AI Cinema Studio turns 18 camera moves into a shot

WaveSpeedAI's AI Cinema Studio, announced on October 2, 2026, turns a camera move into a button. You choose the movement from a list of 18, and the studio writes that move, along with its lighting, lens and atmosphere, into the prompt it hands to the video model. The point of the tool is that you stop describing a camera in prose and start picking one from a menu.

What Is AI Cinema Studio?

It is a director's panel that sits above the prompt box in WaveSpeedAI's generator. Five cards cover the shot: References, Film setup, Camera, Color palette and Lighting, and each opens a set of visual presets with a preview you can hover to watch. WaveSpeedAI says the studio is available now in the generator sidebar, right after AI Chat. The panel is the product: instead of guessing the words for a look, you click the look.

How the Director's Panel Maps to a Shot

Each card controls one part of the frame. Film setup covers genre (Epic, Drama, Noir, Comedy, Horror, Action), era (1960s to the present day) and tempo, which runs from a single continuous shot to fast, chaotic cuts.

The Camera card splits into a body and a lens. The body is modern digital, 35mm film, 8mm film or a DV camcorder. The lens is clean, anamorphic, vintage anamorphic, warm vintage or halation, and the aperture sits at f/1.4, f/4 or f/11. Color palette offers 17 named presets, from Neon Monsoon to Silver Gelatin, or you can build your own from an image or a color picker. Lighting covers silhouette, practicals, window light, overhead, contre-jour and soft cross light, and you can place up to four lights yourself, each with its own color, brightness, softness and direction.

Anything you leave on Auto is up to the model. WaveSpeedAI says you can open "Prompt sent to the model" at any time to see exactly what your choices added.

What the 18 Camera Moves Cover

Open the Camera card and switch to Movement. The 18 moves are static, push in, pull out, pan left, pan right, tilt up, tilt down, truck left, truck right, crane up, crane down, orbit left, orbit right, a full 360 degree orbit, dolly zoom, crash zoom, handheld and aerial pullback. Pick one and WaveSpeedAI says the model follows it from the first frame to the last, so you do not need camera jargon in the prompt. The blog shows the toolkit on a single scene, a lighthouse keeper climbing spiral stairs with a lantern in a storm, restaged as a crane up with Practicals lighting in a Noir look, and it describes the dolly zoom as the subject keeping its size while the background stretches away.

What the Studio Writes Before You Generate

Before generation, the studio rewrites your scene the way a director of photography would plan it. According to WaveSpeedAI, the rewrite adds a deliberate lighting design (warm practical lamps against cool ambient light, in its example), atmosphere such as rain, haze and dust, set and wardrobe detail, lens and depth of field, the camera move with its start and end framing, and a soundscape. The rewrite keeps your subject and action as written, follows every panel choice, and comes out in the language you use. WaveSpeedAI says expanding a scene into the detailed shot uses a small amount of LLM credit.

References and the First Frame

The References card is where you attach outside material. Add images, videos and audio clips as references, then name them in the prompt with an @, as in @image1. You can mark an image as the first frame to start the shot from it, save any upload or result to Assets to reuse a character or a location in later shots, and pull an earlier result back in with Use as reference. That is the workflow a multi-shot sequence runs on: fix a face or a place once, then call the same reference into the next shot.

Which Models It Runs and What It Costs

The panel drives four model families. ByteDance's Seedance 2.5 and its Turbo variant run 4 to 30 seconds at up to 4K (Turbo at 720p to 1080p) with native audio, and support generate, edit and extend. Seedance 2.0, in its Fast, Mini and Turbo builds, runs 4 to 15 seconds at up to 4K with native audio. Alibaba's Wan 3.0 and Wan 3.0 Prime run 2 to 30 seconds at up to 1080p with native audio and rich reference inputs. MiniMax H3 runs 3 to 15 seconds at up to 1080p, including ultra-wide 21:9 and tall 9:21 frames. Generate, Edit and Extend share one menu, you can generate up to four takes at once, and Recreate brings a finished result's settings back into the panel. Pricing is per video at the selected model's rate with no subscription, and the price shows on the Generate button before you start.

How the Rivals Let You Specify a Move

Take one question, how each tool lets you set a camera movement, and the field splits into three habits.

Kuaishou's Kling works like WaveSpeedAI, from presets. Its camera control guide, published November 24, 2025, lists six basic movements (horizontal, vertical, zoom, pan, tilt and roll) plus four Master Shots: move left and zoom in, move right and zoom in, move forward and zoom up, and move down and zoom out. Kling says the function lets you move the camera in the video frame with absolute commands, and set the extent of the movement by adjusting a displacement parameter. That is the same preset-panel idea as AI Cinema Studio, on a shorter list.

Runway's Gen-4.5 takes the opposite route: the move is text. Runway's own prompt guide recommends the shape "The camera [motion description] as the subject [action]", and its camera terms reference pairs each term, from macro to aerial, with a written prompt and a sample output. The example Runway uses on its Gen-4.5 research page opens with: "Camera is panning and trucking across a kitchen table with different objects placed on the counter." There is no bank of named buttons, so the vocabulary lives in your prompt.

Google's Veo 3.1 in Flow also puts the move in the prompt, and adds a second lever: first and last frames. Google's blog on the Flow update describes controlling the shot from start to finish by providing a starting and ending image, and the Veo API docs list frame-specific generation as a feature. Google pins the endpoints of a shot; it does not offer a menu of moves.

Luma's Ray3.2 controls motion through keyframes instead of a move list. Its product page says Multi-Keyframe lets you set up to 16 keyframes inside a single clip, and its announcement says creators can choreograph exact narrative beats, camera paths and visual progressions to match their storyboards and client expectations. You place the motion in time rather than choosing it from a menu.

What It Means for Creators

A pick-a-move panel lowers the barrier to a deliberate camera. The 18 moves cover the standard vocabulary, so a creator who does not know a truck from a dolly still gets a clean one, and the studio writes the start and end framing for them. The trade is range: a named list cannot produce a move that is not on it. A prompt-driven system like Runway or Veo can chase any description you can write, at the cost of a less predictable result, and keyframe control on Luma sits between the two, more precise in time and more work up front. What to watch next is whether preset lists keep growing, and whether first-frame and @-reference workflows become the common way to hold a character across shots.

AI Cinema Studio is live in the WaveSpeedAI generator sidebar now. Open the Cinema Studio page, describe your scene in plain words, then set the film look and a camera move and press Generate, which shows the selected model's price before you start.