ZENOS Standard 1.0

The new gold standard for game data.

Most world models learn from pixels. Under every pixel of a shipped game sits what the engine knew: depth, identity, state and intent. We capture all of it, on one clock, licensed from the studio.

Licensed at the sourceStudio paid on every saleNo SDK, no dev build, no source code

What every capture carries

30 Hzone tick grid, every stream
7buffers, 1920 × 1080
48 kHzaudio, locked to the grid
60 Hzinput readings
384label classes across 97 labels
3provenance levels
Everyfile carries its licence ID
The problem

Pixels make a model guess what the engine already knew.

Most gameplay data is video and a log of key presses. A model trained on it has to guess depth, identity and state before it can learn how the world works, and some of what it needs was never recorded.

The engine computed every one of those answers exactly, on every frame, and threw them away.

What video and inputs cannot teach a model

Missing

No state, so no cause.

A model sees a hostile fall. It never sees its health, its AI state or the player’s reload timer, so it has no way to learn why anything on screen happened.

In the Standard: 04 World state
Guessed

Estimated depth, inherited error.

Depth, motion and camera pose have to be estimated from the picture. Every error the estimator makes becomes training data.

In the Standard: 01 Rendering · 08 Camera and 3D
Ambiguous

Key presses, without what they did.

A left click is Fire in a fight and Click in a menu. A log of buttons records the press and leaves the model to work out which one the player meant.

4.2× better latent-action learning from true action labels on 2.5% of the data, where distractors were present. Nikulin et al. · ICML 2025 [1]In the Standard: 03 Inputs
Burned in

HUD and audio, already mixed.

The HUD is drawn into every frame and every sound is mixed into one track. A model learns the health bar as part of the world, and neither can be pulled apart afterwards.

+18.8% motion quality from the same clips without HUD overlays. Shen, Jin, Chen · arXiv 2026, preprint [2]In the Standard: 01 Rendering · 02 Audio

All four are in the Standard.
Train against the engine’s answers, not guesswork.

Third-party results, cited in full below. None is a measurement of ZENOS data.

The standard

The gold standard, defined.

ZENOS Standard 1.0 sets nine properties, one for each chapter below, and each one you can open in the files and check. Hold any gameplay dataset to the same list.

ZENOS Standard 1.0
PropertyTypical gameplay dataZENOS Standard 1.0Every capture, every tick
01RenderingThe final frame, HUD burned inEvery buffer the engine drewHUDPre-HUDNormalsDepthMotion
02AudioOne mixed trackThe game's own mix, splitMaster MixMixer-Category Stems48 kHz OpusTick-Locked Samples
03InputsKeys and mouse, or actions inferred from videoEvery press, and what it meantKeyboardMouseGamepads60 Hz ReadingsGame ActionsAction States
04World stateNone, or read off the HUDEvery entity, on screen or offEntity ClassTransforms in MetresSource ValuesEvery Tracked Entity
05LabelsGenerated by a model, rules unseenA locked, traceable ontology384 Label Classes97 Labels16 DomainsDefinitionsRulesProvenance
06SegmentationNone, or predicted by a modelEvery object, separated and classedInstance IDsSemanticPanopticBoundsObject CountsOcclusion
07Pixel to entityNoneEvery pixel resolves to its entityPixel-to-Entity LookupPer-Tick Entity StateCross-Frame Identity
08Camera, 3D and multiviewUnknown, or estimated; one view at a timeCalibrated on every tick, synced across viewsfx · fy · cx · cyCamera PoseProjection TermsMultiview Sync
09ProvenanceA licence claimAuditable to the fileEmbedded Licence IDContent Validation
Met in full0 of 99 of 9

metpartly metmissing

How we capture it. Every row above comes from the shipped build, with no studio integration and no lead time. PRISM reads the game while people play it, so any licensed title can meet the standard.No SDKNo Dev BuildNo Source CodeAny Engine

Why it is safe to train on. Every title is licensed from the studio that owns it, under terms both sides can point to.Per-Title LicenceStudio Revenue ShareNo-Recreation Clause

01Rendering

Every buffer the engine drew, pixel-aligned.

