Paul Schrader backs AI in filmmaking-and is writing for an all-AI feature
Writer-director Paul Schrader is leaning into artificial intelligence for his next move. In a recent conversation with Vanity Fair, he said he's developing a screenplay he believes is "perfectly suited" for the first fully AI-based feature. "I think we're only two years away from the first AI feature," he said. His stance is simple: "It's just a tool."
Schrader frames AI as a natural extension of an author's craft. Instead of describing a facial reaction with prose, you "pixelate" it with code and prompts. "You can create the face, and you can create the emotion on the face," he explained-sculpting a reaction in moving images the way a novelist sculpts it in text.
AI as a new writing code
For writers, this shifts the job from only describing scenes to directing systems. Your words become production instructions: tone, blocking, shot framing, lighting, performance nuance. Clarity on the page turns into clarity on screen-because a model can only output what you define with precision.
This doesn't erase authorship. It changes the surface area where you apply it. You're still choosing theme, subtext, and pacing. You're just translating them into parameters that guide models as surely as stage directions guide actors.
He thinks AI could improve film criticism
Schrader also argues that AI may beat average coverage at consistency because it's not swayed by who's paying the bill. "Often, when you are doing coverage, you get a hint that the person who is paying you wants you to like this. You cannot give that information to AI."
He's even floated "robotic reviews" that emulate legendary voices. In his words, "It should be fairly simple to program ChatGPT to review a new film in the manner of, say, Kael, Sarris or Farber." If you're curious about building style-conditioned critiques, start with a general model like ChatGPT. For reference on those voices: Pauline Kael.
What this means for working writers
- Spec scripts will bifurcate: one version for human casting and production, another optimized for text-to-video pipelines. Expect more explicit visual beats, camera notes, and emotional gradients the model can read.
- Character bibles and style sheets matter more. Define micro-expressions, speech rhythm, posture, wardrobe anchors, and scene energy. Consistency is the difference between "coherent" and "muddled."
- Think in modular assets. Scenes, shots, and lines should be iterable blocks. You'll test dozens of variants fast-so label everything like an editor, not a diarist.
- Credits and rights become a real conversation. Track which model, version, prompt chain, and dataset policy you used. It protects you and clarifies authorship.
- Critique becomes a loop. Use AI to stress-test your draft from different critical lenses, then rewrite with intent instead of vibes.
Try this simple AI-first workflow
- Outline the premise in one paragraph. State genre, tone, and a one-line moral argument.
- Write an 8-12 beat sheet. Add camera mood, color palette, and pacing per beat.
- Draft scene cards with clear objectives, conflicts, reversals, and exit images. Include 3-5 shot ideas per scene.
- Generate mood frames or animatics to test visual language. Note what reads and what doesn't.
- Run an AI critique pass in different voices (e.g., analytical, sensory, cynical). Extract actionable notes, not generalities.
- Revise the script with tighter verbs, shorter beats, and explicit emotional cues that a model can stage.
- Log everything: prompts, seeds, settings, and outputs. Reproducibility saves hours.
Practical prompt scaffolds for writers
- Performance direction: "Deliver [line] with [emotion intensity 0-100], [breath pattern], [eye focus], [micro-expression: X ms], [pace: words/min]."
- Shot shaping: "Medium close-up, 35mm equivalent, low-key contrast 70/30, backlight hair rim, shoulder lead-in, 3-second hold after last word."
- Continuity guardrails: "Persist wardrobe code C2, maintain scar coordinate (x,y), keep ambient rain SFX layer at -14 LUFS, match color temp 4300K."
- Critical lens: "Review this scene in the style of [critic], with focus on [performance], [structure], [theme]. Output specific line edits and one paragraph on audience impact."
Guardrails worth setting
- Bias and voice: lock your values up front. Define what the story will not say or show, then enforce it across prompts.
- Citations: if you ask for facts or quotations, require sources and reject unsourced claims.
- Disclosure: keep a simple note of what was AI-assisted. Transparency builds trust with collaborators and clients.
- IP checks: avoid style mimicry of living artists or protected characters. Set rules for "inspired by" rather than "copy."
- Human edit pass: final drafts get a human ear for subtext, rhythm, and taste. Don't ship raw outputs.
The takeaway for writers
Schrader isn't handing the pen to machines. He's widening the canvas the pen can reach. If you can describe an emotion precisely, you can direct it-on the page and on the screen-faster than before.
Writers who learn to spec, direct, and critique with AI will ship more experiments, keep the wins, and cut the waste. That's the leverage.
Skill up, fast
If you want a structured path to build these skills, explore focused resources for your role and prompt systems that fit real writing work:
Your membership also unlocks:
