AI Stock Service
Summary
Regardless of job title—whether researcher, marketer, or developer—and regardless of one’s proficiency with AI tools, we envisioned an environment where anyone could
quickly and easily download high-quality content for use in product detail pages, social media, and other applications.
This idea naturally evolved into the concept of AI STOCK—a centralized platform designed to generate, store, and provide access to a vast library of creative content.
Background
While many consumers make purchases influenced by advertising and marketing, when it comes to cosmetics that come into direct contact with the skin, the experience of testing the product is considered crucial.
This raised the question: What does this “experience” look like in an online setting? That question became the starting point of this project. Online experiences are just as important as those that occur offline.
Although online experiences are more limited in scope, they can still spark curiosity and engagement through images and videos that vividly convey color, texture, and formulation.
This sensory appeal often motivates consumers to visit physical stores.
For this reason, many beauty brands continue to produce visual cuts that emphasize color representation and texture expression — elements that remain essential for effective product storytelling and description.
Process 1. Categorize
Before generating images, we organized and categorized various business areas, taking into account user convenience and potential future applications in the workplace.
Process 2. AI tool study
#RunwayML
The text-to-video results were highly unpredictable, while image-to-video produced overly simplistic outputs. By training the model with reference images using Image + Text to Video, then removing the reference images and running Text to Video, we occasionally achieved more distinctive results. However, the outcomes remained inconsistent and random — a clear solution is still needed.
#Midjourney
Generating character-based images was relatively straightforward. Facial expressions and poses were generally well-rendered according to the prompts. However, more detailed prompts were required when depicting product textures. While makeup products such as lipstick and foundation were easily generated, basic skincare items like creams and ampoules proved more difficult. The consistency and viscosity were not accurately reproduced from the start — the generated images often resembled ice cream, whipped cream, or slime, which made them appear somewhat ambiguous as cosmetic textures. To improve accuracy, I incorporated reference photos of real cosmetic textures during generation. Moving forward, I plan to deepen my understanding of cosmetic formulations and refine my prompt-writing skills.
#Hailuo
To save time, I experimented with an additional video tool. However, since the two tools differ significantly in functionality, the overall workflow speed did not improve as much as expected. Compared to RunwayML, Hailuo is more passive, making it better suited for applying subtle motion effects while maintaining the integrity of the original image.
#Magnific AI
This tool enhances and upscales generated images — not only improving resolution but also eliminating the subtle AI-model look often seen in Midjourney outputs. It reproduces skin textures with impressive realism. It also performs fairly well when generating images directly from text prompts.
The text-to-video results were highly unpredictable, while image-to-video produced overly simplistic outputs. By training the model with reference images using Image + Text to Video, then removing the reference images and running Text to Video, we occasionally achieved more distinctive results. However, the outcomes remained inconsistent and random — a clear solution is still needed.
#Midjourney
Generating character-based images was relatively straightforward. Facial expressions and poses were generally well-rendered according to the prompts. However, more detailed prompts were required when depicting product textures. While makeup products such as lipstick and foundation were easily generated, basic skincare items like creams and ampoules proved more difficult. The consistency and viscosity were not accurately reproduced from the start — the generated images often resembled ice cream, whipped cream, or slime, which made them appear somewhat ambiguous as cosmetic textures. To improve accuracy, I incorporated reference photos of real cosmetic textures during generation. Moving forward, I plan to deepen my understanding of cosmetic formulations and refine my prompt-writing skills.
#Hailuo
To save time, I experimented with an additional video tool. However, since the two tools differ significantly in functionality, the overall workflow speed did not improve as much as expected. Compared to RunwayML, Hailuo is more passive, making it better suited for applying subtle motion effects while maintaining the integrity of the original image.
#Magnific AI
This tool enhances and upscales generated images — not only improving resolution but also eliminating the subtle AI-model look often seen in Midjourney outputs. It reproduces skin textures with impressive realism. It also performs fairly well when generating images directly from text prompts.
Process 3. Apply
To test whether the images created with Midjourney could be seamlessly integrated into product detail pages or Instagram without feeling out of place, we planned a virtual feed layout.
The product selected for this test was Happy Bath’s Micro Micellar.
All the photos and videos used on the actual product detail page were shot in-house at our studio.
However, due to studio limitations, it was difficult to capture dynamic scenes such as splashing water or the face-washing process, so the final cuts only depicted the post-cleansing stage.
To supplement this, I used Midjourney to create visuals representing the washing process that could not be produced in real life.
We also conducted an Instagram application test, selecting LANEIGE as the brand.
The chosen products were a cream and a lip gloss, and we focused on generating images such as product shots, model visuals, texture details, and lifestyle-oriented cuts.
For the lifestyle shots, I aimed to create a natural and authentic look by using prompts like “shot with an iPhone.”
We brought together the developed images and videos to design the AI Stock page.
The project is currently in the content development phase, with plans to launch the service once a sufficient library has been established.
- Amorepacific Creatives
- 기획 및 디자인
- 박영주, 이윤진, 천나리
