cs.CVOct 22, 2026
SaveFilmFlow: Agentic System for 3D Virtual Film Cinematography
Organizations: Cornell University · Shanghai Jiao Tong University · University of California, Los Angeles · Shanghai Artificial Intelligence Laboratory
Abstract
Complex camera motion is central to visual storytelling, yet translating cinematographic intent into controllable, executable camera trajectories remains challenging. We introduce FilmFlow, an agentic framework that converts natural-language instructions or reference videos into explicit camera trajectories for dynamic 3D environments. To generate these trajectories, a Director agent constructs executable camera plans by composing high-level motions from concurrent atomic operations that jointly control translation, rotation, and framing. The Director is supported by three specialized agents: an Interpreter that extracts scene-grounded cinematographic intent, a Scout that selects initial viewpoints, and a Critic that evaluates rendered results and provides visual feedback for iterative plan refinement. We also introduce CinemaBench, comprising 150 Blender scenes from self-created and licensed sources, with five camera instructions per scene for 750 text-directed tasks and 60 reference videos associated with a subset of scenes. CinemaBench supports evaluation of both text-directed camera generation and video reference-directed camera generation. Across a diverse set of evaluated baselines, FilmFlow achieves the strongest benchmark performance and the highest user preference, demonstrating effective spatial reasoning, temporal coordination, and visual composition for executable 3D cinematography.
Figures & tables
Figure 1: FilmFlow at a glance. Cooperating agents use visual feedback to iteratively refine executable camera plans, enabling text-directed camera generation and reference-video directed camera generation in static or dynamic 3D scenes.
Figure 2: System design of FilmFlow. The Interpreter grounds the instruction in the 3D scene; the Scout searches for an opening viewpoint; the Director composes concurrent atomic operations into a camera plan, optionally conditioned on a retrieved demonstration (ICL). A shared Critic combines hard checks, visual assessment, and temporal quality to guide both the Scout and Director.
| Function | Operations |
|---|---|
| Position | Dolly In, Dolly Out, Truck, Pedestal Up, Pedestal Down, Orbit, Path Follow, Position Follow, Height Follow, Distance Follow |
| Orientation | Pan, Tilt, Aim Follow, Screen Position Lock, Roll, Horizon Lock |
| Framing | Zoom, Frame Subject |
| Focus | Focus Pull, Focus Follow |
| Perturbation | Shake / Handheld |
| Hold | Static Shot |
Table 1: The 22 atomic camera operations supported by FilmFlow.
| T2V | V2V | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Method | Align. | Quality | Motion | Probe | Method | Align. | Quality | Motion | Probe |
| LAMP | -1.747 | 41.54 | 0.209 | 0.564 | Zero-shot VLM | -1.417 | 35.86 | 0.041 | 0.502 |
| GenDoP | -1.768 | 39.63 | 0.257 | 0.597 | LingBot-Map | -1.472 | 35.55 | 0.278 | 0.521 |
| CinemaTraj | -1.788 | 44.65 | -0.068 | 0.618 | VGGT- | -1.491 | 34.88 | 0.176 | 0.533 |
| FilmFlow | -1.461 | 47.60 | 0.555 | 0.829 | FilmFlow | -1.224 | 41.77 | 0.289 | 0.750 |
Table 2: Quantitative comparison on CinemaBench. Results are averaged over all 500 T2V and 40 V2V tasks. Align and Motion report VideoReward text alignment and motion quality, respectively; Quality reports the DOVER composite score (0–100). Higher is better for all metrics. Bold marks the best score in each column.
Figure 3: Qualitative comparison for text-directed camera generation. The instruction and target 3D scene are shown on the left; rows show each method’s output at the indicated timestamps. While the baselines mostly show walls or obstructed views, FilmFlow reveals the pedestrians, turns toward the boulevard, and brings the target green sedan into view.
Figure 4: Qualitative comparison for video reference-directed camera generation. Reference frames and the target 3D scene are shown on the left; rows on the right show each method’s output at the indicated timestamps. FilmFlow adjusts the viewpoint as the subject moves and keeps the subject visible across the sampled frames. LingBot-Map misses the subject in the opening frames, VGGT- crops the subject in later frames, and the zero-shot VLM loses the subject in the final frame. The comparison concerns camera behavior rather than scene appearance.
| T2V | V2V | |||||||
|---|---|---|---|---|---|---|---|---|
| Variant | Align. | Quality | Motion | Probe | Align. | Quality | Motion | Probe |
| w/o Scout | -1.542 | 46.11 | 0.387 | 0.770 | -1.344 | 40.77 | 0.243 | 0.673 |
| Sequential-only Operations | -1.547 | 45.06 | 0.536 | 0.775 | -1.314 | 40.78 | 0.278 | 0.717 |
| w/o Critic | -1.499 | 45.25 | 0.430 | 0.770 | -1.288 | 37.21 | 0.191 | 0.716 |
| w/o ICL | -1.474 | 46.12 | 0.447 | 0.724 | -1.296 | 39.04 | 0.172 | 0.766 |
| Full FilmFlow | -1.461 | 47.60 | 0.555 | 0.829 | -1.224 | 41.77 | 0.289 | 0.750 |
Table 3: Ablation study on CinemaBench. Full FilmFlow achieves the highest alignment, video quality, and motion quality scores in both tasks, and the highest T2V Probe score. Probe is the mean match probability from the CameraBench-based evaluator. Higher is better; bold marks the best score in each column.
Figure 5: Qualitative ablation for text-directed camera generation. FilmFlow shows close views of the red, green, and dark-clothed characters before returning to the red character. Without Scout, the opening remains a distant group view; without Critic or ICL, foreground characters obstruct the intended subject in intermediate frames. The instruction and target scene are shown on the left; columns show frames at the indicated timestamps.
