[PR] I tried making a 30-second ad video from a product poster | Pollo AI | GPT Image 2 | Seedance 2.0

*This article is a PR article.

🐾 Introduction

You have product images, but no video assets.

You don't have the budget to book a photo studio, and you don't have a video team.

Even so, social media and ad slots demand video. I think this is a common story in the world of e-commerce and advertising.

Recently, I made a poster for a fictional dog supplement brand. I've posted the making-of on X (formerly Twitter).

I tried to see if I could make a 30-second ad video from that poster, which is just a single product image.

In conclusion, including design and editing, it took half a day to complete the 30-second video.

The finished video is at the end of the article, so first, I will write about the process of making it.

🛠 Tools used this time: Pollo AI

The generative AI platform I used for this video creation is called Pollo AI.

To describe its features in one word, it is an all-in-one type where you can create both images and videos in one place. Moreover, it has almost all the major models in the industry, not just their own, such as Veo, Sora, Kling, Runway, Hailuo, and Pika, which you can switch between according to your needs.

Similarly for images, you can choose from multiple models such as Nano Banana, Flux, Imagen, DALL-E, and Stable Diffusion.

This time, I used GPT Image 2 for images and Seedance 2.0 for video.

When viewed in the context of e-commerce operators, the functions available in Pollo AI are roughly as follows.

Image to Video (turn images into video). If you have one product image, it becomes a video asset with movement. Even for products that are difficult to photograph, you can turn them into video if you have an image. This was the core of how I used it this time.

URL to Video (create video from product page). It seems there is also a flow where you can provide the URL of a product page from Shopify or Rakuten and create ad video assets from it.

Template/Multi-version generation. You can create multiple versions from the same material at once and run A/B tests. It is a usage that also fits mass production for seasonal campaigns.

Dedicated Apps for specific purposes. Templates for specific purposes such as "Product Demo," "SNS Ads," and "Cinematic Trailer" are prepared, so you can start creating in the intended format without having to build from scratch.

Open your browser, choose a model, and write a prompt. That's all it takes to generate both images and videos. Even though multiple models are lined up, the operation feel remains the same, so I was able to move from images to videos within the same screen.

Being able to try all the major models with a single account was a surprisingly effective factor when I actually started creating. I could switch between them according to my needs, thinking, 'This model is good for this kind of image' or 'This model has smoother motion,' and refine my work.

🐶 Step 1: Decide what to sell

The first thing I did wasn't to touch the tool, but to brainstorm with generative AI.

What is the video selling? How many seconds long, for whom, and what is the message? I threw these questions at the generative AI and answered the questions it returned to organize my thoughts.

Deciding beforehand makes things easier later

If I hadn't decided this first, I probably would have gotten lost while writing the prompts. In fact, at first, I just had the feeling of 'I guess I'll just animate the poster,' but with that mindset, I didn't know where to start.

So I decided on just three things.

What is the product? A premium dog supplement from a fictional brand. The product bottle must remain clearly visible on the screen.

What is the length? One 30-second video. A realistic length that people will watch until the end even if posted on social media.

What is the tone? Carry over the same dark and quiet worldview as the poster into the video.

When it comes to product videos, the product is the star. The worldview is there to support it. Just by keeping this consistent, it becomes easier to give instructions to the generative AI.

🎨 Step 2: Derive from the product image

I started by creating the 'first frame' of each scene in the video as an image. These are the keyframe images.

Why start with an image?

At first, I didn't understand why I had to create an image first. Since it's a video, shouldn't I just create a video from the start?

I realized later that the model I'm using this time is a system that creates video from images (Image to Video). That's why I need the first frame for as many scenes as I want to turn into video.

Conversely, this also means that if you have one product image, you can derive each scene of the video from it.

I created six images this time. The first three represent a quiet world, and the last three represent the world after it has transformed. I designed it so that the first and second halves correspond to each other, creating a before-and-after effect.

The problem where the product logo changes shape every time

This is where I stumbled once.

