AI models and their rules
Last updated: October 22, 2026
Showlight AI uses AI models made by other companies. When you use a feature, the content that job needs goes to the company that runs it. Our Privacy Policy explains what we send and what they keep.
You agree to follow each model maker's use rules as well as ours. If a model maker's rule is stricter than our Terms, the stricter rule applies. A model may also refuse a request on its own. If it does, your credits come back.
Most models run through fal.ai, a company that hosts AI models. fal's Acceptable Use Policy applies to every model it runs. Two exceptions: every Omni Flash video and every Extend goes straight to Google's Gemini API, where Google's Gemini API terms apply; and Seedance 2.5 runs through Replicate, another company that hosts AI models, where Replicate's Terms of Service apply.
Video
| Model in the app | Made by | Use rules |
|---|---|---|
| Omni Flash 1.1 | Google. Every video and Extend runs on Google's Gemini API. | Generative AI Prohibited Use Policy |
| Wan 3.0 | Alibaba, run through fal.ai | fal's Acceptable Use Policy |
| Seedance 2.5 | ByteDance, run through Replicate | ByteDance's video model terms and Replicate's Terms of Service. This model may refuse photos of real people. |
| Upscale (a sharper copy of a finished video), ByteDance video upscaler | ByteDance, run through fal.ai | fal's Acceptable Use Policy |
Images, voice, music and captions
| Feature | Model | Made by | Use rules |
|---|---|---|---|
| Images (from your words, or your words plus up to 4 of your own images) | GPT Image 2 | OpenAI | Usage policies |
| VoiceOver (stock voices) | Multilingual v2 | ElevenLabs | Use Policy |
| VoiceOver in your saved voices, and Generate a voice (a new synthetic voice from a description you type; we refuse descriptions of a real or famous person's voice) | Qwen3-TTS | Alibaba | fal's Acceptable Use Policy. The model itself is open source (Apache 2.0 license). |
| AI music | MiniMax Music 2.6 | MiniMax, run through fal.ai | MiniMax's Terms of Service and fal's Acceptable Use Policy |
| Captions | Scribe (speech-to-text) | ElevenLabs, run through fal.ai | ElevenLabs' Use Policy and fal's Acceptable Use Policy |
| Saved voices | fal media tools (sound levelling, voice isolation) | fal.ai | Acceptable Use Policy |
Writing and checks
| Feature | Model | Made by | Use rules |
|---|---|---|---|
| Ideas, clip prompts, VoiceOver scripts, post text. Inside a channel, this includes the channel's Channel text reference, the words of its 5 newest videos, and its most recent ideas | Claude Haiku 5.5 | Anthropic | Usage Policy |
| Protected-material check (your words, before we generate a video, image, song, voiceover or voice) | Claude Haiku 5.5 | Anthropic | Usage Policy |
| Photo check (each image you upload) | Claude Haiku 5.5 | Anthropic | Usage Policy |
When models change
We add and remove models often. The model picker in the app always shows what's available, with the maker's name. When we add a model from a new maker, we add it here on the same day.