Figure 6: Qualitative ablation for video reference-directed camera generation. FilmFlow maintains more consistent target framing across the sampled frames, while the ablated variants show less consistent subject scale, obstructed views, or loss of the target. Reference frames and the target 3D scene are shown on the left; columns show outputs at the indicated timestamps.
| T2V | V2V | ||||
|---|---|---|---|---|---|
| Method | Pass Rate | Top-1 Rate | Method | Pass Rate | Top-1 Rate |
| LAMP | 42.1% | 12.1% | Zero-shot VLM | 63.0% | 2.3% |
| GenDoP | 46.2% | 5.7% | LingBot-Map | 77.3% | 17.4% |
| CinemaTraj | 63.8% | 21.2% | VGGT- | 73.3% | 17.1% |
| FilmFlow | 84.3% | 61.0% | FilmFlow | 82.1% | 63.2% |
Table 4: User study on CinemaBench. FilmFlow has the highest pass rate in both tasks and is ranked first in 61.0% of T2V judgments and 63.2% of V2V judgments. Pass Rate is the proportion of completed judgments marked as pass; Top-1 Rate is the proportion of rankings in which a method is placed first. Higher is better; bold marks the best result within each task and metric.
| T2V | V2V | ||||||
|---|---|---|---|---|---|---|---|
| Method | Align. | Quality | Motion | Method | Align. | Quality | Motion |
| LAMP | 2.6 | 2.9 | 2.7 | Zero-shot VLM | 2.5 | 2.8 | 2.7 |
| GenDoP | 2.8 | 3.0 | 2.9 | LingBot-Map | 3.2 | 3.5 | 3.3 |
| CinemaTraj | 3.5 | 3.7 | 3.6 | VGGT- | 3.4 | 3.6 | 3.5 |
| FilmFlow | 4.2 | 4.4 | 4.3 | FilmFlow | 4.0 | 4.3 | 4.0 |
Table 5: Human ratings on CinemaBench. FilmFlow receives the highest mean ratings for alignment, visual quality, and motion smoothness in both tasks. Align., Quality, and Motion denote the corresponding ratings on a five-point scale. For V2V, alignment measures how well the camera behavior matches the reference, rather than similarity in scene appearance. Higher is better; bold marks the best result within each task and criterion.
Appendix figures & tables9 assets
Supplementary material from the paper’s appendix.
Appendix
| Operation | Effect | Parameters (defaults) |
|---|---|---|
| Dolly In | Move toward the starting target anchor. | distance_scale=0.72 ( ). |
| Dolly Out | Move away from the starting target anchor. | distance_scale=1.35 ( ). |
| Truck | Translate along the initial camera-right axis. | distance_scale=0.75 ; direction=right . |
| Pedestal Up | Translate along world . | height_scale=0.65 . |
| Pedestal Down | Translate along world . | height_scale=-0.55 . |
| Orbit | Circle a target in the world plane. | orbit_angle_deg=28 ; direction=right ; radius_scale=1.0 . |
Table 6: Atomic operations, their effects, and operation-specific parameters. Common parameters are given above.
| Dev | Test | Total | |
| Scenes (dynamic / static) | 27 / 23 | 46 / 54 | 73 / 77 |
| T2V tasks (easy / medium / hard) | 50 / 100 / 100 | 100 / 200 / 200 | 150 / 300 / 300 |
| V2V tasks | 20 | 40 | 60 |
| Human-written instructions | 20 | 30 | 50 |
Table 7: CinemaBench statistics.
Figure 7: Additional qualitative comparison for text-directed camera generation. The instruction requests a bullet-time orbit that starts from a rear three-quarter view, sweeps to the front of a crouching, backward-leaning figure, and settles into a locked-off hold. The instruction and target 3D scene are shown on the left; rows show each method’s output at the indicated timestamps. LAMP and GenDoP drift away from the figure toward the wall or empty frames, and CinemaTraj approaches the figure but leaves it off-center under an oblique view. FilmFlow keeps the figure centered as the rooftop slides behind it and ends on a stable frontal view of the backbend.
Figure 8: Additional qualitative comparison for video reference-directed camera generation. The reference clip moves from a view of the group toward a single subject and ends on a tight close-up. Reference frames and the target 3D scene are shown on the left; rows show each method’s output at the indicated timestamps. LingBot-Map stays close to a static wide view, VGGT- crops the subject in intermediate frames, and the zero-shot VLM loses the subject in the final frame. FilmFlow reproduces the progression from group view to single subject and ends on a close-up that matches the reference framing. The comparison concerns camera behavior rather than scene appearance.
Figure 9: An example of Scout refinement. The Director-specified opening (w/o Scout) and the Scout-selected opening are shown for the same instruction, with the Scout’s selection rationale and the corresponding camera states. Elevation denotes the camera’s pitch angle; the Scout-selected camera is placed higher and pitched downward toward the subject.
Figure 10: An example of Director refinement under a fixed opening. Frame sequences and operation timelines before and after revision show how Critic feedback guides changes to operation parameters and temporal composition.
Figure 11: An example of in-context conditioning across different scenes. The retrieved demonstration is displayed separately from the current task. Results with and without ICL use the same current instruction and target scene.
Figure 12: Grey model render as a video-generation condition (group close-ups). The top row shows the grey model render of a FilmFlow trajectory that gives successive close-ups to figures in a group. Each subsequent row is generated from the same render with a different text prompt. Camera motion, framing, and subject placement follow the render, while appearance follows the prompt.
Figure 13: Grey model render as a video-generation condition (leading shot). The top row shows the grey model render of a FilmFlow trajectory that leads two walking subjects while other characters enter from the right. The four generated videos keep the framing of the pair and the timing of the entering characters across different styles.