Summary of Key Points
AIGC (Artificial Intelligence Generated Content) in the form of images is evolving from generating short videos through a “lottery-style” process to creating longer, more narrative-driven films with higher production values. However, there are currently two major obstacles: firstly, the computational power of AI models is limited, making it difficult to simultaneously achieve high precision in both the main characters and complex environments; secondly, AI lacks a physical coordinate system and spatial awareness, which leads to issues such as awkward character movements and misplaced camera shots. To overcome these challenges, creators are beginning to establish digital asset libraries (including reference images of characters, scenes, and props) similar to those used in traditional film production, and they are experimenting with various methods to help AI better understand spatial relationships. At the same time, digital assets have become a focal point of competition within the industry chain, with both tool providers and film companies making strategic investments to advance the industrialization of AI-generated content.
Detailed Analysis
1. Can AI Generate Long Films? First, Overcome the “Spatial” Barrier
Just as traditional commercial films require a studio to ensure consistent scenes and a realistic environment, AI-generated long films also need a “virtual studio.” However, AI still faces significant challenges:
- Asset Quality: Limited computational power results in either clear main characters with blurry backgrounds or attractive backgrounds with distorted characters. For example, during crowd scenes, AI may apply the face of the lead character to all actors or blur out the background actors to save on processing resources.
- Shot Sequencing: AI operates based on two-dimensional images and lacks a concept of three-dimensional space, which means it doesn’t account for factors like gravity and obstacles. In indoor dialogue scenes, characters might pass through table corners, and seat arrangements may become chaotic during camera transitions.
These issues directly affect the visual coherence and realism of long films, posing significant barriers to the advancement of AI-generated content.
2. Digital Assets: The Secret Weapon Against “Homogenization”
Digital assets serve as a “material library” for AI creation, including reference images of characters, scenes, and props (such as the lead character’s clothing, the appearance of monsters, or the clouds in a celestial setting). They are essential to address issues related to insufficient computational power and inconsistent styles:
- Preventing Homogenization: With limited AI capabilities, adding more assets helps prevent every detail from becoming mediocre. By preparing these assets in advance, AI can focus on generating content without wasting processing resources. For instance, the creators of “Zombie Sweeper” separately generated the lead character’s clothing and weapon and clearly defined the character’s characteristics to eliminate errors.
- Cost Savings: Assets can be reused, eliminating the need for repeated trial-and-error processes. For example, a blogger who worked on “Traveling through the Classic of Mountains and Seas” prepared digital assets for monsters and spectacular landscapes, which saved time and ensured consistent styling during filming.
- Enhancing Realism: Some creators use a combination of real-world footage and digital rendering; they shoot actual scenes and props to provide AI with high-quality reference data, making the generated content more physically realistic. For example, director Chen Xiaoyu went to Latin America to collect real footage for his AI-generated film.
3. Digital Assets Become a Valuable Asset in the Industry Chain
Digital assets are not only useful for creators but also represent a business opportunity:
- Tool Providers: Companies like Jellyfish Intelligence and Seko AI offer digital characters and stylized scenes as competitive advantages to attract professional creators.
- Film Companies: There is a demand for “AI actors” and singers, leading to businesses that specialize in creating customized avatars for use in AI-generated content. For instance, the team behind @MaiJu generates such avatars for commercial purposes.
- Intermediate Service Providers: Services that sell digital assets, such as facial and scene materials, help film companies quickly set up virtual studios.
In essence, whoever possesses a large quantity of high-quality digital assets will gain an advantage in the AI-generated content industry.
4. AI Lacks Spatial Awareness? Creators Are Teaching It to Understand Physical Rules
Since AI doesn’t understand three-dimensional space, creators are employing various methods to guide its behavior:
- Providing Multiple Angles: They show AI reference images from different perspectives (front, side, top) to help it construct a three-dimensional understanding of the scene.
- Detailed Instruction: They break down shot sequences into 5-second segments and provide detailed instructions, specifying camera positions and character movements.
- Using Schematic Diagrams: For example, some creators use diagrams to visualize three-dimensional spaces in two dimensions, indicating character positions and movement directions.
- Cutting-edge Research: The industry is exploring “world models” that give AI an internal physical engine, enabling it to understand gravity and coordinate systems, thus avoiding errors like characters passing through objects or cameras moving in unrealistic ways.
5. The Future: Spatial Sequencing Is Key to the Industrialization of AI-generated Content
For AI-generated content to be produced on a large scale like in traditional film-making, solving spatial sequencing issues is essential:
- Only by enabling AI to understand three-dimensional space can we create films with compelling visual elements, such as complex fight scenes and emotional shots.
- The ability to handle spatial sequences will become a benchmark for distinguishing the skills of creators; beginners may produce fragmented footage, while experts can use AI to create coherent and realistic long films.
In the short term, these challenges will persist, but with technological advancements and the accumulation of digital assets, AI-generated content is on its way to achieving industrialized production, similar to traditional film-making processes.
Conclusion
AI-generated content is evolving from a casual hobby to a professional field. However, issues related to spatial understanding and asset management remain critical hurdles. Creators are making progress by establishing digital asset libraries and teaching AI about spatial rules, while the industry chain is working to commercialize this technology. In the future, those who successfully address these challenges will take the lead in the AI-generated content market.