omniveo
// Defining the logical standard for Gemini Omni

From Random Pixels
To Physical Logic.

OmniVeo Labs is the Engineering Benchmark for Generative World Models. We engineer logic-native video prompts, instruction-based editing templates, and frame-accurate deconstructions. Engineered for precision, not merely curated.

Who it's for: prompt engineers, AI researchers, and video-generation enthusiasts exploring logic-native world models.

Blackboard Study ↓Remix Template ↓
model
Omni-Model
build
2026.05
templates
5 Templates
cases
7 Case Studies
getting started

Pick your entry point

Whether you are prompting for the first time or evaluating frame-level logic, start here.

Beginner

Start with Templates

Copy-ready prompt scaffolds for instruction remix, symbol continuity, and scene edits.

Browse templates →
Advanced

Watch Case Studies

Frame-accurate deconstructions with Logic Score panels, TL;DR summaries, and observations.

View cases →
Deep dive

Read Research

RFC-style notes on temporal coherence, physics benchmarks, and world-model architecture.

Open research →
latest / benchmarks

Recent updates

Model versions, new case studies, templates, and research notes as they ship to the lab.

case_study / 014 · blackboard_derivation

Professor at the Blackboard — Frame-Accurate Deconstruction

Three observation windows isolate Veo 4 / omni-model's symbol continuity, spatial anchoring under occlusion, and logic-native layout planning.

rec · omni-trace
SEC: 0.00
Model: fbard_eac_video_generation_omni
Resolution: 1280×720
FPS: 24
Logic Score: 9.8 / 10
./timeline.log
./agent.stream
[00:00]SystemTokenizing handwritten LaTeX-equivalent glyphs...
[00:00]Agent Loading chalk-on-slate texture priors.
[00:00]OK Stroke order matches handwritten temporal prior.

Playback is synchronized with the timeline and agent trace.

framework / instruction-remix

Three-Stage Agentic Instruction Sequence

A canonical prompt scaffold for omni-model's native multimodal understanding — base, modify, refine — without falling back to pixel substitution.

01 · base

Define the initial visual scene

> A city composed entirely of fiber-optic strands, suspended in a
  windless midnight atmosphere. Static establishing shot, anamorphic lens.
02 · modification

Inject a dynamic logic variable

> The fibers begin to pulse like blood vessels — rhythm locked to a
  human breathing cadence (0.25 Hz). Pulse propagates from the city core
  outward.
03 · refinement

Constrain environment & physics

> Remove all background noise. Recolor to deep indigo (#0A1530).
  Maintain physical consistency: tension, gravity, optical bloom unchanged
  across edits.
./compiled_prompt.txt
role: senior_prompt_engineer @ omniveo
target_model: omni-2026.05
subject: <SUBJECT>

stage[1].base        := scene(fiber_city, static_shot)
stage[2].modify      := dynamic(pulse, sync=breath@0.25Hz)
stage[3].refine      := constrain(noise=0, palette=#0A1530,
                                  physics=preserve)

assert: native_multimodal_understanding == true
assert: pixel_replacement              == false
research / open notebooks

Recent technical notes