Seven buffers from one render, each 1920 × 1080 at 30 frames a second and never scaled or cropped. HUD over Pre-HUD is the frame the player saw.

Depth, Segmentation, the HUD matte and the Normals mask are lossless, so what you decode is exactly what the engine wrote. Pre-HUD on its own is the frame with no interface on it, the footage the HUD-free study above trained on [2].

example frame · The Desolation · one tick, every bufferselect a buffer
same pixel, every buffer
composited every buffer

HUD

Every element of the interface as colour plus an exact alpha matte, so a model can learn the screen with or without it.

CharactersProps

filergba_hud.mp4hud_colorHEVC 4:4:4, 8-bit, constant QP 26hud_alpha8-bit, lossless
02Audio

Every sound on its own track, on the pixels' clock.

The master mix plus one stem per category of the game's own mixer, recorded from the engine rather than separated afterwards. Every file is locked to the capture grid: tick k starts at sample 1,600 × k, so any sound can be read against the frame and the state of the same tick.

example · The Desolation · 8 s · master mix and five stemssolo and mute any stem

tick 48,653−4.00 s

ambience
effects
weapons
interface
music
−4 s−2 stick k+2 s+4 s
example tick

A game's stems are its mixer's named buses, or categories it declares by keyword. Either way the split is the engine's own, which is why a model can hear the weapon, the world and the interface as different things.

Spec
Codec
Ogg Opus, 48 kHz, mono or stereo, 192 kb/s VBR
Files
audio_master.opus, plus audio_stem_N.opus per category
Clock
sample n sits at the start fence + n / 48,000 s
Placement
master within 5 ms of the grid; stems within 10.7 to 25 ms, by audio engine
03Inputs

The press, the intent, and the tick it landed on.

What was pressed and what it meant, on the same tick as the frame. Keyboard, mouse and up to four gamepads are read twice per tick, and beside them sits the game's own action system, recorded after its key bindings: every action, its value, and whether it was idle, pending or active.

example · The Desolation · 10 s of input · recording playerscrub to any tick
LR

raw delta−3, +1counts, two readings summed

YXBALTRT

Gamepads in XInput slots 0 to 3 are read the same way, with sticks and triggers as values. This player used keyboard and mouse, so the pad is idle here.

held on this tickheld across these 10 s

What the game understood

actiondefault bindingvaluestate MoveW A S D1activeFireleft mouse1activeAimright mouse1activeTurnTowardsByMousemouse movement1.41activePointerPressleft mouse1activeSecondSkillE0idle

Actions are read after the game's own key bindings. Rebind Fire to any key or button and this row still says Fire.

tick 48,773 12:09:10.033 UTC

No press lost

Every device is read twice per tick. A key down for a single 1/60 s reading still lands in that tick's row, and mouse motion is summed from raw counts, never sampled, so a flick is never dropped or counted twice.

Survives a rebind

Actions are the game's own reading of the player, after its bindings, so intent stays readable however the controls were set. Each action's default bindings ship in the game's data beside it.

On the frame's tick

Input row k is frame k of every video and state file. No offsets to estimate, no alignment pass between what was pressed and what was drawn.

Spec
Readings
60 Hz, two per 30 Hz tick, on the capture grid
Devices
keyboard, mouse, gamepads in XInput slots 0 to 3
Mouse
buttons, raw delta, wheel, cursor in the game window
Actions
name, value type, value, state: idle, pending or active
04World state

The whole world on every tick, including what the camera can't see.

