# FAQ

### What's the difference between this repo and ai-humanize?

[ai-humanize](https://github.com/molly554/ai-humanize) documents 4 theoretical approaches. This repo (`humanize-text`) provides the **actual working Standard pipeline** with real code you can run.

### What API keys do I need?

- **LLM provider** (Steps 1–2) — choose one:
  - **DeepSeek** (default, free tier available) — set `api_keys.deepseek_api_key`
  - **OpenRouter** (optional) — set `api_keys.openrouter_api_key` and `[llm].provider = "openrouter"`
- **Niutrans** (free tier available) — for translation Step 4

Google Translate (Step 3) uses the free public API and doesn't require a key.

See [configuration.md](configuration.md) for provider setup, model slugs, and environment variable overrides.

### Can I use OpenRouter instead of DeepSeek?

Yes. OpenRouter exposes many models through a single OpenAI-compatible API. In `config.toml`:

```toml
[api_keys]
openrouter_api_key = "sk-or-..."

[llm]
provider = "openrouter"
model = "deepseek/deepseek-chat"
```

You can also override the endpoint with `[llm].base_url` or the `LLM_BASE_URL` environment variable.

### How long does processing take?

Typically 10-30 seconds per text, depending on length. The LLM rewriting steps (1-2) take the most time.

### Can I change the language chain?

The current chain (Chinese → Japanese → Finnish → English) is optimized for English input. Changing it requires modifying `src/standard/pipeline.py` or setting `[pipeline].intermediate_lang` in `config.toml`. Different chains may produce different quality levels.

### Why temperature 1.3?

Lower temperatures (1.0) produce more predictable, AI-detectable output. Higher temperatures (>1.5) cause incoherence. 1.3 is the sweet spot found through testing.

### How does this compare to Lynote.ai?

This repo provides the Standard tier only. [Lynote.ai](https://lynote.ai) combines Standard + Advanced + Focus tiers and automatically selects the best approach for each text passage.