Summary of Key Points
AI films have been making quite a splash recently: The Hundred Flowers Awards have established a dedicated category for them, Youaiteng has launched certified AI-generated content, and projects that have met the Dragon Standard are preparing to hit theaters. There have also been long-form AI films shown at the Cannes Film Festival abroad. However, beneath the excitement lie several challenges—costs are rising sharply (the price of generating video has increased by 20 times), production timelines are unpredictable (the number of trial and error attempts using random generation methods is overwhelming), and model upgrades can render previous efforts futile. To overcome these issues, it ultimately comes down to people: individual creators need to gain experience in understanding how the models work, teams should adopt a collaborative approach with both “traditional directors” and AI-based directors, and meticulous production management is essential to mitigate uncertainties. The ultimate test for AI films, like for all content, lies in the story and the visual and auditory quality; technical gimmicks won’t hold up for long.
Detailed Analysis
1. AI Films Are Now Possible, but Far from Being Mastered
There have indeed been significant advancements in AI film production: The Hundred Flowers Awards introduced a category for AIGC (Artificial Intelligence Generated Content) for the first time, receiving over 2,000 submissions in one month; films like “Qitan” and “Reheating Chen Island” have been released on Youaiteng, and “Sanxingdui: Past and Future” has obtained the Dragon Standard, qualifying it for theatrical release. The international film “Hell Grind” just returned from Cannes, with the team working on a new 114-minute version. However, most of these works are still at the “usable” stage, far from being “user-friendly":
- Stiff Character Performances: AI-generated movements and expressions feel artificial; for example, in group conversations, characters’ positions become disordered after just a few seconds.
- Poor Special Effects: There is a lack of consistency in lighting and props; in indoor scenes, the position of doors and windows or the angle of lights can change abruptly, pulling the audience out of the experience.
- Difficult Team Collaboration: AI generates diverse characters, and different team members using different models result in inconsistent visual styles, making it challenging to achieve a unified look.
In short, while AI can help transform ideas into visuals, creating smooth, appealing, and emotionally engaging content remains beyond its current capabilities.
2. Three Major Challenges in Making AI Films: Costly and Troublesome
Making AI films involves three major challenges that cause significant headaches for creators:
- Uncontrolled Costs: Generating content with AI used to be inexpensive; now, generating a 720p 15-second video using Seedance2.5 costs $39 (about $2 per second), which is equivalent to the cost of a lunch. Individual creators can’t afford to make many mistakes—“Hell Grind” generated over 10,000 images and 16,000 videos in its first 22 minutes, but only 253 were deemed usable, each requiring more than 60 attempts, resulting in substantial financial losses.
- Unpredictable Production Times: AI-generated content often has inconsistencies (e.g., changes in facial shapes or clothing folds), necessitating frame-by-frame editing. The director of “Qitan” noted that maintaining consistency in indoor scenes is particularly difficult, as even minor discrepancies are noticeable to viewers, making the production timeline highly uncertain.
- Lost Efforts Due to Model Upgrades: Any progress made in understanding a model’s requirements can be undone by a model upgrade. For example, Li Muyang’s “Dream Thief Agent” used a combination of models, resulting in clear and refined visuals, but re-generation would incur significantly higher costs.
These combined issues make the industrialization of AI film production even more challenging than that of traditional films.
3. No Matter How Advanced AI Is, It Still Needs Human Guidance
The only solution to these uncertainties lies with humans:
- Individual Creators: They need to develop their own “model usage manuals” by testing various prompts to understand what works best for creating stable visuals.
- Team Collaboration: A hybrid approach of traditional and AI-based directors is effective. For instance, in “Qitan,” a traditional director focused on the story and artistic expression, while an AI director transformed the creative concepts into model commands. Teams can divide tasks between traditional roles (directors, screenwriters, artists) and emerging roles (AI content creators, prompt engineers) to complement each other’s skills.
- Production Management: Using data to manage uncertainties is crucial. Experienced teams lock in character and scene assets early on, set reference images and shot standards during production, and use detailed project management tools to track the cost, responsible parties, and quality for each shot. For example, the “Qitan” team quantified prompt design and generation costs to balance creativity with practicality.
As Bai Yikun, CEO of Linghe Culture, said: “AI is just a tool; whether it’s used well depends on the person using it.”
4. The Ultimate Test for AI Films Lies in the Story and Visual Quality
The novelty of AI technology won’t last long. Audiences may be intrigued by the “AI-made” aspect at first, but they will soon start criticizing details like changing faces or artificial expressions. AI films must compete on the same terms as traditional and animated films without any special privileges:
- Lack of Star Appeal: Few stars authorize their images for AI use, and AI films haven’t yet created their own star talents, so audiences won’t pay extra for them.
- No Special Treatment: AI films are not treated differently; viewers judge them based on the quality of the story and visuals, not on whether they were generated by AI.
- Limited Ability to Create Unique Styles: AI can only imitate existing styles; true artistic innovation requires top-tier creators.
Therefore, the ultimate goal for AI films is to forget the “AI” label and focus on telling a good story and using compelling visual and auditory elements to connect with audiences.
In Conclusion
AI films are not magic; they lower the barrier to production but increase the difficulty of creating high-quality content. The key to success lies in human creativity, aesthetic judgment, and management skills. There’s still a long way to go from being able to make AI films to making them truly outstanding.