38 lenses.

One invoice.

Up to 30 seconds in one unbroken shot — the longest we run. Reads up to 30 reference images, lip-syncs to a voice you upload, and renders 1080p in 10-bit HDR.

Google's flagship. Native audio, strong physics, and the cleanest motion of anything here.

Coherent long takes with sound. Portrait or landscape, four to twelve seconds.

Fifteen seconds at 1080p, and the most permissive of our engines with reference photographs.

Built around reference images — up to seven — for keeping a character consistent shot to shot.

MiniMax H3 through their own door — expressive character motion, any length from 4 to 15 seconds at either tier.

Fifteen seconds at 1080p with a proper negative prompt. Good value for longer work.

Runway's flagship — precise camera language, any length from two to ten seconds.

Ray 3.2 through Luma's own door — smooth, natural motion from a prompt or a still. Five or ten seconds, up to 1080p.

Black Forest Labs' own door. Eight shapes, from square to 9:16.

The quick FLUX. Nine shapes, the widest choice of any image engine here.

xAI's own door. Seven shapes.

The slower Grok pass, for when the first one is nearly right. Seven shapes.

Kling's newest stills, at 1K or 2K for the same price. Eight shapes, and it takes a reference picture.

Kuaishou's own door, newest of the three. Takes any shape the picker offers.

Reads up to six reference images and holds a product or a face across every one. The workhorse behind Product and Cast.

Black Forest Labs' highest-fidelity image model. Fine detail and text that stays readable.

Strong composition and colour, and unusually good at following a long, specific brief.

Alibaba. Up to 15 seconds at 1080p, from a written prompt. 55 credits for five seconds.

ByteDance. Up to 15 seconds at 1080p, from a prompt or from a still you hand it. 95 credits for five seconds.

Vidu. Up to 8 seconds at 1080p, from a prompt or from a still you hand it. 130 credits for five seconds.

Vidu. Up to 5 seconds at 1080p, from a written prompt. 130 credits for five seconds.

Vidu. Up to 8 seconds at 1080p, from a prompt or from a still you hand it. 160 credits for five seconds.

Kuaishou. Up to 15 seconds at 1080p, from a prompt or from a still you hand it. 260 credits for five seconds.

Your photo, performing the movement of a clip you upload. As long as the clip, up to 30 seconds.

Changes a video you already have — restyle it, relight it, take something out. Priced on the clip's own length.

xAI's own. One to fifteen seconds at up to 1080p with synchronised sound, and an exact last frame.

Scored to length — up to 5 minutes in one pass.

Original score from a written brief. Instrumental or with a vocal line.

Original score from a written brief. Instrumental or with a vocal line.

Original score from a written brief. Instrumental or with a vocal line.

Original score from a written brief. Instrumental or with a vocal line.

Scored to length — up to 4 minutes in one pass.

Scored to length — up to 30 seconds in one pass.

Ten stock voices, plus any voice you clone from your own recordings. Thirty-two languages.

One clear photo of a face becomes a presenter. Best for short lines.

Takes a clip you already have and matches the mouth to new audio.

One photograph becomes an actual mesh — a .glb for Blender, Unity or a game engine, plus a turntable render so you can see it without installing anything.