PROMPTING.md
| 1 | # Writing target text for Higgs Audio v3 TTS — control tags |
| 2 | |
| 3 | How to embed control tags in the `input` text to steer emotion, prosody, style, and sound effects. |
| 4 | For where this fits in the overall workflow, see [AGENTS.md](./AGENTS.md). |
| 5 | |
| 6 | ## Format rule (read first) |
| 7 | |
| 8 | Every tag is `<|category:tag|>`. There are **two placements**: |
| 9 | |
| 10 | - **Sentence-level** — emotion, style, and prosody's `speed_* / pitch_* / expressive_*`. |
| 11 | Put at the **start of the sentence**; it colors the whole sentence. |
| 12 | - **Inline** — sound effects (`sfx`) and prosody's `pause / long_pause`. |
| 13 | Insert **at the exact position** in the sentence where the effect should occur. |
| 14 | |
| 15 | **`sfx` gotcha:** format is `<|sfx:tag|>onomatopoeia, then the line` — the tag comes **first**, |
| 16 | immediately followed by the onomatopoeia with **no space** between them. |
| 17 | |
| 18 | ## Examples |
| 19 | |
| 20 | Sentence-level: |
| 21 | ``` |
| 22 | <|emotion:elation|>Welcome aboard, we are absolutely thrilled to have you here! |
| 23 | <|style:whispering|>Come closer, I have a little secret to share. |
| 24 | <|prosody:speed_slow|>Take your time, there's really no need to rush. |
| 25 | ``` |
| 26 | |
| 27 | Inline sfx (tag first, onomatopoeia attached, no space): |
| 28 | ``` |
| 29 | <|sfx:cough|>Ahem, welcome everyone, let's get started. |
| 30 | <|sfx:laughter|>Haha, so glad you could make it! |
| 31 | ``` |
| 32 | |
| 33 | Inline pause (between phrases): |
| 34 | ``` |
| 35 | Hello there <|prosody:pause|> and welcome to the show. |
| 36 | ``` |
| 37 | |
| 38 | Stacking tags (sentence-level emotion + inline sfx in one line): |
| 39 | ``` |
| 40 | <|emotion:elation|><|sfx:laughter|>Haha, welcome, welcome, we're so happy you're here! |
| 41 | <|sfx:sigh|>Haah, what a day — but welcome, please make yourself at home. |
| 42 | ``` |
| 43 | |
| 44 | ## Tips |
| 45 | |
| 46 | - You can stack tags in one sentence (e.g. a leading emotion tag plus an inline sfx). |
| 47 | - `speed_very_slow` only slows the model to roughly ~5s; for slower delivery, insert |
| 48 | `<|prosody:long_pause|>` between phrases instead. |
| 49 | - Only the tags below are recognized — anything else degrades output or gets read literally. |
| 50 | |
| 51 | ## Full tag catalog (43) |
| 52 | |
| 53 | ### Emotion (21) — sentence-level |
| 54 | `affection`, `amusement`, `anger`, `arousal`, `awe`, `bitterness`, `confusion`, `contemplation`, |
| 55 | `contentment`, `determination`, `disgust`, `elation`, `enthusiasm`, `fear`, `helplessness`, |
| 56 | `longing`, `pride`, `relief`, `sadness`, `shame`, `surprise` |
| 57 | |
| 58 | Syntax: `<|emotion:elation|>` |
| 59 | |
| 60 | ### Prosody (10) |
| 61 | - Sentence-level: `speed_very_slow`, `speed_slow`, `speed_fast`, `speed_very_fast`, |
| 62 | `pitch_low`, `pitch_high`, `expressive_high`, `expressive_low` |
| 63 | - Inline: `pause`, `long_pause` |
| 64 | |
| 65 | Syntax: `<|prosody:speed_slow|>`, `<|prosody:pause|>` |
| 66 | |
| 67 | ### Style (3) — sentence-level |
| 68 | `singing`, `shouting`, `whispering` |
| 69 | |
| 70 | Syntax: `<|style:whispering|>` |
| 71 | |
| 72 | ### Sound effects (9) — inline |
| 73 | `cough`, `laughter`, `crying`, `screaming`, `burping`, `humming`, `sigh`, `sniff`, `sneeze` |
| 74 | |
| 75 | Syntax: `<|sfx:cough|>Ahem, ...` (tag first, onomatopoeia attached, no space) |
| 76 | |