Make videos on the computer you already own. FreeVideo runs MiniMax H3 in as little as 8 GB of VRAM and 16 GB of RAM, and adapts its acceleration path to your hardware.
Release 0.4.19
Bowen Xue8310c35ffc54
diff --git a/freevideo_engine/release_notes.json b/freevideo_engine/release_notes.jsonindex 9ff162f..dbfb481 100644--- a/freevideo_engine/release_notes.json+++ b/freevideo_engine/release_notes.json@@ -1,41 +1,70 @@{"schema": 1,- "product_version": "0.4.18",+ "product_version": "0.4.19","en": {- "summary": "FreeVideo can be installed into deeply nested folders on Windows; many failure messages state the cause and the next step; Windows computers with little free memory can complete generation; consecutive requests run faster on computers with ample memory.",+ "summary": "Several pictures can be deleted from the image history at once; prompt enhancement states the reason when it cannot finish.","changes": [- "Install: FreeVideo can be installed into deeply nested folders on Windows; when an update cannot be installed and the previous version is restored, the notice follows the interface language.",- "Launcher: errors state the step that failed and what to do next, and Report a problem includes a line about the environment; when the saved installation folder cannot be opened, Start opens the setup page with the saved paths and an explanation; a graphics driver older than 580 gets a notice with the installed version and the update steps; a damaged launcher settings file is kept as a copy before it is replaced; the launcher keeps working when the prompt model file list is missing, and that row appears once the list is available.",- "Model downloads: slow sources are retried with a longer time limit, and the last failure reason of each source and route is listed; when a source has problems, the launcher shows a dedicated card with a button to switch sources; the image model has its own download list with links and checksums, exported to a separate folder.",- "Failure messages: a failed video generation names the stage where it stopped (loading, sampling or decoding); failure feedback starts with the last error; automatic subtitles, subtitle downloads, media import and subtitle import in the editor state the reason when they fail, and GBK-encoded SRT subtitles can be imported; image generation, adding workflows and share export state when the disk is full; an unreadable queued reference image prompts for a different image; a failed part of a long video shows a short status, with details in the error card and the name of any missing file.",- "Generation: fixed an error when submitting a generation; when first/last frames and references cannot be combined, the failure card says which media to turn off; Generate with AI explains when the media list is full at 32 items; the estimate before generation marks first-time preparation, and a slow estimate gives a reason and a remedy; a queued request shows its wait in minutes or the number of requests ahead; a queued export shows Queued on its button; LoRA files in more naming styles are recognized, and an acceleration LoRA made for the original model gets its own explanation.",- "Prompt enhancement: scenes with audio, connected nodes or videos can be enhanced, with audio and connected nodes skipped and one frame read from each video; with reference images, a rewrite that leaves out an image reference is sent back to the model once to add it.",- "Reference images: uploading and reading show Uploading reference image and Reading reference image N%; the + button in the editor shows a spinner while loading; library import shows uploading, queued and processing in turn, and finishes once progress reaches 100%.",- "Autopilot: it can be stopped while waiting for a reply and shows