This is a question raised by a group member that resonates with me. I share similar doubts because current AI-generated realistic styles suffer from a strong uncanny valley effect, with Veo3 perhaps being an exception. In other words, audiences may immediately recognize AI-generated realistic works as artificial, breaking their immersion. They tend to be much more tolerant of animated works.
The second issue is that realistic styles can usually be generated directly by AI image generation software. This means the barrier to entry is relatively low, though creating high-quality work still requires skill. However, this results in an abundance of raw AI-generated realistic content flooding the market, leading to audience fatigue with this aesthetic.
In the AI era, realism is a stylistic choice rather than a “live-action” production method. This means AI can essentially produce whatever style we desire. Therefore, realism is simply an option. If you have a passion for realism, or like me, focus primarily on long-form narrative cinematic realism, you can choose realism as your style.
Creating animated styles might be easier for gaining traffic and recognition because there’s no uncanny valley effect, and the results can be more fluid. If you’re chasing views, you might want to reconsider obsessing over long-form narrative realistic cinema. AI currently has significant issues with long-form narratives as well. In short, don’t be too fixated on what can’t be achieved. Don’t be like me, stubbornly pursuing one thing to the end. Finding where your interests and strengths lie is most important.
If you enjoy animated styles or have expertise in them, working with animation may be the most suitable option at this stage. Both I and other AI creators have questioned the current AI-generated realistic styles. This skepticism isn’t unfounded—it’s a conclusion drawn from practical experience. Realistic styles have too many flaws and imperfections. They’re far from reaching film-industry standards and likely need another 1-2 years to mature.
Since I haven’t studied non-realistic styles, although I’m aware of current AI’s realistic style limitations, I’m unable to offer advice on creating non-realistic styles. I’ve found the path I want to pursue. Although there are current difficulties, I aim to expand the boundaries of AI filmmaking. Having watched so many great and terrible films, they’ve greatly influenced my aesthetic sensibilities.
This is why I work in realistic styles—I want to create AI films that rival live-action cinema. This is also due to my Computational Universe series, which depicts a world that should be portrayed with real actors, even though many of the characters are aliens. Therefore, I need powerful consistency control methods, specifically the three major consistencies: character consistency, scene consistency, and style consistency.
If you choose realism as your style and work with image-to-video workflows, you must upscale your images to remove the greasy appearance of facial features. This is a crucial detail that makes your work appear professional. While it extends creation time, it’s essential.
Next, you need to learn cinematic audiovisual language. I’m still learning this myself, as I’m a creator who crossed over from writing. I can’t immediately master storytelling through audiovisual language. However, I believe that when AI creates realistic styles, it should be benchmarked against live-action films. Audiences won’t lower their aesthetic standards just because it’s AI-generated.
Then consider stylization to distinguish your work from generic realistic styles. Here’s where our old friend color grading comes in—you can grade images once and videos once. This double color grading ensures your film has its own unified tone, distinguishing it from direct Midjourney or Veo3 outputs. Color grading is a powerful tool that can exponentially enhance the artistic quality of your work.
For realistic styles, character consistency is also crucial. This is why I’m currently only testing Veo3’s text-to-video but haven’t fully committed to text-to-video workflows. Image-to-video offers far more control than text-to-video, especially for complex clothing. I enjoy seeing characters in elaborate costumes performing my directed scenes in fantastical settings. This has always been my pursuit—I don’t like limiting characters to plain white t-shirts.
Why do we pursue seemingly impossible tasks? This relates to one’s life philosophy. Some believe the pursuit itself has value—that this is true innovation and excellence. Others consider it overambitious. So, is using AI for realistic styles a similar topic?
Follow me as we explore AI filmmaking together.