Every entity the game tracks, on screen or off, with its world transform, its class and its exact kind. Every field records the game object and path it was read from, so any value traces back to the engine. Positions ship twice: as the game stored them, and normalised to metres on one right-handed, Y-up frame, so titles from different engines line up.

example · The Desolation · 10 s, top-downselect an entity
camera · squad 1 · 17 m upsquad 1squad 2squad 3bosshunter10 m

Squad 1 (recording)

UniqueID356e4d9833216d00Health2,400 / 2,400Shields1,540 / 1,540PrimaryWeaponNameUca_SMG_01_aAmmoClip10 / 28IsManualFiring1MovementState2position (m)413.70, 0.00, -6.22

The camera keeps the boss in view for all 10 s, and the state still holds five breeding eggs outside it. On tick 48,782 one egg’s Health drops to 0 and a hunter enters the state beside it, off screen. Trails show each entity’s previous 10 seconds, fading with age. The camera hangs above and behind squad 1 looking down: the light wedge is its view from above, the shaded area the ground it sees on the selected tick.

Spec
Format
VTX, the game's own buckets and schema in the file header, Apache-2.0
Rate
30 Hz, the newest game state at each tick
Frame
right-handed, +Y up, +Z forward, metres; the game's value beside it
Rotations
quaternions when normalised; Euler stays Euler in its source form
Every entity
world transform, class identifier, instance identifier, 64-bit id
05Labels

Labels that show their working.

One locked vocabulary for every title: 384 label classes across 97 labels in 16 domains, each label with a written definition. Values are computed from captured state by stated rules, never by watching the pixels. Any value traces back to the game state it was computed from and the rule that computed it, on any tick.

example · The Desolation · 10 s of labels · squad member 1click any moment on any row
running
firing
engaged
false
low
full
full
alive
active
combat_zone
tick 48,623300 ticks · 30 Hztick 48,922
Frame of tick 48,623Frame of tick 48,638Frame of tick 48,653Frame of tick 48,668Frame of tick 48,683Frame of tick 48,698Frame of tick 48,713Frame of tick 48,728Frame of tick 48,743Frame of tick 48,758Frame of tick 48,773Frame of tick 48,788Frame of tick 48,803Frame of tick 48,818Frame of tick 48,833Frame of tick 48,848Frame of tick 48,863Frame of tick 48,878Frame of tick 48,893Frame of tick 48,908
tick 48,773 · 12:09:10.033 UTC · video frame 49,711
movement.state

Gait = running

Horizontal gait tier.

valuesidle · walking · running · sprinting

Provenance on this tick

computed · one game field, its values mapped onto the ontology's own

MovementState 0 → idleMovementState 2 → runningMovementState 3 → sprinting
fieldthis tickpreviousPlayer.MovementState20

Read from data_game_state.vtx at the same tick. Stamped with ontology 3.33.1 and band set blake2b:51c5753d….

Every value on these bars is computed from that tick's game state by the rule shown, with V1's band edges, so the same inputs give the same value on every run. A label made by a model can change with the model version, the prompt or the context, and nothing records why. A rule can be changed once and every hour already captured re-derives from its stored state, with no recapture.

provenance, per labelwhat a value rests on

AssertedSet by a person in the game’s label profile, for a fact that holds across the whole game, such as every level being outdoors in daytime. Fixed for the title, it never changes during play.

ComputedDeterministic logic over captured values: a game field mapped onto the ontology’s values, bands, thresholds, comparisons, deltas, state machines.

InferredAnything probabilistic, from a model or vision. Named as such, never mixed in silently.

coverage, per captureno silent gaps

ApplicableThe game has the system and this capture labels it, with the rule written beside it.

Not capturedThe game may have it, but this capture does not read it. Listed, so you know.

Not applicableThe game has no such system, so the column is empty.

a banded labelvitals.health.state
0.15.35.65.95

critical · low · medium · high · full, from health as a fraction of its range. Not written while the entity is dead or respawning, so a corpse never reads as critical.

Browse the full ontology · 384 label classes, 97 labels, 16 domains
label ontology V1 · version 3.33.1 · lockedselect a domain, then a label

The condition of the entity itself, health, armor, shield, stamina and other gauges, life state and damage taken.

vitals.health.state

Health

Discrete health band derived from health.percent. There is no dead band: death is vitals.life_state = dead, and 0% health bands as critical.

valuescritical · low · medium · high · fulltypeenum · ordered · bandedapplies toany entity

not written when vitals.life_state is dead or respawning

65 enum labels hold one value at a time; 32 boolean labels are independent flags. 18 are bands of a measured number, with edges fixed by the ontology so a band means the same thing in every title.

