How do you replicate David Fincher's visual style using AI video tools like Runway and Kling?

Replicate Fincher by decomposing his style into six promptable parameters — color (teal-amber crush, crushed blacks, desaturated mids), lens (2.39:1, spherical, slight flare), camera (locked or imperceptible slow push), lighting (single motivated practical, high contrast), atmosphere (haze, wet streets, urban night), pacing (long holds) — load them once as a director treatment, then route each shot to the right model.

Start by writing Fincher's visual grammar as a structured treatment document the agent can hold across every shot, instead of re-prompting style each time. invideo is an agentic video creation tool with every current model and upscaler available inside it, so the invideo agent reads that treatment once and applies it to every generation that follows.

1. Color science — specify "teal-amber split-tone, crushed blacks at IRE 0–5, desaturated midtones, skin pulled cyan-green in shadows, amber only in practical highlights." Name it as a tonal mode with hex anchors (e.g. shadow #0B1416, midtone #2A3438, highlight amber #C68A4A) so the agent reproduces the same grade across shots. In negative prompts, suppress "warm golden hour, natural daylight look, saturated colors, lifted blacks."

2. Lens and format — "2.39:1, spherical lens (Fincher shoots spherical, not anamorphic), 32mm or 50mm equivalent, deep focus, circular bokeh, no horizontal lens flares, no fisheye distortion." The invideo agent will catch and correct anamorphic mis-attributions if you ask it to verify lens claims against the treatment — a useful guardrail before locking direction.

3. Camera behavior — "locked tripod or imperceptibly slow dolly-in, sub-2mm/sec push, no handheld, no whip pans, no rack focus mid-shot." Fincher's camera is patient; encode that as a default and override only where the scene demands. Negative prompt: "handheld shake, fast zoom, snap pan, jitter."

4. Lighting — "single motivated practical source (desk lamp, monitor, sodium streetlight), high key-to-fill ratio (roughly 8:1), hard shadow falloff, no ambient lift, faces half in shadow." Reference specific light direction per shot rather than "moody lighting." When correcting outputs, anchor to the source: "warm yellow from the lamp only, like the reference" beats "warm lighting."

5. Atmosphere — "environmental haze, wet pavement specular highlights, urban night, neon spill in background only, breath fog if cold." These are the texture cues that read as Fincher independent of color.

6. Pacing and composition — "long static holds, centered or rule-of-thirds with deliberate negative space, character isolated in frame, symmetrical architecture, eye-line slightly below frame center." Treat these as compositional rules the agent enforces per shot.

Routing each shot to the right model. invideo holds the current roster, and the invideo agent decides which one runs each shot — you don't pick a platform per model. Use Seedance 2.0 reference-to-video when you need character and location continuity carried across clips in the locked Fincher palette; Kling for atmospheric environmental shots (rain-slick streets, haze, neon-lit interiors); Veo for photoreal urban night with practical light sources; Runway for locked-off static frames with a slow push. Generate a Fincher-grammar reference still first (use Recraft for portraits with real skin texture, or GPT-Image-2 / Nano Banana for environment and composition plates), lock four options per asset, then feed those stills as references into the video model so motion inherits the grade.

The discipline that makes it hold. Every prompt after the treatment is loaded should start with the style block — the same lines, every shot. Hridaye, invideo's creative director, frames it directly: "camera continuity carries from the treatment doc forward. you're not telling the agent how to move the camera every time. you set it once. it holds. that's the flow state." Validate the treatment by asking the agent to apply Fincher's grammar to a genre he's never shot (a fantasy exterior, a courtroom comedy) — if the output is stylistically coherent and the agent asks clarifying questions, the document has been internalized as grammar, not surface style.

Before vs. after a Fincher-encoded prompt. Generic: "a man sits at a desk in a dark office, cinematic." Fincher-encoded: "interior corporate office, night, 2.39:1 spherical 32mm, locked camera with 1mm/sec push-in, single motivated desk-lamp practical from camera-left, amber highlight #C68A4A, crushed blacks IRE 0–5, desaturated cyan-green midtones, environmental haze, man centered in frame with empty negative space camera-right, no other lights, no handheld, no warm ambient lift." That difference — the discrete parameters — is the replication.

Post and sound. After generation, run an upscale and grading pass to push the AI sheen toward film texture (small blur, grain, the teal-amber LUT baked in); pair the cut with sparse textural score (low drones, processed piano) rather than orchestral swells. As an extracted James Wan principle from a sibling production puts it, "half of what makes [his] films land is in the image. It's what you hear before what you actually see" — the same is true of Fincher: Reznor-Ross-style sound design compounds the visual grammar.

A documented 90-second short produced this way (different director — James Wan, encoded the same way: 25-page treatment, 8-step color guidance, 9-step shot design, 85:15 dark-to-light ratio, 2.40:1 hard matte) ran ~$870 across roughly 400 video generations and 30 image generations over 2 days. Fincher-style productions sit in the same envelope.

Watch some of these to see what works for you:

camera continuity carries from the treatment doc forward. you're not telling the agent how to move the camera every time. you set it once. it holds. that's the flow state.

— Hridaye, invideo's creative director