I recently wrapped a short film that required a complex night exterior sequence—low light, practical neon, and a Steadicam move through a narrow alley. Traditional pre-viz would have meant a location scout, hand-drawn storyboards, then either a 3D layout in Blender or a rudimentary stand-in shoot with actors. For our micro-budget, that process typically took three days and cost more than we had.
So I tried something different: I used an AI pre-viz tool (ours at ZipX has a “Director’s Room” with archetypes—I selected “noir” for that sequence) to generate a sequence of shot concepts from a text description of the narrative intent. Within six hours, I had a reusable animatic with 14 shots, including camera movement and lighting notes. The time savings were real: 3 days → 6 hours.
But here’s what broke: the AI completely ignored the spatial continuity of the alley. It generated a wide shot where the protagonist stood on the left, then a close-up where he was on the right. No 180-degree rule, no awareness of eyelines. I had to manually reposition characters across frames—which took another two hours. The tool is great at mood and composition, but terrible at spatial logic. That’s where the director’s eye still matters.
This experience aligns with something I’ve been watching: ByteDance’s Seedance 2.0 is generating serious revenue ($1B monthly, reportedly), but the pricing power is shifting to the platform owners, not the creators. Indie filmmakers face a choice: adopt these tools or risk being left behind, but also risk becoming dependent on platforms that dictate terms. It’s the gold rush, but who really owns the picks and shovels?
For me, the lesson is clear: AI pre-viz is a collaborator, not a replacement. It saved me days, but I had to reassert directorial control over blocking and continuity. I’d be curious how others are handling this—have you found workarounds for AI’s spatial blindness, or do you prefer to keep pre-viz fully manual for certain types of shots?
I just love what you guys do. Without you we scribblers would have..... scribbles.
3 people like this
Cartney Wong sounds like you could have done the same shot list in Poser or Daz3d or Unreal Engine or Blender or Unity (all free apps), in less time, with optically correct virtual cameras. You'd use basic shapes for the environment for speed, but the process will reflect actual spatial layout. AI only guesses the next pixel in an image or sequence. It doesn't and cannot simulate an environment or effect. 3D process is about simulating reality. That's my go-to. AI can generate a basic "shot list" without attention to such matters from your script in moments, and from there select/revise which ones you want to storyboard out for visuals.
1 person likes this
Pencil and paper still work too. If someone cannot visualize it from there then they have no business in this business.
I was just going to say that. Pencil and paper is what I use.
1 person likes this
Cartney Wong I can often finish a storyboard in 30 minutes or less and create cinematic previsuals that are very close to what I see in my head. My advice to anyone getting into AI visuals is to learn the fundamentals first. Even basic graphic design, composition, lighting, color theory, image manipulation, and photobashing will make your AI results significantly better. AI is most powerful when it's combined with traditional visual-development skills, not used as a replacement for them.
Consistency, facial expressions, composition changes, and visual iteration become much easier once you understand the workflow.