06Segmentation

Every pixel knows what it is, and which one.

Instance, semantic and panoptic segmentation from the engine's own object-ID pass, with Bounds drawn from the same IDs. Every object in the scene is separated, classed and boxed on every frame, stable from one frame to the next and exact at the pixel.

example frame · The Desolation · one tickone ID pass, four views
A bunker corridor from above: the player, three hostiles and the props around them The same frame as instance IDs: every object a flat, distinct colour The same frame with a bounding box around every character, prop and pickup The same frame with every pixel coloured by its class
Every flat colour is one object, from the engine’s object-ID pass. Walls and floors are segmented too.

Game entitiesProps

At training resolution · objects kept

Instance IDs at 1920 by 1080
113 1920 × 1080
Instance IDs at 455 by 256
109 455 × 256

World models usually train at 256 on the short side. Of the 113 objects in this frame, 109 keep their own ID at 455 × 256, and 109 squashed to 256 × 256.

The structure of the scene

What is where, how many, which one, and what is in front of what: the structure of the scene, separated object by object and held at low resolution.

A judge on every frame

Masks, boxes and counts with exact answers on every tick, so a model's own predictions can be scored automatically. The ground truth for self-improvement loops, at any volume.

Boxes, counts and occlusion for free

Bounds, object counts and what hides what all fall out of the same IDs. No annotators, no detector to correct.

07Pixel to entity

Point at a pixel. Read the entity behind it.

Every pixel resolves to the entity that drew it, and every entity resolves to its full state on that tick. The instance ID under the pixel is the key: one lookup turns a patch of colour into a player, an enemy or a door, with its health, weapon, pose and every other property the game holds. Vision and state arrive as one table, already joined.

example · The Desolation · 4 s of playplay · point at any object · click to pin
A bunker corridor from above: the player, the hostiles and the props around them
tick 550 · 21:28:00.074 UTC · 1920 × 1080

Masks that mean something

Every mask carries the entity's class, health and weapon on that tick.

Identity across frames

Instance IDs hold from frame to frame, so tracking needs no association step and no re-identification model.

Supervision from the answer

Train a model to read state from pixels with the true state under every pixel, on every tick, in every title.

Point at any object · click to pin it here

1 · Pixel to instance
pixel(951, 663)
instance id26,715 px
2 · Resolves to Game entityPlayer id 9223372041149743518
3 · Its state on this tick
movement.phase2
speed (m/s)3.51
weapon_handling.manual_firing1
weapon_handling.reload_time_left (s)0.00
position (m)-537.48, 40.00, -8.45
+ more fields in the game state …
4 · Its labels on this tick
movement.staterunningcombat.stateengagedcombat.actionfiringcombat.weapon.ammo_statelowvitals.health.statefullaim.stateprecise
+ more labels on this tick …
08Camera and 3D

Three players. Three machines. One world.

Every tick carries the pixel intrinsics and projection of the camera that drew it, and the player camera's pose in metres. With them, depth becomes distance and every pixel goes back to where it stood in the world. Every tick is also placed on UTC from public time servers, so captures of one match recorded on different machines, in different sessions, line up to the tick and land on the same ground.

example · one moment, three players' viewsthree captures, three machines
A boss arena rebuilt as a point cloud from three players' depth and camera data, with their three camera frustums The same point cloud coloured by the view each point came from camera 1fov 34° vcamera 2fov 34° vcamera 3fov 28° vsquad 1HP 2,400squad 2HP 2,400squad 3HP 858
fx fy cx cypixel intrinsics on every tick, ready as an OpenCV camera matrix
metrescamera pose, right-handed, Y up
0.066%depth precision per code step, at any range: 1.3 mm at 2 m
UTCevery tick, from public time servers

One moment rebuilt from the depth and camera files of three captures on three machines, each point coloured by its view's pre-HUD frame. If the intrinsics, poses or depth were off, the three views would not land on the same ground.