its current action; after three minutes without a progress update it says generation is still running; a background tab title starts with Confirmation needed, Failed or Done; Edit prompt puts the summary back in the input box; the confirmation card shows the size and duration; when the entry check fails, the entry stays and the check can be retried; an unreadable saved conversation or film leaves the rest available; failure reasons are readable in the English interface.",- "Multi-shot films: for long videos and multi-shot films made by Autopilot, music is added on request: when requested, one continuous score is laid over the whole film after the shots are joined, and a music file from the user can be used directly.",- "Create and Your creations: saving or loading a prompt can be retried after a failure; key presses stay inside the creation panel; in the image editor, the History entry stays visible after a picture is placed from the history; the end of the Your creations list shows how many records cannot be read.",- "Memory and speed: when a Windows computer is refused only because of the default memory reserve, part of the reserve is used, keeping at least 1 GiB, so generation can start with about 5 GiB free; on computers with ample memory, consecutive requests skip reloading the models and run faster.",- "Mac: encoding with many reference images is much faster; accelerated computation falls back to the standard method when its results look wrong; after a one-click update, a new version that starts slowly is kept."+ "Install: setup no longer stops when some graphics card details cannot be read; command-line generation works on Windows when FreeVideo is installed in a very deep folder.",+ "Image history: after Select, several pictures can be checked and deleted at once, with a second click to confirm; pictures that are in use or cannot be deleted stay in place with the reason shown.",+ "Prompt enhancement: when enhancement cannot finish and the original prompt is used, the reason is shown; a rewrite that lacks the video or audio description is sent back to the model to add it.",+ "Editor: Extend on a finished video without a description asks for a description first instead of showing an English validation error; when the description box is cleared, the description carried over for that segment is used.",+ "The Linux package now includes curl, so installation and downloads work on systems without it.",+ "Download list: the check no longer hashes every model file again each time it runs; files already hashed are skipped while their size, times and inode are unchanged.",+ "Keeps the failed phase, step count, error type and memory readings of a GPU error that raised the compatibility level, so a diagnostic report sent after a later run still shows where it stopped.",+ "The chat no longer shows the notice for an edited waiting message as an error with a report button; it is a plain status line."]},"zh": {- "summary": "Windows 上可以安装到层级较深的文件夹;多处失败提示写明原因和下一步操作;可用内存较少的 Windows 电脑可以继续完成生成;内存宽裕的电脑连续生成时速度更快。",+ "summary": "生图历史可以一次删除多张;提示词增强未能完成时写明原因。","changes": [- "安装:Windows 上可以安装到层级较深的文件夹;更新未能装上并换回旧版时,提示按界面语言显示。",- "启动器:出错时写明出错的步骤和下一步操作,“报告问题”附带一行环境信息;已保存的安装文件夹无法打开时,点击“启动”会进入安装页,显示已保存的路径和说明;显卡驱动版本低于 580 时,显示当前驱动版本和更新步骤;启动设置文件损坏时,先保留一份副本再保存;提示词模型的文件清单缺失时,启动器照常工作,相应一行在清单可用后显示。",- "模型下载:下载源较慢时以更长的时限重试,并分别列出每个下载源和线路最近一次失败的原因;下载源出现问题时,启动器显示专门的卡片和切换下载源的按钮;图像模型有单独的下载清单,列出链接和校验值,并导出到单独的文件夹。",- "失败提示:视频生成失败时写明出错的阶段(载入、采样或解码);生成失败后的反馈以最后一条错误开头;剪辑页的自动字幕、字幕下载、素材导入和字幕导入失败时写明原因,GBK 编码的 SRT 字幕可以导入;图像生成、补写工作流和分享导出遇到磁盘已满时写明原因;排队中的参考图无法读取时,提示更换图片;长视频的某一段失败时显示简短状态,详细内容在错误卡片中,并写出缺失文件的名称。",- "生成:修复了部分情况下提交生成时报错的问题;首尾帧与参考素材不能同时使用时,失败卡片说明原因并提示关闭哪一类素材;素材列表已满 32 项时,“AI 生成”说明原因;生成前的时间估计注明“含首次准备”,预计较慢时给出一个原因和一个处理办法;排队时显示“约 N 分钟后开始”或“前面还有 N 