Skip to content

Formerly KrillinAI, VideoTranslation and Dubbing Tool for 120+ Languages. Open-source AI workspace for creators, powered by Codex. Video downloader, video translation, create videos, images, audios, and edits with Agents in one place. For your Social Media like YouTube, TikTok, Instagram, etc.

5.7 KB
Raw Blame
[app]
    segment_duration = 5 # 音频切分处理间隔,单位:分钟,建议值:5-10,如果视频中话语较少可以适当提高
    transcribe_parallel_num = 1 # 并发进行转录的数量上限,建议值:1-3,如果使用了本地模型,最好调成1
    translate_parallel_num = 3 # 并发进行翻译的数量上限,建议值:3,倍于转录的并发量,如果使用TPM限制严格的API,可以适当调低
    transcribe_max_attempts = 3 # 转录最大尝试次数,建议值:3
    translate_max_attempts = 5 # 翻译最大尝试次数,建议值:5,如果模型参数量较少或翻译失败率较高可以适当调高
    max_sentence_length = 60 # 原译每行生成目标;60时验收允许至65,其他值按配置硬上限;每个Unicode字素簇计1,包含空格和标点
    enable_block_vtt_batch = false # 是否启用块级VTT批量翻译,建议值:false(默认关闭)
    vtt_batch_size = 10 # VTT批量翻译的批次大小,建议值:10
    target_language_first = true # 双语字幕中目标语言是否在上方,建议值:true(目标语言在上)
    short_subtitle_max_chars = 20 # 短字幕英文每行最大字符数,建议值:15-25
    proxy = "" # 网络代理地址,格式如http://127.0.0.1:7890,可不填
    youtube_proxy = "" # YouTube/yt-dlp 专用代理;留空时回退到 proxy

[server]
    host = "127.0.0.1"
    port = 8888

# 下方的配置不是都要填,请结合文档说明进行配置

[llm] #支持openai,deepseek,通义千问等所有兼容openai请求格式的模型服务
    base_url = "" # 自定义base url,可配合转发站密钥使用,留空为openai官方api
    api_key = "" # API密钥
    model = "" # 指定模型名,可通过此字段结合base_url使用外部任何与OpenAI API兼容的大模型服务,留空默认为gpt-4o-mini
    json = false # 所使用的llm接口是否支持json格式,如果支持请设置为true,若不知道这是什么,请保持为false

[transcribe] # 视频转文本支持多种方案,配置时先填provider,再填对应的配置
    provider = "openai" #语音识别,当前可选值:openai,fasterwhisper,whisperkit,whisper.cpp,aliyun,volcengine。(fasterwhisper不支持macOS,whisperkit只支持M芯片)
    enable_gpu_acceleration = false # 给fasterwhisper进行GPU加速选项,50系显卡请务必开启,否则无法正常运行
    [transcribe.openai]
        base_url = ""
        api_key = ""
        model = "whisper-1"
    [transcribe.fasterwhisper]
        model = "medium" # fasterwhisper的本地模型可选值:tiny,medium,large-v2。建议medium及以上
    [transcribe.whisperkit]
        model = "large-v2" # whisperkit的本地模型可选值:large-v2
    [transcribe.whispercpp]
        model = "tiny" # whispercpp的本地模型可选值:tiny,medium,large-v2,large-v3-turbo
    [transcribe.aliyun] # provider选aliyun这块就都要填
        [transcribe.aliyun.oss]
            access_key_id = ""
            access_key_secret = ""
            bucket = ""
            region = "cn-shanghai" # Bucket 所在地域;旧配置或留空时默认上海,例如 ap-southeast-1
            endpoint = "" # 可选,不含 Bucket 的公网 OSS Endpoint;留空时由 region 生成
        [transcribe.aliyun.speech]
            access_key_id = ""
            access_key_secret = ""
            app_key= ""
    [transcribe.volcengine] # 豆包录音文件识别大模型,provider选volcengine时填写
        app_id = ""
        access_token = ""
        resource_id = "volc.seedasr.auc"
        base_url = "https://openspeech.bytedance.com"

[tts]
    provider = "aliyun" # 可选值:openai,aliyun,edge-tts,minimax,volcengine
    [tts.openai]
        base_url = ""
        api_key = ""
        model = "" # gpt-4o-mini-tts, tts-1, tts-1-hd
    [tts.minimax] # MiniMax TTS(T2A v2),provider选minimax时填写
        base_url = "" # 留空默认海外版 https://api.minimax.io,国内可填 https://api.minimaxi.com
        api_key = "" # MiniMax API密钥
        model = "" # 留空默认 speech-2.8-hd,可选 speech-2.8-turbo, speech-2.6-hd, speech-2.6-turbo
    [tts.aliyun] # 阿里云百炼语音合成,仅 API Key 必填
        base_url = "https://dashscope.aliyuncs.com/api/v1"
        api_key = ""
        model = "qwen3-tts-flash"
    [tts.volcengine] # 豆包语音合成。小模型 volcano_tts 走 V1;seed-tts-2.0 / seed-icl-2.0 走 V3
        app_id = ""
        access_token = ""
        cluster = "volcano_tts" # volcano_tts | seed-tts-2.0 | seed-tts-1.0 | seed-icl-2.0 | seed-icl-1.0 | volcano_icl
        default_voice_id = "BV001_streaming" # 2.0 可填 zh_female_gaolengyujie_uranus_bigtts;复刻填 S_ 开头 Speaker ID
        base_url = "https://openspeech.bytedance.com"

[dubbing]
    min_subtitle_duration = 2.5 # 最短配音字幕时长,短句会优先合并
    max_chunk_size = 5 # 单个配音 chunk 最多合并字幕条数
    gap_tolerance = 1.5 # 可吸收的相邻字幕空隙,单位秒
    speed_min = 0.95 # 允许的最慢调速倍率
    speed_accept = 1.15 # 推荐的最大自然调速倍率
    speed_max = 1.30 # 硬上限,超过后优先改写文本
    enable_text_rewrite = true # 是否允许 LLM 改写为自然口播
    rewrite_max_attempts = 2 # 单条字幕最多改写次数
    estimator = "statistical" # 估时器,当前支持 statistical

[image] # 封面生图配置
    provider = "openai-compatible" # 当前支持openai-compatible,使用OpenAI Images API兼容格式
    [image.openai]
        base_url = "" # 生图接口base url,留空使用OpenAI官方接口;自定义转发站通常填写以/v1结尾的地址
        api_key = "" # 生图API密钥
        model = "gpt-image-1" # 生图模型,例如gpt-image-1,或转发站支持的兼容模型