IMA Studio TTS

IMA TTS (Text-to-Speech)

Overview

Call IMA Open API to create text-to-speech audio. Same flow as other IMA creation skills: query products → create task → poll until done. Task type is text_to_speech. This skill targets seed-tts-2.0 only — seed-tts-1.1 is not supported; the script defaults to seed-tts-2.0 when no model is specified.

⚙️ How This Skill Works

This skill uses a bundled Python script scripts/ima_tts_create.py to call the IMA Open API:

• Sends text (prompt) to https://api.imastudio.com
• Uses --user-id only locally for preference storage
• Returns an audio URL when synthesis is complete
• Reflection mechanism: on create failure, retries up to 3 times with parameter adjustments

What gets sent to IMA: prompt (text to speak), model selection, parameters (e.g. voice_id, speed). Not sent: API key in prompt body; user_id is local only.

Agent Execution

Use the bundled script:

# List available TTS models (optional; default is seed-tts-2.0)
python3 {baseDir}/scripts/ima_tts_create.py --api-key $IMA_API_KEY --list-models

# Generate speech (default model: seed-tts-2.0; omit --model-id to use default)
python3 {baseDir}/scripts/ima_tts_create.py \
  --api-key $IMA_API_KEY \
  --model-id seed-tts-2.0 \
  --prompt "Text to be spoken here." \
  --user-id {user_id} \
  --output-json

Script outputs JSON; parse it for url and pass to the user via the UX protocol below.

Environment

Base URL: https://api.imastudio.com

⚠️ MANDATORY: Always Query Product List First

You MUST call /open/v1/product/list with category=text_to_speech before creating any task. attribute_id is required; if 0 or missing → "Invalid product attribute" and task fails.

GET /open/v1/product/list?app=ima&platform=web&category=text_to_speech

Then traverse the V2 tree: type=2 = model groups, type=3 = versions (leaves). Only type=3 nodes have credit_rules and form_config. Use a leaf’s model_id, id (= model_version), and credit_rules[0].attribute_id / points for create.

Core Flow

1. GET /open/v1/product/list?app=ima&platform=web&category=text_to_speech
   → Get attribute_id, credit, model_version, form_config

2. POST /open/v1/tasks/create
   → task_type: "text_to_speech", parameters[].parameters.prompt = text to speak

3. POST /open/v1/tasks/detail  { "task_id": "..." }
   → Poll every 2–5s until medias[].resource_status == 1 and status != "failed"
   → Read medias[].url (and optional duration_str, format)

Task Detail API — Actual Response Shape

Poll POST /open/v1/tasks/detail until completion. Response uses the same structure as other IMA audio tasks:

Completed success example:

{
  "id": "task_xxx",
  "medias": [{
    "resource_status": 1,
    "status": "success",
    "url": "https://cdn.../output.mp3",
    "duration_str": "12s",
    "format": "mp3"
  }]
}

Rules:

• Treat resource_status: null as 0 (processing).
• Success only when all medias have resource_status == 1 and status != "failed".
• On resource_status == 2 or status == "failed", stop and handle error (e.g. use error_msg / remark).

API 2: Create Task

POST /open/v1/tasks/create

text_to_speech — no image input. src_img_url: [], input_images: [].

{
  "task_type": "text_to_speech",
  "enable_multi_model": false,
  "src_img_url": [],
  "parameters": [{
    "attribute_id":  "<from credit_rules>",
    "model_id":      "<model_id>",
    "model_name":    "<model_name>",
    "model_version": "<version_id>",
    "app":           "ima",
    "platform":      "web",
    "category":      "text_to_speech",
    "credit":        "<points>",
    "parameters": {
      "prompt":       "Text to be spoken.",
      "n":            1,
      "input_images": [],
      "cast":         {"points": "<points>", "attribute_id": "<attribute_id>"}
    }
  }]
}

prompt must be inside parameters[].parameters, not at top level. Extra fields (e.g. voice_id, speed) come from product form_config; include only those present in the product’s credit_rules/form_config.