个”;导出排队时按钮显示“排队中”;更多命名方式的 LoRA 文件可以正确识别,为原版模型制作的加速 LoRA 会显示专门说明。",- "提示词增强:场景中有音频、连接节点或视频时也可以增强,音频和连接节点会被跳过,视频读取一帧画面;带参考图时,改写结果遗漏图片引用,会请模型补上一次。",- "参考图:上传和读取时显示“正在上传参考图”和“正在读取参考图 N%”;剪辑页的“+”按钮在加载时显示转圈;素材库导入依次显示上传中、排队中和处理中,进度到 100% 后直接完成。",- "AI 全自动:等待回复时可以停止,并显示当前动作;三分钟没有进度更新时,提示仍在生成中;后台标签页的标题前显示“待确认”“失败”或“已完成”;“改描述”会把摘要放回输入框;确认卡片显示尺寸和时长;入口检查失败时保留入口,可以重试;某个已保存的对话或影片无法读取时,其余内容照常可用;英文界面下的失败原因清楚可读。",- "多镜头作品:AI 全自动制作长视频和多镜头作品时,配乐按需添加:需要配乐时,在各镜头拼接完成后为整片铺一段连贯的配乐,用户提供的音乐文件也可以直接使用。",- "创作面板与我的作品:提示词保存或读取失败时可以重试;按键操作只作用于创作面板;图像编辑中从历史记录放入画布后,“历史”入口保持可见;“我的作品”列表末尾显示无法读取的记录数量。",- "内存与速度:Windows 电脑只因默认预留空间不足而无法开始生成时,会使用部分预留空间(至少保留 1 GiB),可用内存约 5 GiB 时也可以开始生成;连续生成多个请求时,内存宽裕的电脑无需重复加载模型,速度更快。",- "Mac:使用大量参考图时,编码速度明显提升;加速计算的结果异常时,自动改用标准计算;一键更新后,启动较慢的新版本同样会保留。"+ "安装:部分显卡信息读不到时,安装不再中断;Windows 上 FreeVideo 安装在很深的文件夹时,命令行生成可以正常运行。",+ "生图历史:点“选择”后可以勾选多张一次删除,删除前需要再次确认;被占用或无法删除的图片会保留并写明原因。",+ "提示词增强:增强未能完成、改用原提示词时,写明原因;改写结果缺少视频或音频描述时,会请模型补上。",+ "剪辑页:在没有描述的成片后“接着生成”时,会提示先写一句描述,不再显示英文校验错误;描述框被清空时,沿用这一段原有的描述。",+ "Linux 安装包自带 curl,系统未安装 curl 时也能正常安装和下载。",+ "下载清单:检查不再每次重新校验所有模型文件;已经校验过的文件在大小、时间和 inode 都没变时直接跳过。",+ "保留导致兼容档位升高的显卡错误的出错阶段、步数、错误类型和内存读数,之后一次运行发送的诊断报告仍能看出它停在哪里。",+ "对话中修改了正在等待 Agent 的消息后,提示不再按错误显示、不再带“报告问题”按钮,改为普通状态提示。"]},"history": [+ {+ "product_version": "0.4.18",+ "en": {+ "summary": "FreeVideo can be installed into deeply nested folders on Windows; many failure messages state the cause and the next step; Windows computers with little free memory can complete generation; consecutive requests run faster on computers with ample memory.",+ "changes": [+ "Install: FreeVideo can be installed into deeply nested folders on Windows; when an update cannot be installed and the previous version is restored, the notice follows the interface language.",+ "Launcher: errors state the step that failed and what to do next, and Report a problem includes a line about the environment; when the saved installation folder cannot be opened, Start opens the setup page with the saved paths and an explanation; a graphics driver older than 580 gets a notice with the installed version and the update steps; a damaged launcher settings file is kept as a copy before it is replaced; the launcher keeps working when the prompt model file list is missing, and that row appears once the list is available.",+ "Model downloads: slow sources are retried with a longer time limit, and the last failure reason of each source and route is listed; when a source has problems, the launcher shows a dedicated card with a button to switch sources; the image model has its own download list with links and checksums, exported to a separate folder.",+ "Failure messages: a failed video generation names the stage where it stopped (loading, sampling or decoding); failure feedback starts with the last error; automatic subtitles, subtitle downloads, media import and subtitle import in the editor state the reason when they fail, and GBK-encoded SRT subtitles can be imported; image generation, adding workflows and share export state when the disk is full; an unreadable queued reference image prompts for a different image; a failed part of a long video shows a short status, with details in the error card and the name of any missing file.",+ "Generation: fixed an error when submitting a generation; when first/last frames and references cannot be combined, the failure card says which media to turn off; Generate with AI explains when the media list is full at 32 items; the estimate before generation marks first-time preparation, and a slow estimate gives a reason and a remedy; a queued request shows its wait in minutes or the number of requests ahead; a queued export shows Queued on its button; LoRA files in more naming styles are recognized, and an acceleration LoRA made for the original