Everything Pyro says is generated by artificial intelligence. Pyro is a fictional AI mascot character, not a person. No human writes, reviews, or approves individual roasts before you see them.
Where AI is used
- Roast text. Every roast, reply, and chat message from Pyro is produced by a large language model in real time.
- Photo understanding. When you send a photo, a vision model describes it so Pyro can roast what's in it. The models see only what you submit.
- Reaction clips. Pyro's on-screen animations (idle, laugh, sneer, explode) are pre-rendered clips; the AI only selects which one to play.
Models we use
- Text: Claude (Anthropic), accessed via OpenRouter.
- Vision: Gemini (Google), accessed via OpenRouter.
- Backup: GPT (OpenAI), accessed via OpenRouter, only when the models above are unavailable.
Nothing is sent to these models until you allow it in the App. All requests are restricted to Zero Data Retention endpoints — your prompts, photos, and Pyro's outputs are not kept by these services and are not used to train their models. Details in the Privacy Policy §5, and how photos and face data are handled in §2.
How the App tells you
- A first-launch disclosure explains that all Pyro content is AI-generated.
- Before anything is sent to an AI service, the App asks for your permission, naming what is sent (photos, including faces; text; chat messages) and who receives it. You can withdraw it in Settings → AI data sharing.
- A permanent AI-content disclosure row lives in the App's Settings.
- Every roast is attributed in the UI to the Pyro mascot character.
- Exported share-cards carry the Roast Master watermark, marking their origin.
Limits of AI comedy
Pyro's roasts are designed to punch at choices — the outfit, the texts, the vibe — never at who someone is. Safety systems keep output inside an edgy-comedy band and deflect attempts to target protected classes, minors, or non-consenting individuals. AI output is probabilistic, though: a specific response can miss the mark. If one does, report it.
Reporting AI content
Long-press any Pyro response in the App and tap Report, or email pyro@thunder.red with a screenshot. Reports are triaged within 24 hours; confirmed violations are taken down immediately. See Terms of Service §6.
Contact
See also: Privacy Policy · Terms of Service