Response: data.id = task_id for polling.

Supported Task Type & Models

Models: This skill supports seed-tts-2.0 only (seed-tts-1.1 is not supported). The script defaults to --model-id seed-tts-2.0 when none is provided. For current attribute_id and credit, the script reads from the product list at runtime.

seed-tts-2.0 — Verified request parameters

The following parameters[].parameters shape has been verified to work for seed-tts-2.0 (attribute_id/credit come from product list and may differ by app/platform):

Script example with extra params:

python3 ima_tts_create.py --api-key $IMA_API_KEY --model-id seed-tts-2.0 \
  --prompt "阳光青年音色测试，你好世界。" \
  --extra-params '{"model":"seed-tts-2.0-expressive","speaker":"zh_male_sophie_uranus_bigtts","audio_params":{"emotion":"neutral"},"additions":{"explicit_language":"crosslingual","context_texts":[]}}' \
  --output-json

Note: The script gets attribute_id and credit from the product list (e.g. app=ima&platform=web → often 2 pts / attribute_id 4419 for seed-tts-2.0). If you have a different app/platform (e.g. webAgent), the product list may return different credit_rules (e.g. 5 pts / attribute_id 8987); the script uses whatever the product list returns for the chosen model.

Speaker / 音色列表（seed-tts-2.0 兼容火山引擎音色）： 完整音色 ID 与场景分类见项目内 volcengine_tts_timbre_list.json。该文件来自 火山引擎豆包语音合成音色列表，使用原生 voice_type 格式（如 zh_male_sophie_uranus_bigtts 魅力苏菲、zh_female_vv_uranus_bigtts Vivi）。⚠️ 注意： IMA API 只支持原生格式（*_uranus_bigtts 系列），不支持 BV*_streaming 豆包音色 ID。

与火山引擎 2.0 文档对照： 上述参数与 HTTP Chunked/SSE 单向流式 V3 请求 Body 一致：req_params.text → prompt，req_params.speaker → speaker（必填项），req_params.model → model（expressive/standard），req_params.audio_params（emotion、speech_rate、loudness_rate 等），req_params.additions（如 explicit_language）。2.0 能力说明见 豆包语音合成2.0能力介绍（语音指令、引用上文、语音标签等）。

🎤 当用户说「帮我制作旁白/配音」时如何询问

当用户表达「帮我制作旁白」「做一段配音」「把这段文字读出来」等意图时，必须先收集关键信息再调用脚本，避免缺参或盲目默认。

必问

建议问（让用户选择）

可选问（按需补充）

脚本对应： --prompt 必填；--speaker、--emotion 直接支持；语速/音量/模型等通过 --extra-params 传入 JSON（见上文 Script example）。

📥 User Input Parsing (Parameter Recognition)

Map user intent to parameters using product form_config (e.g. voice, speed):

If the user does not specify, use form_config defaults. Do not send parameters not present in the product’s credit_rules/attributes or form_config (reflection will strip them on retry).

🧠 User Preference Memory

Storage: ~/.openclaw/memory/ima_prefs.json

{
  "user_{user_id}": {
    "text_to_speech": {
      "model_id": "...",
      "model_name": "...",
      "credit": 2,
      "last_used": "..."
    }
  }
}

• Before generation: Load prefs; if user_{user_id}.text_to_speech exists, use that model and optionally mention it.
• After success: Save used model to user_{user_id}.text_to_speech.
• On explicit change: e.g. “换成XXX” / “以后都用XXX” → switch and save.

💬 User Experience Protocol (IM / Feishu / Discord)

TTS usually completes in a few seconds to tens of seconds. Do not leave users without feedback.

Step 0 — Initial Acknowledgment (Normal Reply)

First reply with a short acknowledgment (normal reply, not message tool), e.g.:

• 好的，正在帮你把这段文字转成语音。
• OK, converting this text to speech.