model gets its own explanation.",+ "Prompt enhancement: scenes with audio, connected nodes or videos can be enhanced, with audio and connected nodes skipped and one frame read from each video; with reference images, a rewrite that leaves out an image reference is sent back to the model once to add it.",+ "Reference images: uploading and reading show Uploading reference image and Reading reference image N%; the + button in the editor shows a spinner while loading; library import shows uploading, queued and processing in turn, and finishes once progress reaches 100%.",+ "Autopilot: it can be stopped while waiting for a reply and shows its current action; after three minutes without a progress update it says generation is still running; a background tab title starts with Confirmation needed, Failed or Done; Edit prompt puts the summary back in the input box; the confirmation card shows the size and duration; when the entry check fails, the entry stays and the check can be retried; an unreadable saved conversation or film leaves the rest available; failure reasons are readable in the English interface.",+ "Multi-shot films: for long videos and multi-shot films made by Autopilot, music is added on request: when requested, one continuous score is laid over the whole film after the shots are joined, and a music file from the user can be used directly.",+ "Create and Your creations: saving or loading a prompt can be retried after a failure; key presses stay inside the creation panel; in the image editor, the History entry stays visible after a picture is placed from the history; the end of the Your creations list shows how many records cannot be read.",+ "Memory and speed: when a Windows computer is refused only because of the default memory reserve, part of the reserve is used, keeping at least 1 GiB, so generation can start with about 5 GiB free; on computers with ample memory, consecutive requests skip reloading the models and run faster.",+ "Mac: encoding with many reference images is much faster; accelerated computation falls back to the standard method when its results look wrong; after a one-click update, a new version that starts slowly is kept."+ ]+ },+ "zh": {+ "summary": "Windows 上可以安装到层级较深的文件夹;多处失败提示写明原因和下一步操作;可用内存较少的 Windows 电脑可以继续完成生成;内存宽裕的电脑连续生成时速度更快。",+ "changes": [+ "安装:Windows 上可以安装到层级较深的文件夹;更新未能装上并换回旧版时,提示按界面语言显示。",+ "启动器:出错时写明出错的步骤和下一步操作,“报告问题”附带一行环境信息;已保存的安装文件夹无法打开时,点击“启动”会进入安装页,显示已保存的路径和说明;显卡驱动版本低于 580 时,显示当前驱动版本和更新步骤;启动设置文件损坏时,先保留一份副本再保存;提示词模型的文件清单缺失时,启动器照常工作,相应一行在清单可用后显示。",+ "模型下载:下载源较慢时以更长的时限重试,并分别列出每个下载源和线路最近一次失败的原因;下载源出现问题时,启动器显示专门的卡片和切换下载源的按钮;图像模型有单独的下载清单,列出链接和校验值,并导出到单独的文件夹。",+ "失败提示:视频生成失败时写明出错的阶段(载入、采样或解码);生成失败后的反馈以最后一条错误开头;剪辑页的自动字幕、字幕下载、素材导入和字幕导入失败时写明原因,GBK 编码的 SRT 字幕可以导入;图像生成、补写工作流和分享导出遇到磁盘已满时写明原因;排队中的参考图无法读取时,提示更换图片;长视频的某一段失败时显示简短状态,详细内容在错误卡片中,并写出缺失文件的名称。",+ "生成:修复了部分情况下提交生成时报错的问题;首尾帧与参考素材不能同时使用时,失败卡片说明原因并提示关闭哪一类素材;素材列表已满 32 项时,“AI 生成”说明原因;生成前的时间估计注明“含首次准备”,预计较慢时给出一个原因和一个处理办法;排队时显示“约 N 分钟后开始”或“前面还有 N 个”;导出排队时按钮显示“排队中”;更多命名方式的 LoRA 文件可以正确识别,为原版模型制作的加速 LoRA 会显示专门说明。",+ "提示词增强:场景中有音频、连接节点或视频时也可以增强,音频和连接节点会被跳过,视频读取一帧画面;带参考图时,改写结果遗漏图片引用,会请模型补上一次。",+ "参考图:上传和读取时显示“正在上传参考图”和“正在读取参考图 N%”;剪辑页的“+”按钮在加载时显示转圈;素材库导入依次显示上传中、排队中和处理中,进度到 100% 后直接完成。",+ "AI 全自动:等待回复时可以停止,并显示当前动作;三分钟没有进度更新时,提示仍在生成中;后台标签页的标题前显示“待确认”“失败”或“已完成”;“改描述”会把摘要放回输入框;确认卡片显示尺寸和时长;入口检查失败时保留入口,可以重试;某个已保存的对话或影片无法读取时,其余内容照常可用;英文界面下的失败原因清楚可读。",+ "多镜头作品:AI 全自动制作长视频和多镜头作品时,配乐按需添加:需要配乐时,在各镜头拼接完成后为整片铺一段连贯的配乐,用户提供的音乐文件也可以直接使用。",+ "创作面板与我的作品:提示词保存或读取失败时可以重试;按键操作只作用于创作面板;图像编辑中从历史记录放入画布后,“历史”入口保持可见;“我的作品”列表末尾显示无法读取的记录数量。",+ "内存与速度:Windows 电脑只因默认预留空间不足而无法开始生成时,会使用部分预留空间(至少保留 1 GiB),可用内存约 5 GiB 时也可以开始生成;连续生成多个请求时,内存宽裕的电脑无需重复加载模型,速度更快。",+ "Mac:使用大量参考图时,编码速度明显提升;加速计算的结果异常时,自动改用标准计算;一键更新后,启动较慢的新版本同样会保留。"+ ]+ }+ },{"product_version": "0.4.17","en": {