I’ve been testing the capabilities of Nano Banana Pro, and I wanted to try something different. Instead of generating random isolated images, I wanted to create a cohesive storyboard sequence, a visual narrative that flows logically from one panel to the next.

The result is a mini-project I call “Tidebound.” Here is a quick look behind the scenes at the workflow that made it happen.

---

The “Context Chain“

The core idea here was visual memory. I didn’t want the model to guess the continuity; I wanted it to see it. By feeding the previous generated images back into the model as context, Nano Banana Pro could “reason” about the sequence.

Here is the logic I applied for the sequence. As the story progressed, the context window grew:

• Panel 1 (img1): Prompt 0 (The Seed)
• Panel 2 (img2): @img1 + Prompt 1
• Panel 3 (img3): @img1 + @img2 + Prompt 2
• …
• Panel 9 (img9): @img1 + … + @img8 + Prompt 8

By the end, the model is essentially looking at the entire storyboard so far to generate the finale.

The Process

My first step was simply to generate a compelling starting image. I’ve been diving deeper into animation and anime aesthetics lately. This gave me the first “hero shot” for our story. With that visual seed in hand, I uploaded it to Gemini 3.0 Pro inside AI Studio to begin the narrative brainstorming. Here’s the prompt for First Image (@img1):

A wide, establishing shot of a surreal anime movie still set on a steep, cobbled street in a coastal European city resembling Lisbon. In the foreground, a female university student with round glasses and a yellow cardigan is riding a bicycle downhill, heavily motion-blurred to convey speed and excitement. Behind her, a yellow tram is traveling up the street, but the tracks defy physics, curving smoothly upward into the sky at a 90-degree angle towards floating islands of city blocks in the distance.
The visual style maintains a strict separation: The student and the diverse crowd of pedestrians are rendered in vibrant anime cel-shading with expressive outlines, while the textured cobblestones, the peeling paint on the tram, and the distant ocean are rendered with photorealistic, gritty textures.
Focus on atmospheric detail: The mood is sentimental and energetic. Fallen autumn leaves are swept up by the wind created by the cyclist. The crowd on the sidewalks is reading newspapers or drinking espresso, treating the gravity-defying tram as a mundane part of daily life.
Lighting Physics: Cinematic “Golden Hour” lighting. The low sun casts long, dramatic shadows across the cobblestones. The tram’s windows reflect the orange sky realistically. There is a slight chromatic aberration and lens flare on the edges of the frame to enhance the cinematic feel.

From that single image, Gemini helped me outline a complete storyboard, detailing the specific action for each panel.

With the storyboard ready, I began generating the subsequent visuals using the “Context Chain” formula mentioned earlier. For each new panel, I produced a batch of three variations using slightly different prompts—the images you see here are the “best of three.” Naturally, some attempts were more successful than others. This wasn’t always a failure of the prompt itself, but sometimes a reflection of an initial plan that needed re-evaluation.

Prompt List

prompt1: Generate the next shot. A medium-shot tracking the girl from the side as she pedals. She looks worriedly at a digital display board on the street corner that shows a countdown timer. The background shows the city buildings fading away as the road begins to rise above the rooftops. Keep the same anime art style and golden hour lighting.

prompt2: Generate the next continuous shot. The scene shifts to a dramatic low-angle worm’s-eye view placed near the ground. The girl has left the city streets and is now riding onto the base of the massive Sky-Ramp. The ground texture transitions visibly from the old gray cobblestones to a smooth, futuristic white pavement. The bridge curves sharply upward into the sky above her, stretching endlessly toward the floating islands. The girl and her bicycle look small and vulnerable against the massive scale of the bridge structure. The lighting is still golden, but shadows are stretching longer. Keep the same anime art style.

prompt3: Generate the next continuous shot. A high-angle shot looking down at the girl from behind and above. She is pedaling her yellow bike into a swirling mass of fog. The red backpack stands out as the only bright color in a scene dominated by grays and whites. The road beneath her wheels is wet with condensation. To the left and right of the bridge, there is nothing but an endless drop into the clouds below. The destination is completely hidden.

prompt4: Generate the next continuous shot. A tight profile shot from the side. The wind has picked up significantly, blowing her hair and clothes horizontally backward. She is hunching over the handlebars to reduce drag. The lighting is flat and cool, indicating she is deep within the shadow of the cloud layer. She is wearing a red backpack that stays tight against her back. We see the moisture of sweat on her skin and the reflection of the gray clouds in her glasses. The floating islands are no longer visible, obscured by the mist she is riding into.

prompt5: Generate the next continuous shot. A stunning Wide Shot from a high angle. The girl rides her bicycle out of the dense gray fog and breaks through the top surface of the clouds. Below her, the clouds look like a vast, white ocean floor stretching to the horizon. Above her, the sky is crystal clear and colored with a deep orange and violet sunset. In the distance, the Floating City (like in the image 3) hovers majestically, its windows glowing with warm evening lights. The road continues upward towards the city, bathed in golden light.

prompt6: Generate the next continuous shot. A Medium Shot from the side. The girl has finally stopped her yellow bicycle on a cobblestone path at the edge of the floating island. She is straddling the bike, leaning heavily over the handlebars, gasping for breath. Her yellow cardigan is rumpled, and her hair is messy. She is wearing her red backpack. To the right, we see the entrance to a charming, rustic house with a warm glowing lantern by the wooden door. Behind her to the left, we see the sheer drop-off of the island edge and the purple clouds far below.

prompt7: Generate the final shot. An Over-the-Shoulder shot from behind the boy inside the house looking out. We see the girl standing on the doorstep, framed by the doorframe. The warm light from the house hits her face fully, highlighting her her glasses, and her messy hair from the wind. She looks exhausted but happy to have arrived. She is still wearing her yellow cardigan and red backpack, looking tired but offering a shy, sheepish smile. Behind her, we see the street lamp from the previous shot and the purple clouds of the high altitude. It is a moment of quiet intimacy.

prompt8: Generate the final concluding shot. An Extreme Wide Shot from a high aerial angle. We see the entire floating island drifting in the deep blue twilight sky. The stone house is just a small, warm speck of yellow light on the edge of the rock. We can see tiny silhouettes of the boy and girl sitting on the edge of the cliff, looking out. The massive white bridge snakes away from the island, disappearing into the clouds below. The first stars are appearing in the upper gradient of the sky. The mood is lonely but peaceful.

Once all the panels were generated and the sequence felt complete, I moved into a final editing phase to polish and unify the storyboard. It was here that I performed the crucial continuity fixes, like removing the red bow tie from the first panel, and added narrative details, like the “Cloud Tide Rising” text above the countdown timer and painting in extra gray clouds to enhance the atmosphere where needed.

With the visual narrative complete, the final step was to give the project its identity. After exploring a few options, I decided on “Tidebound.”

As this experiment demonstrates, a “context-aware” workflow is a powerful method for maintaining visual consistency in a narrative sequence. A key takeaway for me was realizing you don’t need every previous panel as a reference; being selective (for instance, img5 = img1 + img3) can save input tokens and prevent the model from getting overwhelmed.

While Nano Banana Pro did an admirable job of holding the visual style together, any minor inconsistencies feel less like a fault of the model and more like a reminder of where deeper human creative involvement is still essential. Ultimately, this is just one of many possible workflows. I sincerely hope sharing this process inspires you to adapt it, build upon it, and continue exploring the vast creative potential we’re all just beginning to unlock. I hope you found it useful.