Skip to content

FreeVideo v0.3.23

Choose a tag to compare

View all tags
@github-actions github-actions released this 10 Oct 08:53
· 302 commits to main since this release
v0.3.23
0af617e
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.
FreeVideo

Download / 下载
Windows 10/11 · NVIDIA GPU FreeVideo.exe
macOS 14+ · Apple silicon FreeVideo-Mac-arm64.dmg
Linux x86_64 · NVIDIA GPU FreeVideo-Linux-x86_64.AppImage

Windows: run FreeVideo.exe, choose an existing ComfyUI folder or install a new one, then click Install & launch. Offline packages are on Quark.

macOS: open the DMG and drag FreeVideo.app into Applications. If macOS asks you to confirm the first launch, see the first-open steps.

Linux: make the AppImage executable (chmod +x FreeVideo-Linux-x86_64.AppImage) and open it, as on Windows. On a server or from a terminal, see Linux.

Windows： 运行 FreeVideo.exe，选择已有的 ComfyUI 或安装新的 ComfyUI，点击 安装并启动。离线包见夸克网盘。

macOS： 打开 DMG，将 FreeVideo.app 拖入「应用程序」。首次打开如需确认，请参考首次打开步骤。

Linux： 给 AppImage 加上可执行权限（chmod +x FreeVideo-Linux-x86_64.AppImage）后打开，用法与 Windows 相同。在服务器或终端中使用请参考 Linux 说明。

What's new

Generation and installation progress show one overall percentage with the details that matter, such as the current step and how much has been downloaded; the workspace fills the window; RTX 50 GPUs no longer keep pausing when the driver cache is full, and the first video after a start waits less.

  • The first video after FreeVideo starts waits up to about 15 seconds less for the text encoder. On RTX 40 and RTX 50 cards under Windows, and on Linux, more of the video decoder stays on the graphics card while decoding (all of it on 12 GB cards), so decoding is faster.
  • On Windows with RTX 50 GPUs, generation no longer keeps pausing once the graphics driver's compile cache is full.
  • On Windows, when the video model fits neither in GPU memory nor in RAM, generation waits less for the drive. When another program takes GPU memory during generation, decoding no longer starts over because it ran out of memory.
  • Generation progress shows one overall percentage, the current step and the elapsed time, with a small bar under slow stages, and the remaining time appears after the first two steps; during installation, repair and updates, a download shows what is being downloaded, how much and how fast. Loading the video model no longer hangs when the graphics driver does not answer, and after a long time without progress the page says that generation is still running and offers the report.
  • Error cards now say in one sentence what happened and what to do, with the raw error under Technical details; cases such as a reference video that is too short or a LoRA that cannot be used with the int8 model yet each have their own message. The compatibility setting now describes what each level does.
  • The creative workspace and the Creations window now fill the browser window, so the video preview grows with it. On windows at least 1680 px wide, a Recent videos column appears on the right; clicking a video plays it in the preview, and a button in the preview header hides the column.
  • A custom step count now downloads only a sampling table the first time, up to 1.2 GB, instead of 26 GB of model weights. Any download still needed, as for some LoRAs on FP8, shows its progress, size and speed instead of an unchanging “Loading model”.
  • On NVIDIA graphics cards with 8 GB or more, prompt enhancement needs about 4 GiB of free RAM instead of 11 GiB, and on 8 GB and 10 GB cards a rewrite takes tens of seconds instead of several minutes. While it runs, the status line shows how long it has taken.
  • Each saved video now takes far less disk space, about 40 MB instead of about 790 MB for a 10-second 1344 × 768 video, because the uncompressed video and audio are no longer kept. The first start after updating deletes them from earlier videos as well, and previews can still continue with Run second pass.
  • A new Repair button in the launcher fixes an installation in one step and keeps models and videos; it appears on the error card when the launcher finds a problem it can repair, and in Settings › Environment. A repair that only adds a few packages no longer asks for as much free disk space as a new installation.
  • On a Mac, when unified memory runs short, the second pass switches to a computation that uses less memory; if it still runs short, the first pass is kept and only the second pass runs again. When less memory is available than a video needs, FreeVideo says so as soon as generation starts.
  • Fixes: on Linux, desktops without xdg-open can open FreeVideo in the browser, and the background service starts when the folder name contains %, $, quotes or backslashes; when port 8188 is in use, the launcher starts ComfyUI on a free port and shows the new address; and Captions only in the image editor no longer leaves white specks when it sweeps a caption in several strokes. With a self-installed ComfyUI updated to a recent version, each video appears once in Assets instead of twice, and paths with hidden folders no longer show <PATH><PATH> in the launcher terminal and diagnostic reports.

