Skip to content
OniraOnira
Visual Continuity

Consistent visuals across every scene.

AI generation is powerful scene-by-scene. Without reference management, characters shift appearance, props change shape, and tone drifts between clips. Onira's Visual Continuity Engine keeps every character, setting, and visual tone locked — automatically.

Without references

Character drifts scene-to-scene

With Onira

Locked by reference images
The problem with unmanaged AI generation

Same character. Sixty different faces.

Ask Gemini Flash Image to generate “Dr. Elena Torres” sixty times without reference images and you'll get sixty different people. Hair colour shifts, eye colour changes, age varies, clothing disappears. The visual incoherence breaks immersion immediately. Reference-anchored generation is the step that turns a collection of AI clips into a coherent film.

The problem

Why visual continuity is
hard in AI video.

Every generation call is stateless. Without explicit references, the model has no memory of what a character looked like in scene 1 when it renders scene 47.

Without Reference Images

Scene 01
Brown hair, blue coat
Scene 12
Black hair, grey coat
Scene 29
Blonde, no coat
Scene 47
Different face entirely

Each generation call is stateless. The same character prompt produces a different person every time.

With Onira Reference Anchoring

Scene 01
Consistent, reference-locked
Scene 12
Consistent, reference-locked
Scene 29
Consistent, reference-locked
Scene 47
Consistent, reference-locked

Every prompt is reviewed against the character's approved reference images before generation fires. Drift is corrected before it reaches the API.

How it works

The Visual Continuity
pipeline.

A three-stage process that anchors every image to approved references before generation, and inherits that look into animation.

01

Reference Ingestion

Upload character portraits (FRONT, THREE_QUARTER, FULL_BODY), VisualContext settings, and VisualSubject props. These are stored at org level and applied automatically to every project in your workspace.

Org-scoped · shared across all projects

02

ImageDirector Review

Before each image prompt reaches Gemini Flash Image, an ImageDirector LLM compares it against all active references. Prompts that would produce drift are rewritten. Only approved prompts reach the generation API.

Gemini 3.1 Pro · pre-generation gate

03

Image-to-Video Inheritance

The approved still frame is passed directly into Pixverse v6 as the source image for animation. Motion is generated from the approved look — lighting, colour, and composition carry through to the final clip.

Gemini Flash Image → Pixverse v6

Reference system

Every type of visual
consistency, covered.

Characters, environments, props, tone, narration voice — every dimension of visual identity has a reference type in Onira.

Character Portraits

Every recurring character in a project gets three org-shared reference images: FRONT, THREE_QUARTER, and FULL_BODY. Gemini 3.1 Pro reads these portraits before writing any scene that features that character.

Coverage: FRONT · THREE_QUARTER · FULL_BODY angles

CharactersConsistencyOrg-shared

VisualContext References

Project-level visual context locks in the overall look: tone, lighting style, environment palette, and production design. Every image prompt inherits these constraints automatically.

Coverage: Tone · Lighting · Environment · Palette

SettingsAtmosphereWorld-building

VisualSubject References

Props, vehicles, locations, and recurring objects get their own reference images. A spaceship looks identical in scene 3 and scene 47 because both prompts reference the same VisualSubject.

Coverage: Props · Vehicles · Locations · Objects

PropsContinuityScene-to-scene

ImageDirector Review

Before any image reaches the generation queue, an ImageDirector LLM reviews the prompt against all active references. Prompts that would produce drift are rewritten before the API call fires.

Coverage: Prompt review · Drift correction · Pre-generation

Quality gateLLM reviewGemini 3.1 Pro

Image-to-Video Inheritance

Pixverse v6 animates the approved still frame directly. The motion clip inherits the exact lighting, color, and composition of the reference image — animation cannot diverge from the approved look.

Coverage: Gemini Flash Image → Pixverse v6 pipeline

AnimationPixverse v6No drift

30+ Language Narration

ElevenLabs eleven_v3 voices narrate each scene. The same character voice is reused across every scene in the project, keeping audio identity as consistent as the visuals.

Coverage: ElevenLabs eleven_v3 · Consistent voice per project

NarrationElevenLabsVoice identity

Org-shared references

Character portraits and VisualContext settings are scoped to your whole organisation. Define a character once; every team member and every future project uses the same approved look.

Pre-generation quality gate

The ImageDirector LLM runs before the Gemini Flash Image API call. Drift is corrected at prompt level — no wasted credits on generations that would have failed visual review.

Animation that can't drift

Pixverse v6 animates from the approved still frame directly. The motion clip inherits the exact lighting and composition of the reference — there is no separate animation prompt to go wrong.

Output

What you actually get.

No references set

Character appearance changes every scene · No visual memory between generations

With Onira references

ImageDirector-reviewed · Reference-locked · Pixverse v6 animation inherits approved look

Reference anchoring applied across all scenes · Consistent characters, zero manual prompt-wrangling

Characters that stay
consistent, every time.

Set your references once. Onira's Visual Continuity Engine keeps every scene, character, and setting locked — automatically, across every video you produce.

From $149/mo · Cancel anytime