Skip to content

Releases

v0.3.0

Choose a tag to compare

@github-actions github-actions released this 24 Sep 08:16
91dc14c

Trendlume v0.3.0: 多赛道生产流、云端生成扩展与故事板工作台重构

Trendlume v0.3.0 正式发布!本次更新将视频生成引擎由单一文本流水线全面升级为支持「知识科普、电商带货、故事短剧」的三大赛道生产体系,接入阿里云百炼、Google 与 RunningHub 云端图像与视频生成能力,并重构了解耦的故事板工作室与镜头级精修控制。

🌟 核心新特性与改动

1. 三大赛道生产流程(Multi-Track Production)

  • 知识科普:从选题调研、热榜灵感或现成文案切入,经过事实检索与结构化策划卡,自动拆解镜头大纲并进入故事板精修与本地合成。
  • 电商带货:引入商品事实表(Truth Sheet)与合规约束,支持创意策划与实拍素材强锁定,确保关键商品画面的真实性与前后一致。
  • 故事短剧:支持剧作设定集、角色库、场景库与道具库资产管理,支持多集分镜头拆解与审核审批门禁。

2. 模型与云端生成服务扩展(Provider Ecosystem)

  • 图像生成:新增阿里云百炼(Qwen Image 3.0 Pro,支持角色与场景参考)、Google(Gemini Image)以及 RunningHub 线上 ComfyUI 工作流适配器;保留本地 ComfyUI 与火山方舟 Seedream。
  • 视频生成:新增阿里云百炼(Wan 2.7 图生视频与文生视频)、Google(Veo 3.1 首尾帧与参考图支持)与 RunningHub 视频工作流适配器;保留火山方舟 Seedance 与本地 ComfyUI。
  • 前置说明:云端生成依赖对应平台的 API Key;本地 ComfyUI 需要具备相应显存与运行环境;内置微软 Edge-TTS 依然开箱即用、无需填 Key。

3. 故事板工作室与镜头级精修(Storyboard Workbench)

  • 解耦媒体规划与渲染:生成草稿后保留对每个镜头的修改权,支持单镜头旁白微调、分镜头试听、素材换源与局部重试,拒绝因单镜头不佳全片重跑。
  • 任务流水线中心:集中展示与管理多项目任务进度,支持随时暂停、恢复与失败镜头定位。

4. 安全、持久化与工程基建

  • 密钥安全脱敏:本地敏感凭据采用 AES 算法加密持久化存储,应用日志与状态流中全自动打码脱敏。
  • 纯轻量架构:坚持无重量级外部依赖(SQLite + asyncio 队列 + 本地存储),支持单文件一键打包与本地源码运行。
  • 文档与界面更新:重构中英文 README,同步更新 01~07 全套工作台界面预览截图。

⚠️ 使用边界与前置依赖

  • 视频合成与渲染:本地依赖 FFmpeg 以及 Playwright Chromium(源码运行首次需执行 uv run playwright install chromium)。
  • 社交平台发布:当前版本已实现官方创作者中心扫码发布的渠道为「抖音」;其他平台仍在规划与社区共建中。

Trendlume v0.3.0: Multi-Track Production, Cloud Providers, and Storyboard Workbench

Trendlume v0.3.0 is officially released! This update evolves the creation engine from a single script-to-video pipeline into dedicated multi-track production workflows (Knowledge, E-commerce, and Drama), introduces cloud image and video providers (Aliyun Bailian, Google, and RunningHub), and completely overhauls the storyboard editing experience.

🌟 Key Changes & Improvements

1. Multi-Track Production Workflows

  • Knowledge & Science: Start with topic research, trending news, or custom scripts, generate structured research cards, and inspect storyboard scenes before final local composition.
  • E-Commerce: Introduce product truth sheets and real-media asset pinning to ensure authentic product presentation without hallucinated details.
  • Short Drama: Manage drama bibles, character profiles, scene locations, and props, with episode/shot breakdowns and approval gates before media generation.

2. Expanded Provider Ecosystem

  • Image Generation: Added Aliyun Bailian (Qwen Image 3.0 Pro with character/scene reference support), Google (Gemini Image), and RunningHub cloud ComfyUI workflows, alongside local ComfyUI and Volcano Ark Seedream.
  • Video Generation: Added Aliyun Bailian (Wan 2.7 I2V and T2V), Google (Veo 3.1 with keyframe interpolation), and RunningHub video workflows, alongside Volcano Ark Seedance and local ComfyUI.
  • Prerequisites: Cloud providers require valid external API keys; local ComfyUI requires local GPU resources; built-in Edge-TTS remains zero-key ready to use.

3. Storyboard Workbench & Scene-Level Control

  • Decoupled Media Planning: AI generates the initial draft while creators retain granular control. Edit narrations, preview voiceover, swap footage, and retry individual scenes without regenerating the entire video.
  • Unified Pipeline Hub: Centrally track, pause, resume, and recover generation jobs across projects.

