Skip to content

利用 AI 大模型和自动化工作流,根据主题或关键词一键生成高清短视频。Generate HD short videos from a topic or keyword with an automated AI workflow.

14.2 KB
Raw Blame

English | 中文

Highlights

  • Added official mpt pip packaging. Start the WebUI with mpt webui or generate videos with mpt cli.
  • Added static and slow zoom image effects for local and AI-generated images. Slow zoom remains the default.
  • Added Kimi Code subscription access with browser authorization, alongside improvements to the sign-in dialog, status updates, and cancellation.
  • Added an optional local SDXL image server for the OpenAI-compatible image material workflow, with idle GPU offloading.
  • Added Opper, onomeo, Muse Spark, and EmpirioLabs AI LLM providers, and completed Brazilian Portuguese WebUI translations.
  • Added diagnostic logging controls in the WebUI, including runtime log-level changes and opt-in API request tracing with redaction and size limits.

Fixes and Improvements

  • Fixed grayscale image brightness, transparent image backgrounds, zoom mask alignment, and subtitle background colors. Improved transitions on short clips, background music fade-out, and special-character handling in concatenation paths.
  • Isolated intermediate image files across concurrent jobs, improved image cache identity, and validated cached downloads. Completed videos are retained when another video in the same batch fails.
  • Fixed speech speed, volume, and audio container handling across several providers. Improved remote audio validation and preserved existing audio until replacement output is ready.
  • Improved subtitle timestamp rounding, long-word wrapping, Unicode handling, and pause-tag recognition. Symbol-only lines and separators no longer distort estimated subtitle alignment.
  • Fixed Chinese and Japanese fallback social titles using the entire script. Improved JSON extraction, reasoning removal, Markdown cleanup, and generated search-term validation.
  • Expanded supported local upload formats, including multi-picture camera JPEGs. Improved task history checks, deletion state, subtitle previews, and model connection error messages in the WebUI.
  • Added HEAD support for generated artifacts and improved byte-range downloads, validation errors, and Redis task updates and deletion safeguards.
  • Reduced unsafe speech resubmissions after lost responses or redirects, preserved Upload-Post tracking handles, and improved WaveSpeed submission error detection.
  • Updated Edge TTS to address 403 failures, constrained PyAV for Whisper compatibility, and fixed Windows cache cleanup and UTF-8 BOM handling in locale files.
  • Improved the agent helper's configuration handling and preflight checks, including workflows with prepared narration and visuals that do not need an LLM.

Usage and Upgrade Notes

  • The official PyPI package is mpt, maintained by this project, and requires Python 3.11 or later. Install or upgrade with python -m pip install --upgrade mpt; a virtual environment is recommended. The separately named MoneyPrinterTurbo package is a third-party distribution.
  • The pip installation stores configuration and generated files in your user directory by default; set MPT_HOME to choose another location. Source installations continue to use the project directory. When switching installation methods, migrate your configuration and required resources to the chosen runtime directory.
  • Image motion defaults to slow zoom for existing users. For a still image without transition effects, select Static for image motion and None for transitions. The setting is available in the WebUI, CLI, and API.
  • Kimi Code requires browser sign-in and uses your plan's usage limits. The other new LLM providers require their corresponding API credentials and configuration.
  • The local SDXL server requires separate dependencies, model downloads, and a running GPU service; it is not started automatically with MPT. Its default setup unloads Ollama models before image generation, so avoid concurrent LLM and image work on the same GPU. See the setup instructions.
  • API request tracing is off by default. Prompts and business parameters may appear in logs when enabled; turn tracing off after troubleshooting and check for private content before sharing logs. See the diagnostic logging guide.

What's Changed

Thanks to all contributors and everyone who reported issues and helped test these changes!

Full Changelog: v1.3.8...v1.3.9


重点更新

  • 支持通过 pip 安装官方 mpt 包,使用 mpt webui 启动界面,或通过 mpt cli 生成视频。
  • 新增图片动态效果选项,本地图片和 AI 生成图片均可选择“静止”或“缓慢放大”,默认保留缓慢放大。
  • 新增 Kimi Code 订阅接入,支持浏览器授权登录,并优化登录窗口布局、状态更新和取消操作。
  • 新增可选的本地 SDXL 生图服务,通过 OpenAI 兼容接口接入图片素材生成,支持空闲时释放 GPU 显存。
  • 新增 Opper、onomeo、Muse Spark 和 EmpirioLabs AI 大模型服务,完善巴西葡萄牙语界面翻译。
  • 新增诊断日志设置,支持在 WebUI 中即时调整日志等级,按需记录经过脱敏和长度限制的 API 请求参数。

修复与改进

  • 修复灰度图片亮度、透明图片背景、缩放蒙版对齐和字幕背景颜色问题;改善短片段转场、背景音乐淡出和拼接路径中的特殊字符处理。
  • 隔离并发任务的图片中间文件,改善图片缓存标识并校验下载缓存;批量生成部分失败时,保留已经完成的视频。
  • 修复部分配音服务的语速、音量和音频格式处理,完善远程音频校验;替换音频准备完成前保留已有文件。
  • 改善字幕时间戳舍入、长词换行、Unicode 字符和停顿标签识别,避免纯符号行和分隔线干扰字幕时间轴估算。
  • 修复中文、日文备用社交标题使用整篇脚本的问题;改善 JSON 提取、推理内容过滤、Markdown 文案清理和素材搜索词校验。
  • 补齐支持的本地上传格式,包括多图相机 JPEG;改善 WebUI 历史视频检查、任务删除状态、字幕预览和模型连接错误提示。
  • 支持生成文件的 HEAD 请求,完善范围下载、参数错误提示,以及 Redis 任务状态更新和删除保护。
  • 减少语音请求在响应丢失或重定向后被重复提交的风险,保留 Upload-Post 请求追踪信息,并改善 WaveSpeed 提交失败的识别。
  • 更新 Edge TTS,修复相关 403 问题;限制 PyAV 版本以兼容 Whisper,修复 Windows 缓存清理和带 UTF-8 BOM 的翻译文件读取。
  • 改善 Agent 辅助脚本的配置处理和启动检查,已准备好配音与画面的任务不再额外要求大模型服务。

使用与升级说明

  • 本项目维护的官方 PyPI 包名为 mpt,需要 Python 3.11 或更高版本。使用 python -m pip install --upgrade mpt 安装或升级,建议使用虚拟环境。PyPI 上另一个名为 MoneyPrinterTurbo 的包属于第三方发行版。
  • pip 安装版默认将配置和生成文件保存在用户目录,可通过 MPT_HOME 指定位置。源码版继续使用原有项目目录;切换安装方式时,请将配置和所需资源迁移到选定的运行目录。
  • 图片动态效果默认仍为缓慢放大,兼容已有使用方式。希望画面保持静止时,请同时选择静止和无转场;WebUI、CLI 和 API 均支持该设置。
  • Kimi Code 需要浏览器登录,并使用对应订阅额度;其他新增大模型服务需要配置各自的 API 凭据和服务参数。
  • 本地 SDXL 服务需要单独安装依赖、下载模型并启动 GPU 服务,不会随主程序自动运行。默认配置会在生图前卸载 Ollama 模型,请避免在同一 GPU 上同时运行大模型和生图任务。部署方法见本地生图服务文档。
  • API 请求追踪默认关闭。开启后,提示词和业务参数可能出现在日志中;排查结束后请关闭,分享日志前检查私人内容。详见诊断日志文档。

主要变更

感谢所有贡献代码、反馈问题和参与测试的朋友!

完整变更记录:v1.3.8...v1.3.9