I’ve summarized a concept that frequently appears in games: silent narrative. In games, letters, notes, recordings in the environment, and even conversations with every NPC allow players to independently piece together the complete picture of the story, making players feel as though they’ve created their own unique story.
For instance, in The Elder Scrolls V: Skyrim, there’s a flooded prison where players must read the notes inside to understand what happened there. There’s also a destroyed lighthouse where players piece together the story by reading the diary left behind and killing the monsters within.
This experience is almost exclusively found in games—a narrative technique belonging to the ninth art form.
In AI filmmaking, silent narrative refers to the details that audiences observe beyond the surface story, such as character costume design. Some might argue that AI cannot maintain complete costume consistency or complex clothing designs.
This is where we need to use LoRA and the constantly emerging new consistency tools. Consistency has always been a challenge that major companies are striving to solve. This won’t be a problem in the future. Instead, this is a conceptual issue. Some content creators tell everyone that character consistency isn’t essential—don’t pursue consistency; 70% similarity is enough.
But character consistency is crucial for telling long-form stories. If you’re satisfied with telling short stories, consistency isn’t a problem. However, maintaining consistency in characters’ faces and clothing is fundamental if you want to tell a long story.
I don’t stop at wanting character consistency—I also want characters to wear complex, designed clothing. This overall atmospheric feel is especially critical for science fiction stories. Coco Chanel once said that a person’s clothes are an expression of their inner self.
Currently, using LoRA or multiple references can achieve over 80% similarity. So use these tools—don’t find them troublesome.
As for scene consistency, apart from combining with 3D, using multiple references or Midjourney’s serf can solve part of the problem. However, the spatial sense of scenes and their internal logic can only achieve about 50% of the desired effect. I haven’t successfully trained scene LoRAs, so I mainly rely on multiple references for this part.
Second, in AI filmmaking, establishing shots are either too numerous or too few due to tool limitations—similarly, close-ups are too many or too few. Many segments require voice-over narration to tell the story, and I add more details to enrich the audience’s experience.
This is a balance that many AI films lack. Some films consist entirely of character dialogue or entirely of voice-over narration, making them difficult to watch. This returns to my previous point: if you think shooting short stories is sufficient and narrative isn’t essential, you won’t recognize these problems.
I don’t believe AI films cannot tell stories and can only make trailers. With upgrades to tools like Kling 2.1, tools respond to prompts with increasing precision. Subtle character expressions and camera movements that were previously impossible are all being improved.
Even when AI filmmaking began, creators made films look cinematic through excellent editing techniques. They’ll only become more film-like in the future.
Third, color grading is significant for silent narrative in AI. I’ve found that color grading can cure everything. Especially for images and videos generated with different tools, without color grading, they look very disjointed, like many people shooting together without achieving unity.
Unified style is the foundation of art.
Although I’m not a DaVinci Resolve expert, I’ve discovered that AI films offer two color grading opportunities: once for images and once for videos. You can use high saturation during image grading and add cinematic teal and orange. Then, during video color grading, reduce saturation while maintaining color coherence throughout the footage.
The overall tone of the entire film is also an essential component of a silent narrative. If you’re not a DaVinci Resolve expert like me, you can rely more heavily on image color grading with video color grading as support. In any case, I’m willing to emphasize the importance of color grading multiple times.
I’m obsessed with expanding its possibilities and boundaries whenever I do something. Don’t abandon the pursuit of excellence because of temporary limitations. We’re still in the foundational stage of AI filmmaking, but remember what I said about the importance of silent narrative.