The logo on the product bottle changes shape every time. Even if I write "use the same logo" in the prompt, the AI creates a new "logo with a similar vibe" every time. If the product's identity falls apart as an advertisement, it becomes unclear what video you are watching.

Reference Image When I used this feature, it solved the problem easily. I passed the completed poster image as is and asked it to "use the same bottle as the one here." As a result, both the logo and the shape became consistent.

I think this is an important feature when making product videos. If the product's appearance is inconsistent, it doesn't work as an advertisement. If you use a single product image as a solid reference image, the product will look the same even if you create different scenes.

How to write prompts

A tip for prompts is to write by separating elements. What is the main subject? What kind of space is the background? Where is the light coming from? What is the color combination? Writing these in separate paragraphs made it easier for the AI to pick up the key points.

It is easier to convey information by structuring it into chapters rather than writing everything in one long sentence. It is probably the same as when you explain something to a person.

🎬 Step 3: Turning images into video (the biggest hurdle)

This is where I spent the most time this time.

Seedance 2.0 is a model that creates a video from a single image when you provide one. This is exactly Image to Video. I think this is the feature that will have the most use cases for e-commerce businesses.

When the scene changes, a new image is required. One video from one image. When I understood this, it made sense why I needed six keyframes.

Failures I didn't expect, rather than the ones I did

I was able to sort out the technical stumbling blocks relatively quickly.

If you carefully write down what you don't want it to do, such as "the bottle should not move" or "the text should not change," you can maintain the product's identity. Writing down what you want it to stop doing seems to be an important process for AI.

That wasn't the problem.

It was supposed to be a product ad, but it started leaning toward art

Halfway through, I lost track of what I was making.

At first, I intended to make an "advertising video that conveys the product's benefits." It was supposed to be a simple structure where the product bottle is on the screen, a dog appears, and the product's appeal is conveyed.

However, as I worked on each scene, it gradually became more artistic. The expression of light. The camera movement. The flashiness of the effects. Before I knew it, my focus was pulled toward creating a "cool video," and the original goal of "selling the product" had been left behind somewhere.

For example, I included a cut in one scene where a dog sprints like a flash of light. It feels good as a video. However, when I think about it, the product bottle is not in the frame at all at that moment. I paused for a moment, wondering if that was the right decision for an advertising video.

The moment I stopped

I stopped and rethought several times, "Does this work as a product advertisement?"

The time the bottle is on screen. The moment the product name is conveyed. The relationship between the dog and the product. I took each cut that had leaned toward the artistic side, reviewed them flatly once, and checked, "Does this function as an advertisement?"

The finished video, as a result, is finished with an artistic leaning. However, in the final stage of editing, I placed scenes where the product bottle is clearly visible at the beginning and end, so I think I managed to keep the skeleton as an advertisement.

This was both a point of reflection and a learning experience. Because generative AI has a high degree of freedom, the focus tends to shift more and more toward expression. If you are making a product video, I thought it was safer to decide the "time the product is on screen" from the beginning.

Something like a tip

Also write down things you don't want it to do. "Do not change the text," "Do not go out of the frame," "Do not change the hue." Just adding this at the end reduces accidents.

And then, decide "what to show as an advertisement" from the start. This was the biggest lesson this time. It often works out better to mix atmosphere and specifications rather than writing it like a specification document, but if you lean too much toward atmosphere, you lose sight of the goal.

🪡 Step 4: Connect 6 clips into 1

Once you've reached this point, all that's left is editing.

Connect the six 5-second videos in order using editing software. I always use Microsoft Clipchamp. It comes pre-installed on Windows, and it has enough features for free editing software.

I changed how I connected them depending on the scene. Where it moves from silence to silence, I made it dissolve softly. Where it switches from silence to an explosion, I cut it sharply. I felt that having a change of pace makes it less boring to watch.

I layered the BGM and text overlays in the editing software. Since I had instructed not to include text in the video generation prompts, this is the first time the words are added.

And here is the result.

