Multi-Model AI Routing.
Not every layer needs the same AI model. Onira routes stills to Gemini Flash Image, animates them with Pixverse v6, locks narration first with ElevenLabs eleven_v3, and scores each act with ElevenLabs Music.
The right model for every scene.
Onira's routing engine evaluates scene requirements and selects the optimal AI model. Here's how each model contributes.
Gemini Flash Image
Generates the source still frame for every scene. The ImageDirector agent (Gemini 3.1 Pro) reviews each image prompt before generation, ensuring appearance and composition are locked in before motion is planned.
Used for: All scene stills, character portraits, establishing frames
Pixverse v6
Animates every still frame into a 1–14 second motion clip. The VideoDirector agent (Gemini 3.1 Pro) plans the motion prompt separately from the image prompt — appearance and movement are never conflated.
Used for: All video clips, image-to-video animation, scene motion
ElevenLabs eleven_v3
Generates per-beat narration in 30+ languages. Audio is produced first — narration durations are locked before any visual is generated, so visuals always conform to audio timing.
Used for: Narration per beat, 30+ BCP-47 languages, character voices
ElevenLabs Music
Generates one original soundtrack per act. Genre, tempo, and mood are derived from the video's narrative arc. No stock music — every score is composed fresh for the video.
Used for: Per-act original soundtrack, genre and mood matching
How routing decisions
are made.
Scene Analysis
Narration is generated first via ElevenLabs eleven_v3 — beat durations are locked before any visual is planned. The ImageDirector agent then analyzes each scene for appearance and composition.
Model Scoring
Gemini Flash Image renders the still frame (Nano Banana 2 is the quota-exhaustion fallback). The VideoDirector agent plans the motion prompt separately — appearance and motion prompts never overlap.
Generation & Validation
Pixverse v6 animates each still into a 1–14 second clip. ElevenLabs Music composes one original score per act. Remotion assembles all layers into the finished video.
Why multi-model beats
single-model.
No single AI model is the best at everything. Onira's approach ensures every scene gets the best possible result.
Best quality per scene
Gemini Flash Image for stills, Pixverse v6 for motion, ElevenLabs for narration and score. Each layer gets a dedicated specialist. Single-model tools use the same model for everything — Onira orchestrates each layer individually.
Cost optimization
Not every scene needs the most expensive model. Simple transitions use efficient models, while hero shots get premium ones. Smart allocation reduces cost without sacrificing quality.
Future-proof architecture
When new AI models launch - and they launch constantly - Onira integrates them into the routing engine. Your videos automatically benefit from the latest advances.
Example routing for
a documentary.
A 10-minute deep ocean documentary. Here's how Onira routes different types of scenes.
| Scene Type | Assigned Model | Reason |
|---|---|---|
| Sweeping ocean vista | Gemini Flash Image → Pixverse v6 | Still frame animated into clip |
| Jellyfish bioluminescence | Gemini Flash Image → Pixverse v6 | Detail still, gentle motion |
| Submarine descent sequence | Gemini Flash Image → Pixverse v6 | Mechanical motion via Pixverse |
| Extended coral reef pan | Gemini Flash Image → Pixverse v6 | Wide still, slow pan motion |
| Per-beat narration | ElevenLabs eleven_v3 | Audio locked first, 30+ languages |
| Act soundtrack | ElevenLabs Music | Original score per act |
What you actually get.
Routing decisions · Deep Ocean Documentary
67 scenes routed automatically · Stills → motion clips via Gemini + Pixverse · Narration locked first · Zero manual decisions
Every scene,
the best model.
Experience multi-model AI video production today. Let Onira route every scene to perfection.
From $149/mo · Cancel anytime