flashes of distant lightning”
prompt：”scene_description”: “The rider gallops out from the ruins of the ivy-covered statues, leaving the storm-lit plains behind. The path winds through rugged terrain as the pace remains fast. Ahead, a towering dark castle glows with eerie green light atop jagged cliffs, its spires piercing the stormy sky. Cloaked figures march steadily toward the fortress across a massive stone bridge.”, “visual_style”: “dark epic fantasy, cinematic, continuous POV”, “camera_movement”: “smooth forward gallop, first-person view without cuts, transitioning naturally from the ruined statues across the plains to the castle bridge”, “main_subject”: “the white horse’s head and rider’s gloved hands, centered as they race toward the looming fortress”, “background_setting”: “storm-darkened mountains and cliffs, a vast stone bridge spanning a deep chasm, leading to the glowing green-lit castle”, “lighting_mood”: “ominous twilight with green highlights from the fortress and flashes of distant lightning
Editing these videos together produces this long time-travel effect video.
This case is from X blogger @TechHalla
2. Graffiti - 3D image - video
This case is from X blogger @Alex Patrascu
3. Let characters from famous paintings meet in the real world
This case is from X blogger @Alex Patrascu
4. AI anime
This case is from X blogger @Framer
03. How to use Nano Banana
Google Gemini (requires Pro membership)
On the Gemini official website homepage, select the Gemini 2.5 Pro model, and in the dialog box select Create images under Tool.
At this point, the Nano Banana model is used by default.
Upload an image, enter a prompt, and you can use it.
Google AI Studio (free)
Open the Geogle AI Studio website, and click the settings in the upper right corner.
In the settings, select Nano Banana.
Upload an image and enter a prompt, and you're good to go.
LMArena (free)
On the LMArena homepage, select Direct Chat mode at the top.
Then select the gemini-2.5-flash-image-preview (nano-banana) model, and you can use it directly.
Lovart (free for a limited time)
Click the Nano Banana model entry on the Lovart homepage.
You can then use the Nano Banana model directly, but the free period is limited and ends on September 2.
Freepik
The homepage carousel has a Google Nano Banana entry; click it to enter directly.
Official website:Freepik
Select the Nano Banana model to use it for free, and you can generate 10 images for free every day.
04. Some sharing
In the past, making a high-quality set of images with consistent details required professional skills and also took a lot of time.
Now with Nano Banana, just one short instruction is enough; whatever you say, it can associate and reproduce in a dozen or so seconds.
The industry hit hardest is the content industry. Not many companies truly dare to use AI for advertising creativity now, but after this, things may be very different.
With one prompt from a brand, dozens of versions of advertising creative can be generated within hours.
The same goes for anime and short dramas. Details that used to take an entire team months to polish may in the future be produced by one person while revising on the fly.
In the future, there may be "on-demand generated" plots—as soon as viewers post a barrage comment, the model immediately continues writing.
Content creation will enter a completely new mode of production.
Original link:The most complete collection of Nano Banana玩法大全 across the internet (with 4 free methods to get it for nothing)