Full Breakdown: My AI-Powered Video Production Workflow

Full video transcript

Hey everybody. In this video, I’m going to be sharing with you a breakdown for how I created a 30-second video commercial for a product that I’ve been developing with a client partner. We are developing an AI tool to help coach real estate agents in real time on calls to help them book more appointments. We had an urgency to turn this commercial around in just a few days because the product is about ready to come out of the wait list and we wanted to put together something that was a little bit more visually compelling that shared the message of the problem that we’re solving and the solution of where agents can go to sign up to learn more, to get started for free. And I’m going to show you the commercial first and I’m going to walk you through the full breakdown of how I produced this through the creative process and the tools that we used to tie it all together. So go ahead and jump right in.

The commercial itself, in its final form that we exported the files from, was in Descript. So I’m just going to play the commercial in Descript. “Real estate agents, you’re sitting on money in your CRM, but you burn that money when calls don’t turn into meetings. You need more conversations that become appointments. But in the moment, it’s hard to know what to say. And when the call ends, you’re like, ‘Damn, I could have had it.’ Seeso fixes that. It gives you live prompts during calls so you can book meetings right away. Fast coaching, clear guidance, day-one appointments, even for new agents. Stop burning leads and start booking more meetings. Try Sayso at asksayso.com.” And then it stays on this slide for the next 30 seconds so people have time to go to the website and scan.

But that is the 30-second commercial spot for Ask Sayso, helping real estate agents book more meetings. But how did we get here?

Basically, this process started with a conversation between my client and I in Slack. We made a pivot toward the real estate market and needed to create some new assets for it. So our conversations in Slack consisted of screenshots and voice notes, and I opened up a Claude thread to start this whole process and build the context of everything we were creating. Some of our initial context was basically saying, “Is there any way to create a video, a pitch video, that we could use to demo what this tool is for?” Once we have video demos from the actual software, we’ll be overlaying those into parts of this commercial, so that’ll be an easy addition to update later.

Basically, the value proposition around the commercial was that real estate agents will lose commissions because they’re not making calls, or they don’t know what to say. Seeso is going to be guiding agents on demand without needing to bring in a human coach in real time or on the follow-up side of things. So this entire thread was started by me inside of Claude because I believe Claude is one of the best tools for writing scripts. I find that Claude is really effective with that. We started with taking this initial context and screenshots, then working through a lot of different prompts and iterations to create a storyboard we could follow, mapping out the story and the core value proposition we wanted to articulate in under 60 seconds. We ultimately got it right around 30 seconds.

So what Claude and I did during this process was map out the script and come up with different versions of the prompts that we could ultimately refine and hand off to the next step, which would be running them through Gamma Studio. There was a lot of iteration, a lot of back and forth, a lot of refining based on what the prompts were giving me.

In tandem, I was working inside of Gamma. This is the final Gamma we ended up generating the slides through, but it went through multiple iterations first. There was a bunch of iteration on the design and style based on feedback, having conversations with my client, getting through some initial slides and demonstrations to make sure we were on the same page about exactly what we wanted the visuals and animations to look like.

Claude was giving a full frame-by-frame breakdown, and we were creating prompts to complement each of the frames, while also tightening up the sales script my client partner was speaking. I then took that script and ran it through Descript with his voice-over to tie everything together. More on that in a bit.

The main thing I want to document here is how many iterations happened inside Claude to get to a final set of production notes for that cartoon-superhero style, custom art-style content brief. Once we had that style guide locked down for all the slides and all the prompts, I thought it would be cool to have some music kick in, which you heard at the part where the product gets introduced — “Sayso fixes that” — you can hear the music in the background picking up the energy a bit.

I took that same prompt and created music using Suno AI. You can see two songs; I think this was the one we ended up choosing. Inside Suno, I gave this prompt to Suno’s song description, made it an instrumental, and generated the theme music. I have a license with Suno, so it made an AI song.

From there, coming back to the explainer video, we get to the final slide-creation part. Here’s where we take all of that context into Gamma. There were a bunch of different iterations, with this being the winner — the best one out of everything we generated. All of that context was set up by going into Gamma Studio: you go to Gamma, generate a new Gamma, select “Studio,” and that takes you to the studio view. Studio is great because you can tap into their image models and create far more visually driven scenes versus the more slide-and-text style of regular Gamma editing.