4. Security & Infrastructure

  • Credential Protection: API keys and tokens are encrypted at rest using local AES keys and automatically masked in logs.
  • Zero Heavyweight Dependencies: Continues to run on lightweight SQLite, asyncio queues, and local file storage with standalone executable packaging support.
  • Documentation: Overhauled bilingual READMEs and refreshed UI screenshots (01 to 07).

⚠️ Prerequisites & Boundaries

  • Rendering Dependencies: Local video rendering relies on system FFmpeg and Playwright Chromium (uv run playwright install chromium).
  • Publishing Path: Official creator-center QR authorization is currently verified for Douyin; additional platforms remain on the community roadmap.

v0.2.0

Choose a tag to compare

@github-actions github-actions released this 15 Sep 08:51

Trendlume v0.2.0: 从全网热点到分镜视频

Trendlume v0.2.0 带来了全新的「热点中心」,让视频创作除了“输入主题”和“粘贴文案”外,拥有了第三种更直接的起点:从全网公开热榜捕获灵感,生成选题提案,并一键转入故事板精修。

✨ 新增能力与体验优化

  • 全网热点中心(/trends):
    • 聚合微博、抖音、知乎、今日头条、小红书、B 站、百度 7 大平台公开热榜。
    • 采用双源容错机制(主源 60s API,备用源 xxapi),免去网页爬虫与登录失效困扰。
    • 支持项目级关键词关联与后台自动轮询,及时筛选与你创作领域匹配的高热话题。
  • 热点选题提案(Proposals):
    • 由大模型根据热点内容与项目偏好生成选题切入点、目标受众与内容大纲。
    • 确认提案后一键生成标准视频任务,直接进入镜头级故事板预览和编辑。
  • 画面来源模式(Content Modes)契约化:
    • 统一了 AI 生图、AI 视频、Pexels 素材库视频、自定义素材上传与纯文字排版卡片的领域契约。
    • 故事板局部重试和模板匹配更加可控,排除了旧版状态覆盖问题。
  • ComfyUI 本地生图更健壮:
    • 内置工作流校验自愈机制:当默认的 Flux 工作流在本地环境缺少特定节点或参数验证失败时,自动切换至兼容的 Turbo 基础工作流重试。
    • 提交失败时精准提示报错的节点 ID 和原因,便于本地排错。

⚙️ 环境与升级说明

  • 数据库升级:如果你从 v0.1.0 升级,进入 backend 目录运行 uv run alembic upgrade head 完成数据表迁移。
  • 前置依赖:热点提案生成依赖大模型配置;素材库视频模式依赖 Pexels API Key;其他本地功能与 Edge-TTS 继续保持开箱即用。

Trendlume v0.2.0: From Trending Topics to Storyboard Video

Trendlume v0.2.0 introduces the Trend Center, expanding your creative starting points. In addition to prompt expansion and manual script paste, you can now discover live topics across public feeds, review structured proposals, and convert them directly into editable video storyboards.

Highlights

  • Public Hotspot Dashboard (/trends): Aggregate trending topics from 7 platforms (Weibo, Douyin, Zhihu, Toutiao, Xiaohongshu, Bilibili, and Baidu) using resilient failover feeds without scraping anti-bot platforms.
  • AI Creative Proposals: Turn raw topic observations into reviewable angles, target hooks, and content outlines tailored to your project settings.
  • Idempotent Task Conversion: Approved proposals convert directly into standard video tasks, maintaining scene-level inspectability and retry controls.
  • Formal Content Modes: Unified visual source contracts for generated images, generated videos, stock footage, custom uploads, and kinetic typography.
  • ComfyUI Resilience: Automatic fallback to lightweight compatible workflows on prompt validation errors, paired with node-level diagnostic reporting.

Upgrade Instructions

  • Run database migrations: cd backend && uv run alembic upgrade head

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 12 Sep 08:01

initial release of Trendlume v0.1.0

Trendlume is an open-source AI short-video creation and publishing workbench
designed for creators, supporting full-lifecycle video generation from ideation
to platform distribution.

Key Features & Modules:

  • Dual Creation Modes: Topic-driven research & generation or custom script import
  • Storyboard Editor: Scene-level visual tweaking, duration tuning & prompt editing
  • Asset Pipeline: AI image/video generation, Pexels stock footage & local uploads
  • Audio & Subtitles: Edge-TTS / Volcengine TTS integration, multi-aspect auto alignment
  • Direct Publishing: Web QR code authorization & scheduled publishing to Douyin
  • Deployment: Standalone zero-dependency executable, Docker Compose & local dev setup
    Tech Stack:
  • Backend: FastAPI, SQLAlchemy, Alembic, SQLite, FFmpeg, Playwright, Edge-TTS
  • Frontend: Next.js 14, React 18, Tailwind CSS, TanStack Query, Lucide Icons