synced across sessions · the same moment as abovethree captures, one clock
UTC · public time12:09:10.033 UTCsquad member 1machine A · its own session and clockframe 48,773squad member 2machine B · its own session and clockframe 49,699squad member 3machine C · its own session and clockframe 49,730

Three players in one match, each recorded on their own machine in their own session. Each machine's clock is checked against time.cloudflare.com and time.aws.com through the capture (diamonds: one probe every 37.5 s), and every tick carries its UTC instant with the uncertainty stated. The same moment is frame 48,773 in one file and 49,699 and 49,730 in the others, and they line up without matching sound or pictures. Bar starts and probe marks are drawn to illustrate; the frame numbers are real.

Stamped against public time

Two independent public time sources, asked throughout every capture. Each tick is stamped in UTC with its own uncertainty, so a machine's clock error never reaches the data unmeasured.

Any two captures line up

Different machines, sessions and start times land on one timeline. No clapperboard, no audio matching, no visual alignment pass.

Required for multiview

The standard requires every capture of a shared match to be syncable this way, so several players' views join into one scene with a lookup.

09Provenance

Every file accounted for. Every title licensed.

A capture is a folder you can audit. The licence travels inside every file, every stream is checked for real content, and only a capture with a passed verdict is delivered. Nothing is stitched: every boundary is kept.

chain of custody · every capturedataset contract 2.0

1 · Licence

Signed with the studioPer title, directly with the studio that owns it. The right to train is written into the contract, never assumed from a platform's terms of service.

2 · Stamped

The licence travels with the fileLicence, capture and game IDs are written inside every MP4, Opus, VTX and Parquet file, so a file copied out of its folder still says where it came from. Real player names reach the data only masked or pseudonymised.

3 · Verdict

Every stream checked for real contentBlack, frozen, silent and flatlined streams are measured, each stream's duration is checked against the others, and every hash is recomputed. A failed capture is never delivered.

4 · Alignment

Every frame explainedOne row per tick: capture time, UTC from public time servers with its uncertainty, the game's own clock, and each stream's frame state and content age. A repeated frame always says why.
data_alignment.vtx · what every frame statesno unexplained frames
newnewer content than the frame before repeated · no_new_entry_in_timethe source produced nothing new by the tick repeated · fell_behind_other_streamsa video showed an older game frame than the others repeated · entry_invalidthe entry could not be used; the last good frame repeats repeated · entry_lost_before_copythe entry was overwritten before it was copied repeated · entry_copy_not_finishedthe copy was still running at the tick emptyno content yet, before a stream's first entry nothing drawnthe game drew nothing to that buffer in the frame content agehow old a frame's content is at its tick, in 100 ns units

For every video, new + repeated + empty + nothing drawn equals total. A frame of default values is never written.

Studios paid on every sale

A 50/50 net revenue share with the studio. Non-exclusive.

No recreation

Labs are contractually barred from training models to recreate the game.

Download

Open a minute of it.

One minute of The Desolation with every layer on this page, a readme and a notebook that opens it. Tell us who you are and we will email you your own download link, with an evaluation licence attached.

desolation-minute.zip2.0 GB
desolation-minute/
  README.md
  open_the_minute.ipynb
  LICENSE.md                      evaluation licence
  capture-1794415146022/          player one’s view
  capture-1794415239971/          player two, same moment
    each capture folder holds
    manifest.json                 provenance, hashes
    manifest_system.json          capture machine
    rgb.mp4
    rgb_prehud.mp4
    rgba_hud.mp4
    depth.mp4
    normal_screenspace.mp4
    motion.mp4
    segmentation_instance.mp4
    segmentation_semantic.mp4
    segmentation_panoptic.mp4
    audio_master.opus
    audio_stem_N.opus             22 stems
    data_game_state.vtx
    data_camera.vtx
    data_input.vtx
    data_alignment.vtx
    data_system_health.vtx
    data_segmentation_instance.vtx
    data_segmentation_semantic.vtx
    data_segmentation_panoptic.vtx
    labels.parquet

Request the minute

Required. By requesting the minute you accept the evaluation licence.

