Skip to main content
Legal · Model and content policy

Model and content policy

Model sources, capability boundaries, input restrictions and responsibility for results.

Review draft · Updated 2026-02-15
In short

Ombak does not train models. It calls third-party providers. This page states what is connected, the limits that follow, and the real state of content moderation.

01

Connected models

Image generation: Nano Banana 2 (a Google Gemini-family image model), called through fal.ai. 1K / 2K / 4K and seven aspect ratios, one image per generation, up to five reference images. Every other model in the directory, and all video models, are not connected: they are marked as such and cannot be submitted.

02

Your responsibilities

Do not use the service for unlawful, harmful, infringing or deceptive content, or to process someone’s identity without authorisation; see the acceptable use policy in the terms. Uploading a reference image is a statement that you hold the rights to it.

03

State of content moderation

Ombak does **not** currently run its own automated moderation of prompts or generated output. Today it relies on the safety filtering built into the upstream model provider; a request that provider refuses fails and the credits are returned. Our own prompt and output moderation, moderation logging and abuse-report handling are being built, and this page will be updated before they ship. Until then, report violations to support@ombak.ai and we will review them by hand. Generated content can be inaccurate or differ from the input; review it before publishing or commercial use.

Disputes and contact

For questions about these terms, billing or content handling, contact support@ombak.ai first and we will respond as soon as we can. These terms may change as the product changes; material changes will carry an effective date on this page.