Luxury brand site workflow

How to Build a Luxury Watch Website With AI Prompts

By Shreyas Raj. Written from my YouTube video, about 18 minutes long.

Video: I Built a $10,000 Luxury Watch Website With AI in 30 Minutes on the AI with Shreyas Raj YouTube channel.

Short answer: Use a master visuals prompt with a reference watch image to get first-frame and video prompts, generate the clips, join three of them for the hero, and send a one-shot website prompt to a coding agent. The same prompts work for any watch brand.

What you will learn

Dissect the site first: hero is three joined videos and the rest are images
A master prompt turns any product photo into first-frame and video prompts
Generate several image variants and pick the best clips
Edit the clips together in CapCut and give the result to the coding agent
Use JSON prompt examples to steer image output toward a style

Dissect the website before you build it

The first step is to understand what you are looking at. In this luxury watch site the hero video is just three AI videos joined together. The rest of the page is images with subtle movement, with image overlays in the back and the front. The subtle movement is what makes it feel premium, but they are essentially images.

Everything starts from a guide document with master prompts for visuals and a one-shot website prompt.

I also say the first ten minutes of the video cover the build, and the last part covers how I craft my own prompts and find inspiration, which is the part that turns this from a copy into your own work.

Run the master prompt with a reference image

I copy the master prompt for visuals, paste it into Perplexity and add the reference image of the watch. I found a reference on a watch information site. The subject can be any watch, a Rolex or a Patek, and only the subject image changes while the master prompt stays the same.

The output is a full prompt pack: first-frame image prompts for that watch and a video prompt for each first frame.

That prompt pack arrives in two parts, so you can copy the image prompts into an image tool and the video prompts into a video tool without any extra writing.

Variables such as brand name, product line and edition name are filled into the template. Everything else in the prompt stays fixed, which is why the same one-shot prompt can serve a different watch brand with only the inputs changed.

Generate images and videos

For this build I used Freepik Spaces because my Google Flow credits had run out. Open a new space, paste the prompts, add the reference image to an image generator and generate five variants in 16 by 9. You get a front shot, a face shot, a close-up and so on.

Next, I pick the best image and generate video from it. I used a Hailuo model, which I find the most realistic of the cheaper options. Attach the image, paste the video prompt and generate. If you can use Veo 3, you will get better results. Out of many clips, I kept three that looked good and discarded the wonky ones.

Because my Google Flow credits were over, I accepted lower-end results from cheaper models. I note that you would get a better result with Veo 3. The choice of model is yours and depends on how much you want to spend.

Join the hero clips in CapCut

Open CapCut, import the clips and place them in order to build a single hero video. This is the video that plays at the top of the site. Export it and keep it ready for the coding agent.

I started the sequence with one frame, added the next clip, then the final one. Arrange the order so the movement of the watch flows from one clip into the next, and trim anything that looks off.

Build the site with the one-shot prompt

The document includes a one-shot luxury website prompt with variables you can configure: brand name, product line, edition name. Leave variables blank if they do not apply, for example if a brand has no special editions.

I used GitHub Copilot, which has a free trial for a month and gave me access to Claude Opus 4.6. In VS Code I open a new window, create a folder, sign in to GitHub in the bottom corner, paste the prompt, pick Claude Opus 4.6 and set autopilot. In my experience Opus can one-shot most of what you give it, which means fewer back-and-forth iterations.

On the coding side, I explain why I switched. I had tried Claude Code and Antigravity, and Antigravity gave me problems, which is why I moved to GitHub Copilot with a trial. You can also use several accounts if your limits run out.

Give it the media and fix bugs

I download the images, copy the folder path and tell the agent: these are the images, use them wherever required, and this video goes in the hero section. Then queue a final instruction: deploy locally for me to check.

The first output was decent. It needed the brand logo, and two images were overlapping. I asked it to fix the overlap bug. I also asked Perplexity for a dissection shot: the watch opens up and the internal components spread out in layers, with a starting image prompt and video prompt. That is optional polish.

As I note, the better your video, the better the site will look. My output used cheaper models, so it was not the best, but the structure of the process is the point.

Find inspiration and write your own prompts

For inspiration I search Pinterest, for example luxury website for watch, and screenshot a design I like. I give it to Perplexity and ask for an image prompt in that style and a website prompt to put text over the image.

To avoid generic output, I grab a JSON prompt example from a prompt library site, tell the AI to follow that JSON format for the output, and then use the resulting prompts in an image generator such as Nano Banana Pro in Freepik. Better inputs give better visuals, and the sky is the limit.

I also say that I plan to teach how to sell sites like this, but that is a separate topic from the build itself.

After the first render I noticed the logo was missing and some images overlapped. Those are exactly the kind of issues you expect from a first pass, and each can be handled in one follow-up instruction.

Watch the full video to see the generation, the CapCut edit and the finished luxury website.

Watch the full video See EducateXAI

Results vary and depend on your effort. Read the results disclaimer.

Questions people ask

Can I use this for brands other than watches?

Yes. In the video I say the subject can be any watch, and the same idea applies to other products. The master prompt stays the same and you only swap the reference image and variables.

Why is the hero video three clips?

Single clips from image-to-video tools are short. Joining three good clips in an editor gives you a longer, more varied hero sequence, and lets you discard the clips that look wonky.

Why does the guide use JSON prompts?

A JSON prompt gives the image model more structure than a sentence. I borrow a good JSON prompt as a template and ask the AI to produce new prompts in the same structure for my subject.

What does the coding agent need from me?

A clear prompt, the folder path of your images and videos, and instructions about where each file goes. Then ask it to run the site locally so you can review and request fixes.

More from the channel

Prefer short reads? Start with the EducateXAI guides.