I pasted in the storyboard slides as initial context, and that generated an outline. A key point here: Gamma will take that content — say, 20 slides’ worth — and create a title outline with a bullet point of context per slide. I found that this can throttle the creative direction a bit more than ideal. What I did differently this last time, which got much better results, was: once the 20-slide outline was generated, I went back and copied and pasted the exact prompts from the visual prompt style guide into each individual slide. It took extra time, but made each slide prompt fully fleshed out and specific. I’d recommend that if you need hyper-specific details on any slide, make sure your outline has everything you want in it, because the detail you put into the prompt pays off in the output. This also applies if you’re creating social content with the same image style but different copy variations — you’ll want those copy variations built into the outline too.

Once the slide outline was generated, I selected “just vibes” for text density (minimal text versus a lot of text) and chose a custom theme built with all of Seeso’s brand style guides — colors and themes — so the visual vibe matched the brand. Then I used the AI image model to generate the slides. There’s an “auto select” option that’s fine to use, but there’s also a newer option — Nano Banana Pro HD — which I hadn’t even seen a few days before making this video; I used Nano Banana Pro, which had discounted pricing. Be aware that a generation can cost a meaningful chunk of your credit balance depending on the image model you choose.

Then there’s the custom art-style prompt — beyond the basic ones Gamma provides — which I built out using the Claude thread to really lock in the specific style. That prompt applies across all the slides alongside the theme and other elements. Once you click generate, it produces the output, which you can modify further. From there, I made sure all the images followed the story we were trying to tell. Some slides I duplicated and edited further — an important Gamma workflow: once you generate an image, you might want to make additional edits, so you clone the slide and use Gamma’s AI agent chat (a chat bubble) to request specific changes, like “add this text” or feedback-driven tweaks from the client.

Once the slides were finalized, I shared and exported all of them as PNGs. From there, I brought all of those image files into a Descript project — you can see all the different exported Gamma images on the right-hand side of the project files. An important note when tying everything together in Descript: I wanted to layer the images on top of the actual voice-over from my client. I had my client submit a voice note of the script through the same Slack thread, and once I added that voice note into Descript, it transcribed it word-by-word, letting me edit the words directly.

One thing I like about Descript: under AI tools, you can select Studio Sound, which cleans up and tightens the voice — cutting background noise and making it sound more studio quality. I applied that to the voice-over from the Slack note. Then I started refining: I placed all the images onto the timeline and, based on the cadence of when the voice was changing, resized them larger or smaller and adjusted where each slide appeared and transitioned. After a few iterations, we removed a few slides, combined others, and added transitions timed to specific lines, like where the music begins to swell alongside the video around the point the product name is introduced.

For one image that was a bit beyond Gamma’s capabilities to render well, I exported it, uploaded it into Google Gemini to remove the background, then — since I was having trouble getting a fully clean removal — brought the Gemini output into Canva to build out the rest of that slide. Gemini is really good at specific edits but doesn’t always nail exactly what you’re looking for, so I used Gamma for the final version of that slide and uploaded the resulting PNG directly into Descript.

With all of that in place, we had the voice-over, the theme music coming in around the 13-second mark, and all the slide images matching the stick-figure-plus-superhero style. On the final slide — sourced from one of the other Gamma generations — I added a QR code linking directly to Ask Sayso. Once everything was assembled, it was simply a matter of downloading the video and uploading it to YouTube. From there, the video went live on the website (built in Lovable), where visitors can see the headline, sub-headline, and an opt-in to join the wait list alongside the video.

In summary: extract the idea and context from a source like Slack, and start a Claude thread so all of that context — script writing, style guide, custom art prompt — stays aligned in one place. Create a lot of different prompt iterations there. Use Gamma and its Studio mode to take the final visual slide storyboard, generate and refine an outline, select “just vibes” for text density, choose a custom brand theme, and build a custom art-style prompt — this produces a number of different presentation options, from which you pick the best pieces across a few. Export the slides, upload the voice-over, all the presentation slides, and the theme song generated in Suno into Descript, and assemble everything there. After several rounds of iteration, export the final video and upload it to YouTube, and it now lives on the site.

I hope you find that breakdown useful for how you can tie all these tools together. Also worth noting: Gemini can edit very specific video images and follow specific style guides for combining images. All in all, it’s a handful of tools tied together to create a finished 30-second video sales piece for people learning about this product and business, driving them to sign up and join the wait list. Hope you found that useful — let me know if you have any questions. Good luck, y’all.