AI, a Squirrel, and a Living Scroll
How to Use AI to Promote Cultural Dissemination
Recently, to celebrate its annual natural gas output surpassing 50 billion cubic metres, PetroChina's Southwest Oil & Gasfield Company released a series of cultural merchandise. The star of the collection is a charming panda mascot, designed in a popular cartoon style. Dressed as an energy worker, it wears a bright red safety helmet and a matching uniform jacket, embodying the spirit of the industry with a friendly and accessible look.
Previously, creating such merchandise required professional designers and a lengthy design and production cycle. What set this collection apart was that from initial design to final approval, the entire process was deeply assisted by artificial intelligence, dramatically reducing development time and making high-quality design more accessible.
Making Your First AI-Powered Product
This case reflects a broader trend: AI is making high-quality creative development more accessible than ever. The following is a step-by-step guide on how to create a trendy squirrel character—from a simple idea to a physical figurine and a line of related products—using currently available AI tools.
Step 1: Concept and Visual Generation A successful IP begins with a unique visual design. The first step is to use a text-to-image AI model to turn an abstract idea into a concrete visual draft. This process allows creators to bypass traditional illustration skills and directly translate their imagination into a high-quality image.
For this, we use Jianying’s Jimeng AI (known internationally as Dreamina), a tool whose aesthetic model aligns well with contemporary Chinese tastes. A precise and detailed prompt is crucial. The prompt for our squirrel was: "A cute squirrel IP character, plump body, delicate features, light green hooded jacket, with fluorescent graffiti symbols on the hood and cuffs; the body is defined by hard-edged planes, paired with flip-up black sunglasses. Thick black outlines, high-saturation colours, very cool. The background is pure white with no shadows, close-up shot, professional studio lighting, digital art style."
Step 2: Generating Three-View Drawings To convert the 2D image into a 3D model for manufacturing, precise three-view drawings showing the character from the front, side, and back are necessary. AI tools can now execute this complex task efficiently, ensuring perfect consistency across all angles, which is a common challenge in manual drawing.
We used Google Gemini for this step, providing it with the image from Step 1 and the following prompt: "Based on the attached image of the cute squirrel IP character, please generate its three-view drawing, including front, side, and back views. It is required to strictly maintain the consistency of the character's body proportions, facial features, clothing details, and colour scheme. The three-view drawing must be presented on a pure white background, avoiding any extra decorations or shadows."
Step 3: Automated 3D Modelling With standardised three-view drawings, the next step is to create a digital 3D model. Modern AI modelling tools can now automate this process, generating a production-ready model from 2D images. This significantly reduces the time and technical skill required compared to traditional 3D software.
Using Tencent's Hunyuan3D, we simply uploaded the three-view drawings from the previous step. The tool then automatically produced a high-fidelity 3D model suitable for most figurine and merchandise production requirements, completing a task that would typically take a professional modeller hours or days. The 3D models generated at this stage can also be directly printed using 3D printers, allowing customers to see the final result immediately and provide feedback straight away.
Step 4: Physical Production and Merchandise Expansion Finally, the 3D model can be used for physical production, and the IP can be expanded into a diverse line of merchandise. The 3D model can be exported for small-batch 3D printing or sent to a supply chain partner for mass production, while AI design tools can quickly generate mockups for other products.
To visualize the possibilities, we used Lovart.ai with the following prompt: "Based on the IP character in the photo, please generate design mockups for four types of cultural merchandise, including a plush toy, a blind box figurine, a canvas notebook, and a hooded sweatshirt. The design style should be consistent, with bright colours, and highlight the core features of the IP."
When an Ancient Scroll Comes Alive
The same AI-driven logic used to create the squirrel can be scaled up dramatically. What happens when the object of creation is not a single character, but an entire historical masterpiece?
An exhibition in Paris, "An Era in Jinling," offered visitors the chance to interactively "walk into" a thousand-year-old Chinese painting. The artwork in question, An Era in Jinling, was painted by the Qing dynasty court artist Feng Ning. It is a meticulous depiction of urban life in Song dynasty Nanjing, capturing 533 individual figures amidst a bustling cityscape. The over 10-meter-long scroll is an invaluable historical and artistic document, preserving the intricate details of a prosperous, bygone era.
To create this immersive experience, the curation team first used 3D modelling and motion capture technology. They transformed the architecture, the environment, and all 533 figures from the flat scroll into three-dimensional digital assets, building a living, virtual city.
Visitors then create their own virtual character and wear a UWB (Ultra-Wideband) high-precision positioning wristband. By tracking the wristband's real-time location within the exhibition hall, the system synchronises the visitor's physical movements with their virtual character on a massive screen. This achieves a seamless effect of "the character entering the painting and following in real-time."
Building an AI Brain for Culture
The technology provider for the "An Era in Jinling" exhibition, Unilumin, is now developing an even more ambitious project: a large-scale AI model dedicated to traditional Chinese culture.
According to their plan, the model will be built with traditional Chinese philosophy (such as Confucianism, Daoism, and the Hundred Schools of Thought) as its "soul," and classical literary canons (like the Four Books and Five Classics and Tang poetry) as its "foundation." The initial goal is to construct a language model that can deeply comprehend the wisdom embedded in these texts.
Building on this textual base, the model will be upgraded to become multimodal, enabling it to learn from Chinese classical paintings, sculpture, music, architecture, and even the intangible cultural heritage from rural villages that may not exist in any online text.
One of the motivations for this project was the team's discovery that existing large models have a shallow and often flawed understanding of Chinese culture. Furthermore, they found that some publicly available overseas datasets contained hidden risks, such as sensitive political questions and answers embedded within long tracts of English text. This underscored the necessity of building an autonomous and secure cultural database.
You Can Make It, Too
Everyone knows I'm not exactly blessed with an eye for aesthetics or design skills. Using the methods from the first section, I also made an attempt at a mascot for the Brazilian company Petrobras, using Doubao AI developed by ByteDance.
It's still rough, but it's a nice first try.