📦 Application for EC sellers: As a case study

Although this production is in the form of a commercial for a fictional brand, I think it can also be applied to the actual sites of EC businesses. As a tentative example, I have organized a few.

Case 1: Apparel sellers who only have product images

A case where the shooting budget is limited, and you have product images but no video material.

At this point, if you pass a single product image to Image to Video, it becomes a dynamic video asset. You don't need to create 6 scenes like I did this time; often, 1 or 2 scenes are enough to make a sufficient ad video.

Teasers before a shoot, advance announcements for new products, or assets for social media ad slots. You can turn images into video without waiting for a shooting schedule.

Case 2: Mass-producing holiday campaigns

A case where you need different creative assets every week for seasonal campaigns.

By keeping the product's appearance fixed with a reference image, you only change the scenes and backgrounds. The same product moves on a beach in summer, in a forest with autumn leaves in fall, and in a snowy landscape in winter. If you can make one of these in half a day, you can mass-produce them at a pace of 4 to 5 per month.

Instead of creating videos from scratch every time, you can turn them into templates and rotate them.

Case 3: Speeding up A/B testing

A case where you want to create three videos with different appeal points for the same product to verify which one gets a better response.

You can create different variations just by changing part of the prompt. I focused on one video this time, but if you were to create three variations using the same procedure, I think it would take about 1 to 2 days.

Compared to mobilizing a shooting crew to create materials for A/B testing, this is a realistic sense of speed.

✨ What I learned from trying it out

After making it from start to finish, I noticed a few things.

I was able to make a 30-second ad video in half a day. Honestly, I was surprised by this myself. 1 to 2 hours for design, 1 to 2 hours for image generation, 2 to 3 hours for video generation, and 1 hour for editing. That's half a day including the time for trial and error. Considering the feeling of outsourcing shooting and editing, this is quite a difference.

Everything from image to video was completed in the same place.Pollo AI has the same operational feel even when switching models, so I didn't get confused. Going back and forth between an app for creating images and an app for creating videos makes you tired just from managing files. Being able to proceed within the same screen was a small but significant difference.

It is difficult not to lose sight of the goal. The technical side went more smoothly than I thought. On the other hand, because generative AI has a high degree of freedom, it's easy to get distracted by the visual expression. The biggest lesson for me this time was that what should have been a product advertisement started leaning toward art. I thought it was important to stop and check what kind of video I am making several times.

Video generation still doesn't perfectly match your intentions. I feel that a distance where you generate a few times and choose the one you think is good enough is just right.

If there is anyone who wants to try it, I recommend starting by making just 1 or 2 cuts. Rather than aiming for a full-length video from the start, you'll probably be happier when one 5-second clip goes well.

🌙 Conclusion

When you hear about making ad videos with AI, it sounds somewhat difficult.

But when you actually try it, each step is simpler than you might think. Decide what to sell, create an image, animate it, and put it all together.

Each step is not exactly a new concept. It feels close to what people who make ad videos have been doing for a long time. The difference might be that you can now complete each step by yourself, in half a day.

You don't need to mobilize a film crew or rely on an editing team; if you have one product image, it can become a video. This has become a realistic option for individuals and small business owners alike.

What used to require a large budget and time not long ago—'making a video ad'—can now be finished by the evening, leaving you with a 30-second video in your hands. I think that is quite remarkable.

Afterword

Thank you for reading this far.

I realized halfway through that 'conveying the product's appeal' had been replaced by 'making a cool video,' and I had to stop once because that was a problem. Because generative AI has so many expressive options, the original purpose can sometimes take a backseat.

I am reflecting on that.

Pollo AII hope the appeal of this comes across, even if just a little!

Following or liking this post would be a great encouragement!
I look forward to your continued support!

#GenerativeAI #VideoGenerationAI #PR #VideoMarketing #ECSite #AdProduction #note #ImageGenerationAI #VideoProduction #Marketing #WorkEfficiency #ContentCreation #SNSAttraction #Branding #PolloAI #AIUtilization #TriedIt #MakingOf #Creative #Design