AI Models & Limits
Nano Banana image models, Omni and Veo video models, and technical boundaries.
Serelius Flow Video leverages Google Flow's foundational models for all media generation. This reference outlines model capabilities, resolution constraints, and system limits.
1. Image Generation Models
Image models generate canonical 1:1 character sheets and scene storyboards:
| Model | Profile Tier | Distinctive Capabilities |
|---|---|---|
| Nano Banana 2 Lite | Economic | Swift generation speed with minimal Google Flow credit consumption. |
| Nano Banana 2.1 | Balanced | Standard balance of visual clarity, textural fidelity, and prompt adherence. |
| Nano Banana Pro | Assertive | Maximum rendering fidelity for intricate styles, fine linework, and complex lighting. |
2. Video Generation Models
Video models animate storyboards, render camera movement, and generate synchronized dialogue:
| Model | Tier | Supported Resolutions | Supported Durations | Key Strengths |
|---|---|---|---|---|
| Omni 1.1 Flash | Economic | 360p and 720p | 4, 6, 8, 10 seconds | Fast turnaround, lower credit footprint, flexible scene lengths. |
| Veo 3.1 — Fast | Balanced | 720p HD | 8 seconds | High visual stability and cinematic camera trajectory. |
| Veo 3.1 — Lite | Lightweight | 720p HD | 8 seconds | Lighter resource footprint within the Veo 3.1 architecture. |
| Veo 3.1 — Quality | Assertive | 720p HD | 8 seconds | Peak visual fidelity, rich photorealism, and clean motion clarity. |
3. Text Generation Engine
All script generation, style recommendations, and kinetic Match-Cut prompts are powered by Gemini 2.0 Flash, Google Flow's primary language model.
4. Optimization Profiles
Pre-configured optimization bundles allow quick calibration between speed, credit cost, and aesthetic fidelity:
| Profile | Image & Storyboard Model | Video Model | Video Resolution |
|---|---|---|---|
| Economic | Nano Banana 2 Lite | Omni 1.1 Flash | 360p Draft |
| Assertive | Nano Banana Pro | Veo 3.1 Quality | 720p HD |
| Custom | User-selected image model | User-selected video model | 360p or 720p |
5. Technical Numbers and Boundaries
Every technical limit is enforced deterministically by the tool:
| Parameter | Specification Limit | Operational Context |
|---|---|---|
| Scene Duration | 8 seconds | Default duration benchmark across script scenes. |
| Scene Range Target | 1 to 225 scenes | From single-clip shorts to complete 30-minute productions. |
| Raw Story Input | Up to 30,000 characters | Maximum source story pasted into the AI script tab. |
| Script Batching | 10 scenes per API call | Automated batching prevents text token truncation. |
| Pacing Auto-Repair | Up to 2 rounds per batch | Dialogue refinement ensuring spoken lines fit scene limits. |
| Assets per Scene | Up to 3 visual assets | Maximum props and graphics bound to a single scene. |
| Storyboard References | Up to 10 image inputs | Priority stack (characters, props, location frame). |
| Video References | Up to 5 image inputs | Maximum references attached to a single video prompt. |
| Custom Voice Prompt | 15 to 400 characters | Allowed description length for custom narrator or character voices. |
| Style Name & Header | 3–40 characters / ≤ 200 chars | Style title length and sentence cap for the first two lines. |
| Subtitle Phrase Sizing | Up to 6 words per block | Optimal visual density for burned-in captions. |
| Project JSON Backup | Up to 30 MB | Maximum file size accepted during project import. |
| Reconnect Media Batch | Up to 50 items per selection | Number of assets linked from Flow gallery in one batch. |
| EDL Timeline Standard | CMX 3600 at 24 fps | Base timecode starts at 01:00:00:00 for NLE import. |
Ready to build and launch your SaaS?
Get 100% full source code ownership, no proprietary wrappers, with architecture ready for millions of requests on Cloudflare.