config.toml

# ============= 开发者选项 / Developer Options ===============
[developer]
developer_mode = false
default_llm = "deepseek-chat"
default_vlm = "qwen3-vl-8b-instruct"
print_context = false # 在拦截器打印模型拿到的全部上下文，会很长 / Print full context in interceptor (output will be very long)

# ============= 模型配置 for 体验网页 ===============
[developer.chat_models_config."deepseek-chat"]
base_url = ""
api_key = ""
temperature = 0.1

[developer.chat_models_config."qwen3-vl-8b-instruct"]
base_url = ""
api_key = "" 
timeout = 20.0
temperature = 0.1
max_retries = 2

# ============= 项目路径 / Project Paths ======================
[project]
media_dir = "./outputs/media"
bgm_dir = "./resource/bgms"
outputs_dir = "./outputs"

# ============= 模型配置 for user / Model Config for User =============
[llm]
model = "deepseek-chat"
base_url = ""
api_key = "" 
timeout = 30.0 # 单位：秒
temperature = 0.1
max_retries = 2

[vlm]
model = "qwen3-vl-8b-instruct"
base_url = ""
api_key = "" 
timeout = 20.0 # 单位：秒
temperature = 0.1
max_retries = 2


# ============= MCP Server 相关 / MCP Server Related =============
[local_mcp_server]
server_name = "storyline"
server_cache_dir = ".storyline/.server_cache"
server_transport = "streamable-http"  # server 和 host之间的传输方式 / Transport method between server and host
url_scheme = "http"  
connect_host = "127.0.0.1" # 不要改动 / Do not change
port = 8001                # 如果端口冲突，可以随便用一个有空的端口 / Use any available port if conflict occurs
path = "/mcp"              # 默认值，一般不用改 / Default value, usually unchanged

json_response = true       # 建议用 True / Recommended: True
stateless_http = false     # 强烈建议用 False / Strongly recommended: False
timeout = 600
available_node_pkgs = [
    "open_storyline.nodes.core_nodes"
]
available_nodes = [
    "LoadMediaNode", "SearchMediaNode", "SplitShotsNode",
    "UnderstandClipsNode", "FilterClipsNode", "GroupClipsNode", "GenerateScriptNode", "ScriptTemplateRecomendation",
    "GenerateVoiceoverNode", "SelectBGMNode", "RecommendTransitionNode", "RecommendTextNode",
    "PlanTimelineProNode", "RenderVideoNode"
]

# =========== skills ==========
[skills]
skill_dir = "./.storyline/skills"

# =========== pexels ==========
[search_media]
pexels_api_key = ""

# ============= 镜头分割 / Shot Segmentation =============
[split_shots]
transnet_weights = ".storyline/models/transnetv2-pytorch-weights.pth"
transnet_device = "cpu"

# ============= 视频视觉理解 / Video Visual Understanding =============
[understand_clips]
sample_fps = 2.0                 # 每秒抽几帧 / Frames sampled per second
max_frames = 64                  # 单clip抽帧上限兜底，避免长视频爆 token / Max frames per clip limit to prevent token overflow

# ============= 文案模板 / Script Templates =============
[script_template]
script_template_dir = "./resource/script_templates"
script_template_info_path = "./resource/script_templates/meta.json"

# ============= 配音生成 / Voiceover Generation ===================
[generate_voiceover]
tts_provider_params_path = "./resource/tts/tts_providers.json"

[generate_voiceover.providers.302]
base_url = ""
api_key = ""

[generate_voiceover.providers.bytedance]
uid = ""
appid = ""
access_token = ""

[generate_voiceover.providers.minimax]
base_url = ""
api_key = ""


# ============= BGM选择 / BGM Selection ====================
# 主要是用于计算音乐特征的一些参数 / Mainly parameters for calculating music features
[select_bgm]
sample_rate = 22050
hop_length = 2048   # 每次分析窗口向前跳多少个采样点，越小越精细（但更慢） / Hop length samples; smaller = more precise but slower
frame_length = 2048 # 计算信号的均方根RMS的窗口大小。越大越稳定，但对瞬态不敏感 / Window size for RMS; larger = stable but less sensitive to transients

# ============= 字体推荐 / Font Recommendation ====================
[recommend_text]
font_info_path = "resource/fonts/font_info.json"

# ============= 时间线组织 / Timeline Organization ====================
[plan_timeline]
beat_type_max = 1 # 使用多强的鼓点，例如4/4中，鼓点类似1,2,1,3，其中1是最强的，3最弱 / Beat strength (e.g., in 4/4: 1=strongest, 3=weakest)
title_duration = 0 # 片头时长 (ms) / Intro duration (ms)
bgm_loop = true # 是否允许 bgm 循环 / Allow BGM loop
min_clip_duration = 1000

estimate_text_min = 1500 # 在没有TTS的情况下，估计每段字幕至少上屏多久 / Min on-screen duration for subtitles without TTS
estimate_text_char_per_sec = 6.0 # 在没有TTS的情况下，估计每秒展示几个字 / Estimated characters per second without TTS

image_default_duration = 3000 # 默认的图片播放时长 / Default image duration
group_margin_over_voiceover = 1000 # 在一个group中，画面比配音多出现多久 / Extra visual duration over voiceover in a group

[plan_timeline_pro]

min_single_text_duration = 200      # 单段文字最小时长 (ms) / min single text duration (ms)
max_text_duration        = 5000     # 单句文字最大时长 (ms) / max text sentence duration (ms)
img_default_duration     = 1500     # 默认图片时长 (ms) / default image duration (ms)

min_group_margin = 1500             # 段落/组最小间距 (ms) / min paragraph/group margin (ms)
max_group_margin = 2000             # 段落/组最大间距 (ms) / max paragraph/group margin (ms)

min_clip_duration = 1000            # 最小片段时长 (ms) / min clip duration (ms)

tts_margin_mode = "random"          # random | avg | max | min
min_tts_margin  = 300               # 最小 TTS 间隔 (ms) / min TTS gap (ms)
max_tts_margin  = 400               # 最大 TTS 间隔 (ms) / max TTS gap (ms)

text_tts_offset_mode = "random"     # random | avg | max | min
min_text_tts_offset  = 0            # 最小文字-TTS偏移 (ms) / min text–TTS offset (ms)
max_text_tts_offset  = 0            # 最大文字-TTS偏移 (ms) / max text–TTS offset (ms)

long_short_text_duration = 3000     # 长/短文本阈值 (ms) / long/short text threshold (ms)
long_text_margin_rate   = 0.0       # 长文本起始边距率 / long text start margin rate
short_text_margin_rate  = 0.0       # 短文本起始边距率 / short text start margin rate

text_duration_mode = "with_tts"     # with_tts | with_clip (随配音 | 随片段)
is_text_beats      = false          # 文字对齐音乐节拍 / align text with music beats