30 video engines · 22 image routes

The roster.
Route the shot, keep the Cast.

Every serious model, on one call sheet. Each family below is live in the composer: same Cast, same VX wallet, exact cost quoted before every run. When a new model ships, it joins your pipeline instead of replacing it.

Open the composer
01

Video families

MOTION
ByteDance · BytePlusCoreSeedance 2.5Live on the roster: one continuous take up to 30 seconds, with 30 reference images behind it.VIDEOByteDance · BytePlusCoreSeedance 2.0Every generation of one model family, from prompt probe to native 4K master.VIDEOGoogleVeo 3.1Google video with native audio, in three grades from the finished take to the fast draft.VIDEOGoogleGemini Omni FlashGoogle’s multimodal video model: generate from text, frames or references, or edit a clip by instruction.VIDEOLightricksLTX-2.5Lightricks video up to 4K at 50 fps, 20-second takes, and shots cut to an audio track.VIDEOKuaishouKling 4.0Kling 4.0 takes: any length from three to thirty seconds, from text, frames, references or a clip.VIDEOKuaishouKling 3.0Kling 3.0 up to native 4K, plus 3.0 Omni, 3.0 Turbo, 2.6 and 2.5 Turbo.VIDEOxAIGrok ImagineFast, per-second video and two-speed image generation from xAI.VIDEO & IMAGEAlibabaHappyHorseAlibaba video at native 1080p — and on 1.1, with sound.VIDEOAlibabaWan 3.0One take up to thirty seconds, built from stills, footage and sound at once.VIDEOAlibabaWan 2.7Alibaba’s reference-driven family: sound, an end frame, five refs — and stills.VIDEO & IMAGEMiniMaxMiniMax H3Up to 2K, refined from a native 768p render, in one take of up to fifteen seconds with sound.VIDEOMiniMaxMiniMax H3 MaxA post-trained MiniMax H3 tuned for prompt adherence, with camera paths and a Turbo grade.VIDEOBlack Forest LabsFLUX 3 VideoBlack Forest Labs moves to video, with sound in the same pass.VIDEOBlack Forest LabsFLUX 3 MaxA post-trained FLUX 3 Video, tuned to follow the prompt and to render sooner.VIDEO
02

Image families

STILLS
ByteDance · BytePlusCoreSeedreamSix image variants on flat VX pricing, led by Seedream 5.0 Pro.IMAGEOpenAIGPT Image 2.5Two OpenAI siblings priced the same — Sunburst for fidelity, Flare for speed.IMAGEOpenAIGPT Image 2The previous OpenAI generation, with its earlier grades kept on the roster beneath it.IMAGEGoogleNano Banana ProGoogle’s professional image tier — photographic realism at 4K.IMAGEGoogleNano Banana 2The Google workhorse: 4K output, fourteen references, and type that stays legible.IMAGEBlack Forest LabsFLUX.2The art-direction favourite, in Pro and Max grades.IMAGERecraftRecraft V4.1Recraft V4.1 for brand and product images, with type set from the prompt.IMAGE
Retired from the rosterSora 2 · OpenAI · retired 22 September 2026
03

What a shot costs

RATE CARD

A 5-second text-to-video shot on each first-party engine, and the per-image price of the three image flagships. Every figure is computed from the same pricing engine the composer quotes from, so what you read here is what you are charged. Cost scales with duration and resolution, and the exact VX is always shown before a run.

EngineBest forVX · 5s shot
Seedance 1.5 procheapest engine for look-dev.5 VX · 720p
Seedance 1.0 liteentry probe for prompt testing.6 VX · 720p
Seedance 2.0 Minieconomy 2.0 lane. 720p, joint audio-video, full multimodal refs.7.5 VX · 720p
Seedance 2.0 fastproduction-default speed/quality balance.8.5 VX · 720p
Seedance 2.0native-4K masters, takes up to 15s.11 VX · 720p
Seedance 2.530s single-take, joint audio-video. Up to 1080p, native 4K on Enterprise.16.2 VX · 720p
Image economics

Premium image models, a fraction of the price.

Seedream 5.0 Pro, our flagship, runs 1.3 VX per image, as low as $0.11, priced the same whether you generate one or a thousand. The same wallet buys the other flagships.

1.3 VXSeedream 5.0 Pro · from $0.112.36 VXNano Banana Pro · from $0.203.68 VXGPT Image 2.5 Sunburst · from $0.31

Dollar figures at our best self-serve rate ($0.085/VX at Studio Scale). The exact VX cost is shown before you run.

Guest videoGrok VideoGrok Video 1.5VeoVeo 3.1 FastVeo 3.1 LiteGemini Omni FlashLTX-2.5 FastLTX-2.5 ProKling 2.5 TurboKling 2.6Kling 3.0 TurboKling 3.0Kling 4.0Kling 3.0 OmniFLUX 3 VideoFLUX 3 MaxHappyHorseHappyHorse 1.1Wan 3.0Wan 3.0 PrimeWan 2.7MiniMax H3MiniMax H3 MaxMiniMax H3 Max Turbo
ImageSeedream 4.0Seedream 5.0 LiteSeedream 5.0 ProSeedream 4.5Seedream 5.0 FlashSeedream 3.0GPT Image 2.5 SunburstGPT Image 2.5 FlareGPT Image 2GPT Image 1.5GPT Image 1Nano Banana ProNano Banana 2Nano Banana 2 LiteGrok Image · QualityGrok Image · FastFLUX.2 ProFLUX.2 MaxWan ImageWan Image ProRecraft V4.1 ProRecraft V4.1 Utility Pro

You are refunded down to the clip you got

Where an engine decides the length for itself, the quote holds its maximum so the run can never overdraw you, then refunds the difference against the clip actually delivered. You are charged for the seconds you received, not the ceiling they were quoted at.

Aspect ratio can follow the reference

Set the aspect to auto and the shot takes its shape from the reference you gave it, across the line. A vertical plate makes a vertical shot without a second setting to remember or a crop to fix afterwards.

04

Identity lives in the Cast, not the model

WHY MULTI-ENGINE

Single-model tools tie your campaign to one vendor's roadmap. On VisionX, the Cast carries identity across every family above, so an engine changing or retiring never strands your work. Compare takes across families, keep what wins the edit.

FAQ

Before you pick a model

Which model should I use for a given shot?
There's no single best answer. Compare takes across families for the same shot and keep what wins the edit. Each family's page covers what it's actually good at, and the composer shows the exact VX cost before you commit to a run.
Do all models cost the same VX?
No. Cost varies by engine, mode, and resolution, and the exact figure is shown before every run, never estimated after the fact.
What happens if a vendor discontinues a model?
Identity lives in the Cast, not in any model. A retired engine or a vendor's API change never strands a campaign, because every other family on the roster still resolves to the same verified identity.
How is this different from using Veo, Kling or Wan directly?
Those are single engines. VisionX routes the same Cast across all of them plus the Seedance engines from one wallet, so a Kling shot and a Seedance shot sit in the same sequence without the character changing.
Is every model available in my workspace?
Guest engines are switched on per workspace by an admin. A model not yet enabled shows as a labeled, locked card in the composer rather than disappearing, so you always know what exists on the roster even before it is turned on for you.

22 families, one call sheet.

Start free with 20 VX on signup, no card required.