Step 1 — Pre-Generation Notification (message tool)

Push once:

🔊 开始语音合成，请稍候…
• 模型：[Model Name]
• 预计耗时：[X ~ Y 秒]
• 消耗积分：[N pts]

Step 2 — Progress

Poll every 2–5s. Every 10–15s send a progress update, e.g.:

⏳ 语音合成中… [P]%
已等待 [elapsed]s，预计最长 [max]s

Cap progress at 95% until API returns success.

Step 3 — Success (message tool)

When resource_status == 1 and status != "failed", send the audio and caption:

• media = medias[0].url
• caption example:

✅ 语音合成成功！
• 模型：[Model Name]
• 耗时：[actual]s
• 消耗积分：[N pts]
🔗 原始链接：[url]

Use the URL from the API (do not use local file paths).

Step 4 — Failure (message tool)

On failure or API/network error, send a short, user-friendly message and suggestions:

❌ 语音合成失败
• 原因：[自然语言原因]
• 建议：换个模型重试或检查文本长度/内容

需要我帮你用其他模型重试吗？

Error translation (do not expose raw API/technical errors):

Links: API key — https://www.imaclaw.ai/imaclaw/apikey ；Credits — https://www.imaclaw.ai/imaclaw/subscription

Step 5 — Done

After Step 0–4, no further reply needed. Do not send duplicate confirmations.

Common Mistakes

Security & Local Data

• Network: This skill uses only https://api.imastudio.com (no image upload domain for TTS).
• Local files: ~/.openclaw/memory/ima_prefs.json (preferences), ~/.openclaw/logs/ima_skills/ (logs, e.g. 7-day retention). No prompts or API keys stored.
• API key: Set via environment (e.g. IMA_API_KEY) or agent config; never hardcode.

Python Example (Minimal)

import time
import requests

BASE = "https://api.imastudio.com"
HEADERS = {
    "Authorization": "Bearer ima_your_key",
    "Content-Type": "application/json",
    "x-app-source": "ima_skills",
}

# 1. Product list
r = requests.get(
    f"{BASE}/open/v1/product/list",
    headers=HEADERS,
    params={"app": "ima", "platform": "web", "category": "text_to_speech"},
)
tree = r.json()["data"]
# ... find type=3 node, get attribute_id, model_id, model_version, credit ...

# 2. Create task
body = {
    "task_type": "text_to_speech",
    "enable_multi_model": False,
    "src_img_url": [],
    "parameters": [{
        "attribute_id": attribute_id,
        "model_id": model_id,
        "model_name": model_name,
        "model_version": model_version,
        "app": "ima", "platform": "web",
        "category": "text_to_speech",
        "credit": credit,
        "parameters": {
            "prompt": "Hello, world.",
            "n": 1,
            "input_images": [],
            "cast": {"points": credit, "attribute_id": attribute_id},
        },
    }],
}
r = requests.post(f"{BASE}/open/v1/tasks/create", headers=HEADERS, json=body)
task_id = r.json()["data"]["id"]

# 3. Poll
while True:
    r = requests.post(f"{BASE}/open/v1/tasks/detail", headers=HEADERS, json={"task_id": task_id})
    task = r.json()["data"]
    medias = task.get("medias") or []
    if not medias:
        time.sleep(3)
        continue
    rs = medias[0].get("resource_status")
    if rs is None: rs = 0
    if rs == 2 or (medias[0].get("status") or "").lower() == "failed":
        raise RuntimeError(medias[0].get("error_msg") or "failed")
    if rs == 1 and (medias[0].get("url") or medias[0].get("watermark_url")):
        url = medias[0]["url"] or medias[0]["watermark_url"]
        print(url)  # e.g. https://cdn.../output.mp3
        break
    time.sleep(3)

Quick Reference