[PR] For nights without a shoot or a model. | For small-scale EC sellers | Pollo AI | GPT Image 2 | Seedance 2.0

*This article is a PR piece.
*This is a fictional brand.

Someone is walking through the city at night. They raise a camera. The shutter clicks.

Can you create a 30-second video ad like that without using any shooting equipment or models? This time, I decided to put that to the test.

The subject is a fictional retro camera brand called "Yoakari." The products, the people, and the setting were all constructed using AI.

There were moments along the way where I wanted to double-check, "Wait, this really wasn't filmed, was it?" and each time, I found myself staring at the screen a little more intently.

🎬 Pollo AI Overview

Pollo AI is an AI video and image generation platform.

I mainly used two features this time.

GPT Image 2 generates images based on text and reference images. It is a feature that can also handle images with layouts containing text, such as product posters.

Seedance 2.0 generates videos based on images or storyboards. Since you can specify reference images using @ tags, instructions like "use this character and this camera" are also understood.

🌃 Designing a camera called Yoakari

First, I decided on the product.

The reason for choosing a camera as the subject

I chose a camera as the subject because I have the freedom to decide what to shoot and where to shoot it. When turned into a video, a product that can be showcased through location and movement makes it easier to see the true potential of Seedance 2.0.

The setting is Shibuya at night

I chose Shibuya at night as the setting. The neon lights, the wet road surface, and the flow of people. A place with high density as a scene makes it easier for highlights to stand out on screen.

The name is "Yoakari"

The product name is "Yoakari" (Night Light). I wanted a coined word based on Japanese, so I came up with a few candidates, checked online to see if they existed, and then made my decision. A camera that captures the light of the night.

The look is a silver and black two-tone.

The look is a brushed silver top with a matte black leather-like body. An amber accent on the shutter button. I designed it with the image of a retro rangefinder-style compact mirrorless camera.

Once this was decided, I made the poster.

🖼️ Create a poster with GPT Image 2 and make it.

GPT Image 2 was used to generate the product poster.

The poster is composed of a spec card (F1.7, APS-C, 5-axis 6.0 stops, film simulation), an English tagline, and the product name wordmark. It came out quite clean, even including the small Japanese captions.

The first image that came out had the camera placed on the ground. It looked like it had been dropped, somehow. It looked more like a photo of a lost item than an advertising poster. I changed the composition to remove the ground and have it floating in the center.

As for the color, I initially had it generated with a cream-colored body, but it got a bit buried in the night neon. I switched it to a silver and black two-tone.

After correcting the prompt a few times, it took a shape that I could think of as an advertising poster.

It took time to get here, but once the foundation is solid, the subsequent steps become much easier.

🎞️ Create reference images for the person

When making a video, the person's face can change slightly between cuts. The eye shape might be different in the first half compared to the second, or the angle of the nose might shift.

This time, I designed it from the start to not show the face. Back view, side view, looking down with bangs covering the face, framing out. If it's not shown, it can't fall apart.

Also, I like the sense of distance that comes from not seeing the face, which makes it feel like it could be 'me'.

I made a sheet of the same person in 5 angles side-by-side. This is to ensure that the hairstyle, clothing, and body type do not waver even when the cuts change in the video. I initially tried to make it in a 3:4 ratio, but it was too cramped to line up 5 angles of the full body, so I changed it to a landscape 3:2.

The persona is a traveler in their late 20s walking around Shibuya at night. The camera is not included in the sheet. Since I have it separately from the product image, I keep them separated so they don't get mixed up.

I don't pass this sheet directly to the video; I choose one strong angle from it to pass on. I think of the sheet as a ledger of identity kept on hand.

Even though I put in so many instructions to hide the face in the sheet, when it became a video, Seedance 2.0 tried to be helpful (?) and produced a person with clear facial expressions. I can get closer by increasing the number of back-view cuts or specifying how to frame the shots, but it is harder than I thought to completely restrict the face from appearing.

This part is still under adjustment.

🧪 While working on this, I'm also building GPTs

I took a slight detour here.

I'm also working in parallel on a GPT that writes prompts for Seedance 2.0 using two images: a product image and a person image.

It's still a work in progress. This time, it served as a real-world test for that GPT.

Some parts work well, while others still need manual adjustments. It looks like it will be a process of nurturing it for a while.

🎥 Creating a video with Seedance 2.0

I fed two images—a product image and one chosen person image—into Seedance 2.0.

Seedance 2.0 generates 15-second clips. This time, I joined two clips to create a 30-second video.

The prompt it wrote initially just listed actions: walking, reaching out, posing, shooting, and continuing to walk. Even though there was movement, for some reason, the screen felt static.

I rewrote it. I re-centered the prompt around the reason for the movement—why they are posing, or what makes them stop in surprise.

I intended to connect the end of clip 1 and the beginning of clip 2 by transitioning from 'starting to pose' to 'pose complete.' I included instructions in the prompt to match the poses, but in practice, the transition between the two videos didn't align perfectly.

I used CapCut this time. I added a slight crossfade to soften the cut. I think it's more realistic right now to generate videos with the intention of absorbing the seams through editing, rather than aiming for a perfect connection.

🧭 What is Pollo AI?

Pollo AI is a platform specialized in AI video and image generation.

GPT Image 2 and Seedance 2.0 aside, it offers multiple entry points such as Image to Video, Reference to Video, URL to Video, Templates, and UGC Ads.

There is also a feature that allows you to generate video assets just by entering a product URL. I believe its strength lies in providing accessible entry points even when you have few materials on hand.

💼 Can it be used for small-scale EC?

I tested this with a fictional camera, but the process can be applied directly to real products. If you have photos of your product on hand and an image of the characters you want to feature, you can follow the same steps.

I've listed a few scenarios where I felt it could be useful.

As SNS assets for new products

Even at the stage where you only have product photos, simply adding motion with Image to Video can create assets that easily catch the eye on a timeline.

As a teaser video before a shoot

If you build the world-view first, you can create a vibe before the launch. It also seems useful for gauging market reaction.

For mass-producing sales and campaign content

When you want to roll out multiple versions of the same product for different seasons, events, or color variations, it seems to pair well with template features.

On the other hand, there are parts that take time

You do need a fair amount of time to refine the prompts. In my case, I spent the most time solidifying the poster. Once the foundation is built, the rest goes quickly, so not giving up on creating that first image might be an important point.

🌙 Conclusion

No shooting equipment, no models, no budget. Even so, I was able to place a person walking through the night streets, a camera, and the moment of the shutter click inside the screen.

It's not that AI stands in for a perfect shooting location, but rather that it can be placed as another option in situations where setting up a location is difficult. I feel that alone broadens the range of choices a little.

Yoakari is fictional, but I ended up liking the finished poster more than I expected, and I'm actually thinking I'd want it if it were real.

Afterword

Thank you for reading this far.

Some of you may have noticed, but in the latter half of the video, some of the text from the reference image ended up in the lower part.

I'm currently working on GPTs to see if I can make them without a storyboard.
I'm still in the middle of trial and error with constant failures!

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

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

#PR #PolloAI #GPTImage2 #Seedance #GenerativeAI #VideoGenerationAI #ImageGenerationAI #SmallScaleEC #VideoAds #AdProduction #Marketing #EC #Branding #TriedIt #Verification #Prompt #Creative #ContentProduction #OrganizingThoughts #note