更新内容

生成和安装的进度只显示一个总百分比，并给出在做什么、下了多少等必要的细节；创作面板铺满整个窗口；RTX 50 显卡不再因驱动缓存写满而反复停顿，启动后的第一个视频等待更短。

  • 启动后的第一个视频等文本编码器的时间更短，最多少等约 15 秒。在 Windows 上的 RTX 40、RTX 50 显卡以及 Linux 上，解码时更多的解码器留在显卡上（12 GB 显卡可以全部留下），解码更快。
  • 在使用 RTX 50 系显卡的 Windows 电脑上，显卡驱动的编译缓存写满后，生成时不再反复停顿。
  • 在 Windows 上，显存和内存都放不下整个视频模型时，生成时等硬盘的时间更短。生成期间其他程序占用了显存时，解码不再因显存不足而从头再来一遍。
  • 生成进度只显示一个总百分比、正在做的那一步和已用时间，慢的阶段下面多一条小进度条，跑完两步后再显示预计剩余时间；安装、修复和更新时，下载会写明在下什么、下了多少和速度。加载视频模型时不再因显卡驱动没有响应而一直卡住；长时间没有新的进度时，页面会说明生成仍在进行，并可以下载报告。
  • 出错时，卡片用一句话说明发生了什么、该怎么做，原始报错收进“技术详情”；参考视频太短、LoRA 暂时不能用于 int8 模型等情况各有自己的说明。兼容性设置也改成说明每一档的效果。
  • 创作面板和作品窗口现在铺满浏览器窗口，视频预览随窗口一起变大。窗口宽 1680 像素以上时，右侧会多出“最近作品”一栏，点击其中的视频会在预览中播放，预览标题行的按钮可以收起这一栏。
  • 选用自定义步数时，第一次只需下载一个采样表（最多 1.2 GB），不再先下载 26 GB 的模型权重。仍需下载时（例如 FP8 上的部分 LoRA），会显示进度、大小和速度，不再一直停在“加载模型”。
  • 在 8 GB 及以上的 NVIDIA 显卡上，提示词增强只需约 4 GiB 可用内存（原来要 11 GiB），8 GB 和 10 GB 显卡上的改写从几分钟缩短到几十秒。改写时，状态栏会显示已用时间。
  • 保存的视频占用的磁盘空间大幅减少：10 秒、1344 × 768 的视频从约 790 MB 降到约 40 MB，因为不再保留未压缩的画面和音频。更新后第一次启动时，也会删除之前视频里的这些文件；预览仍然可以点“继续二采”。
  • 启动器新增“修复”按钮，一步修好安装，模型和视频都会保留；发现可以修复的问题时，出错卡片上会出现这个按钮，“设置 → 环境”里也有。只需补装几个包的修复，不再要求和全新安装一样多的空闲磁盘空间。
  • 在 Mac 上，统一内存不足时，第二遍会改用占用内存更少的计算方式；仍然不足时，保留第一遍的结果，只重新运行第二遍。可用内存低于所需时，生成开始时就会给出提示。
  • 修复：Linux 上，没有 xdg-open 的桌面也能在浏览器中打开 FreeVideo，文件夹名带 %、$、引号或反斜杠时后台服务也能启动；本机 8188 端口被占用时，启动器会改用空闲端口并说明新地址；图片编辑器的“只去字幕”分几笔扫过字幕时不再留下白点。自行将 ComfyUI 更新到较新版本后，每个视频在“资产”中不再出现两条记录；启动器终端和诊断报告里，隐去文件夹的路径不再显示为 <PATH><PATH>。
sha256 
75fe6a25c4bdcaab698f8092d2f97d8065ef0b33dcfc913e79a4bf4035396028  FreeVideo.exe
302ad72e28593013f1900016aca4e308b59fe5537ef9358b364f658fbc5ba5db  FreeVideo-Mac-arm64.dmg
2ebd84cad67ad3c55a2d4c38ee2aef313f3b0ae21b6194db3df98a1c5cb8de9f  FreeVideo-Linux-x86_64.AppImage

Discord · QQ Group / QQ 群 · WeChat Group / 微信群

Assets 9
Loading