Received

Your request is in.

Thanks for asking for the minute. Your download link is on its way to the work email you gave us. It is yours alone and works for 14 days.

Need more than a minute? Request the hour →

ZENOS Evaluation Licence

Version 1.1, 12 November 2026

Copyright (c) 2026 Zenos Interactive Limited and its studio licensors. All rights reserved.

This Licence is granted by Zenos Interactive Limited, a company registered in England and Wales with company number 12603895 ("Zenos"), to the organisation that receives the Sample Data from Zenos or from a partner authorised by Zenos, and to its affiliates ("you"). The titles and contents of the sample are listed in the readme or sample manifest that accompanies it.

By accessing or using the Sample Data you accept this Licence. If you do not accept it, do not use the Sample Data, delete any copies, and tell Zenos at [email protected].

1. Definitions

Sample Data means the files Zenos provides to you under this Licence, and any copy or extract of them.

Derived Artefact means any model, checkpoint, weights, embeddings or similar artefact produced from Sample Data.

Evaluation Results means your metrics, benchmark scores, analysis and notes about the Sample Data, provided they contain no Sample Data and no Derived Artefact.

Studio means a studio or publisher that owns or controls a game from which Sample Data is derived.

Public Title means a game that Zenos has itself named publicly as a source of the Sample Data, for example on zenosdata.com or in the readme that accompanies the sample.

Evaluation Period means 90 days from the date you first receive the Sample Data.

2. Grant

Subject to this Licence, Zenos grants you a non-exclusive, non-transferable, non-sublicensable, revocable licence to access, store, reproduce and process the Sample Data during the Evaluation Period, solely to evaluate whether Zenos data is suitable for your artificial intelligence research and development.

3. Permitted use

You may:

  1. inspect the Sample Data and analyse its quality, structure and coverage;
  2. test it against your data pipelines and tooling;
  3. run evaluation, benchmarking, and trial training or fine-tuning experiments on it, provided every Derived Artefact is used only for that evaluation and is deleted by the end of the Evaluation Period, unless you have by then licensed the data from Zenos;
  4. keep your Evaluation Results, provided they identify Studios and games only by genre and data type, or by name where the game is a Public Title;
  5. share the Sample Data with your employees, contractors and affiliates who need it for the evaluation and are bound by confidentiality.

4. Restrictions

You may not:

  1. use the Sample Data to train, fine-tune or improve any model that is deployed, released, made available to any third party, used in any product or service, or retained beyond the Evaluation Period, other than as section 3(c) permits;
  2. publish, post, publicly display, demonstrate, broadcast or otherwise make available any part of the Sample Data or any Derived Artefact to any third party, including in any paper, presentation, marketing material, dataset repository, platform or marketplace;
  3. sell, license, sublicense, rent or otherwise commercialise the Sample Data or any Derived Artefact, or incorporate either into any dataset, catalogue, product or service;
  4. train or fine-tune any model specifically to reproduce a game from which Sample Data is derived as a playable experience, or to produce substantial or verbatim copies or adaptations of it;
  5. reverse engineer, decompile or attempt to derive the source code or shipped binaries of any game from the Sample Data;
  6. remove, obscure, alter or circumvent any notice, watermark, fingerprint, provenance metadata or access control in or applied to the Sample Data.

5. Studio confidentiality

Except for Public Titles, the Studios have not agreed to be identified publicly in connection with AI training. You may not make any public statement identifying a Studio or game in connection with AI training, or disclose the identity of any Studio or game in the Sample Data or in Zenos's catalogue outside your organisation, without Zenos's prior written consent. Zenos will ask the Studio if you request it; the Studio may refuse. You may describe the Sample Data by genre and data type. This section does not stop you naming a Public Title as the source of the Sample Data you evaluated.

The Sample Data and Zenos's catalogue are confidential to Zenos.

