Here's how I created a virtual tour of this restaurant using Runway Gen-3 and Google Veo - honestly, it was a bit of trial and error at first, but I eventually figured out a workflow that works pretty well:
1. Base Material Preparation
So first things first, I photographed the restaurant from various strategic angles. I took this image as a starting point, trying to capture the key elements: the sea view, tropical vegetation, the set table, and natural lighting. I made sure to include those characteristic details like the fan-shaped leaves and the white lamp fixture. Pro tip - take way more photos than you think you'll need. Trust me on this one.
2. Narrative Path Definition
I spent some time planning out the camera movement for the walkaround:
Beginning: overall view from the entrance (though honestly I went back and forth on where to start)
Smooth movement toward the table area
Focus on details - like this area with the plants
Panoramic sweep toward the ocean vista
Return or continuation to other areas (still deciding on the best ending point)
3. Generation with Runway Gen-3
Okay, so I uploaded the image to Runway and started experimenting with prompts:
"Smooth forward camera movement through a coastal restaurant with tropical plants, maintaining the serene atmosphere and natural lighting. Camera glides past the table setting toward the ocean view."
I generated multiple short clips - usually 4-10 seconds each - because I found that shorter clips gave me better control over the visual coherence. Runway Gen-3 is honestly pretty amazing at maintaining consistency of architectural elements and that photographic realism we're after. Though not gonna lie, sometimes it does weird things with reflections.
4. Integration with Google Veo
For the trickier transitions or when I wanted to generate alternative views, I switched over to Veo:
I uploaded key frames and requested smooth interpolations between different angles. Veo really helped me out with the dynamic lighting and shadow movement on the scene - it creates these more cinematic transitions between the different restaurant areas. The results were... actually better than I expected? Sometimes AI surprises you in good ways.
5. Iterative Workflow
My process ended up being super iterative. For each segment, I'd make maybe 3-5 generations, then pick the one with the best spatial coherence and stylistic continuity. I paid particular attention to keeping elements consistent - like plant positions, water color, and light quality. Some days the AI just cooperates better than others, you know?
6. Post-Production and Assembly
I pulled everything together using DaVinci Resolve. This part included:
Color grading to unify clips generated by the different AIs (they have slightly different "looks")
Stabilization where things got a bit wobbly
Smooth transitions between segments
I also added some ambient background audio - ocean waves, light breeze, that sort of thing
7. Prompt Optimization
Through lots of experimentation, I figured out that my most effective prompts always needed to include:
Camera movement type specification (dolly, pan, tilt, etc.)
Atmosphere and lighting description
Mention of key elements to preserve
Movement speed indications (learned this the hard way after getting some overly jerky results)
Advanced prompt example that actually worked well: "Cinematic slow dolly shot moving from left to right through an elegant seaside restaurant. Maintain focus on the tropical fan palm plants in foreground, with soft natural lighting from the large windows. Ocean vista visible in background. Photorealistic, architectural photography style, 24fps smooth motion."
8. Challenges I Ran Into
The biggest pain point? Maintaining geometric coherence of spaces between different clips. Like, the AI would sometimes decide a wall should be in a slightly different place, which... yeah, not great. I eventually solved this by always using very clear reference images and just regenerating the problematic sections with way more specific prompts about perspective. Takes longer, but worth it.
Also, sometimes the plants would mysteriously multiply or disappear between shots. AI logic, am I right?
Final Result
The walkaround ended up being about 45-60 seconds long and honestly captures the vibe of this coastal restaurant pretty well. It's got that immersive quality that makes it perfect for the website or social media. Could it be better? Sure, there's always room for improvement. But for a first real attempt at this workflow, I'm pretty happy with how it turned out.
Would I do anything differently next time? Probably start with even more reference photos and maybe be more patient with the generation process. But hey, we learn as we go!