6. Personal data

The Sample Data is not intended to contain personal data. Capture is carried out by paid professional players using Zenos-controlled accounts and devices, and player names that appear in telemetry or labels are Zenos test accounts. Some sessions take place on public game servers where other players are present, and their in-game names may appear on the HUD in the video. Manifests include hardware and performance telemetry for the capture machines as part of the provenance record. The Sample Data contains no in-game chat, IP addresses or account credentials. Zenos does not warrant that the Sample Data contains no personal data. If you find anything that appears to be personal data, tell Zenos and both parties will cooperate to remove or anonymise it.

7. Term and deletion

This Licence ends at the end of the Evaluation Period, or earlier if Zenos revokes it for breach of section 4 or 5. When it ends, either license the data from Zenos or delete the Sample Data and every Derived Artefact, and confirm the deletion to Zenos by email on request. You may keep your Evaluation Results. Copies in routine backups may be retained while they remain confidential and are not restored to use. Sections 4, 5, 8 and 9 survive.

8. No warranty; no liability

THE SAMPLE DATA IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING WARRANTIES OF ACCURACY, COMPLETENESS, QUALITY OR FITNESS FOR A PARTICULAR PURPOSE. YOU USE IT ENTIRELY AT YOUR OWN RISK. TO THE FULLEST EXTENT PERMITTED BY LAW, ZENOS AND ITS STUDIO LICENSORS HAVE NO LIABILITY OF ANY KIND ARISING FROM THE SAMPLE DATA OR YOUR USE OF IT. Nothing in this Licence limits liability for fraud, or for death or personal injury caused by negligence, or any other liability that cannot be limited by law.

9. Reservation of rights

All rights not expressly granted are reserved. As between you and Zenos, all copyright, database rights and other rights in the Sample Data remain with Zenos and its Studio licensors. Use outside this Licence is unlicensed.

10. Relationship to other agreements

This Licence applies in addition to any agreement between you and Zenos, or between you and a partner authorised by Zenos, under which the Sample Data is provided. If you and Zenos enter into a Zenos AI Lab Data Licence Agreement covering this data, that agreement replaces this Licence from its effective date.

11. Governing law

This Licence is governed by the laws of England and Wales. The courts of England and Wales have exclusive jurisdiction, save that Zenos may seek injunctive relief in any competent jurisdiction to protect its intellectual property or confidential information.


Zenos Interactive Limited, 71-75 Shelton Street, London WC2H 9JQ
[email protected]

Questions

What researchers ask first.

Is this just video plus inputs?

No. Video and inputs are two of the nine rows in the standard. The other seven are what the engine knew.

How do you capture without an SDK?

PRISM reads the running game at the engine level, from the shipped build. No plugins, source code or development build required.

With whose permission?

The studio's. Every title is licensed from its studio, the studio earns from every sale, and no lab may train a model to recreate the game.

Which engines?

All of them, including custom in-house engines. PRISM reads the running game rather than plugging into an engine, so the engine is never the constraint.

Does every title carry every label?

No title has every system, so none carries every label. Each label table lists what applies to its game and what its capture read, so a gap is always stated.

Why 30 Hz, not 60?

Because 60 doubles what you store, move and decode, and we have not seen evidence that it trains better models. What matters at 30 is that nothing is lost. Every frame on the grid is accounted for, and every repeat states its reason. Input is read twice per tick, so even an action shorter than a tick keeps the key or button that fired it.

Is this a list of modalities?

No. A modality list counts every variant and every option a studio integration could unlock. The standard lists what every capture carries, from the shipped build, with no SDK and no lead time. Counted line by line, the nine rows hold about two dozen streams.

Can I train on the minute?

The minute is for evaluation, and its licence says what you can do with it. Training data comes under a licence per order.

References
  1. Nikulin, A. et al. (2025). Latent Action Learning Requires Supervision in the Presence of Distractors. ICML. arxiv.org/abs/2502.00379
  2. Shen, W., Jin, D. and Chen, D. (2026). Game2World Engine: Unlocking In-the-Wild Gameplay Videos for World Model Training. Preprint. arxiv.org/abs/2608.24680

Build on real worlds.

Building frontier AI?Talk to us →
Have a catalogue?List Your Games →