OpenFilm 0.1.6

Patch Changes

• #32 60cb113 Thanks @PazChen! - Settings → About in the desktop app shows the version and its update without going to the menu: it looks for one on opening (unless it looked a few minutes ago), shows "Up to date" with when it last checked, or the new version downloading with its progress, then "Restart to update" once it is ready (it installs at the next quit too). A failed check says why, with Try again. A build run from source says it doesn't update itself. The menu's Check for Updates… is still there.

• #32 60cb113 Thanks @PazChen! - Desktop: when a background task of the agent's own ends after its turn (a command Claude Code ran in the background, a subagent) and it takes it up by itself, the chat shows that work as a turn of its own, under "A background task finished; the agent went on", with its log; before, it was lost. It ends with the agent's result, and as soon as you write. A ring beside the composer's tools shows how full the agent's context is in the conversation, with the numbers on hover. The / menu has the agent's commands from the start (they were there only after a first message), keeps them between launches, finds them by name, and shows each name whole.

• #32 60cb113 Thanks @PazChen! - Desktop: the agent's own commands in the chat, as in its own: type / at the start of a message for the list Claude Code or Codex gives (their built-in commands and your own skills and custom commands), with what each does; it goes to the agent as typed.

• #32 60cb113 Thanks @PazChen! - Desktop: each conversation in the chat is its own session of the agent's, and it goes on where it stopped after a restart, a reload, or switching conversations: the chat keeps the agent's own session id and resumes it, as claude --resume or codex resume would. A new conversation starts a new session (it carried the last one's context before). A session left idle for 20 minutes closes, and its next message resumes it. A picture attached to a message is sent to the agent beside the text, as one pasted in its own chat is, besides being put in the project. The message goes as written (the first one no longer starts with a line about the language), so a slash command typed first works.

• #32 60cb113 Thanks @PazChen! - Desktop: the chat's agents are told to use the app's own openfilm command, not npx openfilm (which an installed OpenFilm skill says), so they work on the Studio in the window at its version; and only not to open Studio in a browser, rather than never to open a browser.

• #32 60cb113 Thanks @PazChen! - Desktop: the model menu has the models the agent offers now. It asks each agent for its options once a launch (what it had from before shows meanwhile), so an updated Claude Code's models replace the ones an earlier version had; and a pick it no longer offers is dropped for the agent's own default.

• #32 60cb113 Thanks @PazChen! - Desktop: write to Claude Code or Codex while it works, as in its own chat. The composer's button turns from Stop to Send once something is written, and the message goes into the running turn (the agent reads it at its next step); it shows as a message of its own, with what the agent does from then on under it. An agent that cannot take a message mid-turn gets it as soon as the turn ends, as before. Esc still stops.

• #32 60cb113 Thanks @PazChen! - Desktop: less work around a running agent. Studio no longer redraws the film when the agent writes a file no page loads (its notes, scripts, logs). The chat around a turn no longer re-renders for every word the agent streams. A model menu whose agent did not answer is asked again only when the menu opens or on Retry, not on every render (each try was a whole session of the agent's). A turn's log open while it runs is read again only once it has grown.

• #32 60cb113 Thanks @PazChen! - Settings → Agents shows whose account each of your agents uses (organization, email and plan for Claude Code; the ChatGPT account for Codex). The agents use the account signed in on your computer: to use another, switch accounts in the agent itself, then disconnect and connect it again in OpenFilm.

• #32 60cb113 Thanks @PazChen! - Desktop: Claude Code and Codex in the chat run as they do in your terminal. They get everything your shell sets (an API key or a gateway such as ANTHROPIC_BASE_URL, an MCP server's token, your certificates, your ssh agent), where before they got only its PATH. Their model and reasoning effort are your own config's until you pick another in the menu (the chat no longer sets High). Codex keeps your own developer_instructions, with OpenFilm's after them. Your ffmpeg and Playwright browsers come first for the commands they run; the app's ffmpeg and browser go only to the openfilm command (and ffmpeg last on the PATH, for a computer that has none). Electron's ELECTRON_RUN_AS_NODE no longer reaches their commands. A session starts without waiting on your shell to be read again.

• #32 60cb113 Thanks @PazChen! - Desktop: Codex, Claude Code and the other agents are found on Windows where they were missed. The PATH is read as Windows has it now, so an agent installed after signing in to Windows is found without signing out; the folders of pnpm, Volta, nvm-windows, Yarn, Bun, Scoop, Chocolatey and winget are searched, with npm's own global folder; and the shims npm, pnpm and Yarn write are read correctly (npm's current shim was taken for its node.exe). A shim that cannot be read is started through cmd.exe. On a Mac too, the login shell's PATH is read again while the app runs, so an agent installed meanwhile is found without restarting the app.
  Settings → Agents: an agent not found has Locate…, which opens the file dialog; the program chosen is tried (it has to answer --version) and used from then on, until Find it automatically. "Why isn't it found?" says what to do (installed only in WSL, files that cannot start, nothing found) and copies a report of every folder searched, to send with an issue.

• #32 60cb113 Thanks @PazChen! - Desktop: the model menu has no Mode: every agent works with full access, and a plan mode picked there before is no longer applied.

• #32 60cb113 Thanks @PazChen! - Settings → Agents says "Not signed in" for an agent that has never been signed in on the computer, instead of "Signed out".

• #32 60cb113 Thanks @PazChen! - Settings → Agents shows what is left of Codex's plan without opening it: each window (five hours, this week) with a bar of how much is used and when it resets, as Codex itself says, and its account's email and plan. Claude Code's is not shown: its own report said 0% of a Team account the Claude app showed half used.

• #24 f91edd3 Thanks @PazChen! - Studio: changing the canvas's shape fits what the film shows to the new canvas, centered, with at least one side meeting its edge. Shapes picked one after another no longer shrink the film further each time, and a film already shrunk grows back to the edge on the next change.
  The ratio menu lists plain ratios (and 21:9), with "Original" first and nothing else, and the inspector drops its note about the canvas. A project opens with the media, the viewer, the inspector and the timeline shown, and the layers hidden.

• #26 e0a80d5 Thanks @PazChen! - Studio: a canvas change is applied...

OpenFilm 0.1.5

Patch Changes

• #22 387d392 Thanks @PazChen! - Exact frames (render, look, Studio's exports): a video's frame on screen is told by how long its frames last, which the host reads with ffprobe. A video woken ahead while hidden, then shown, no longer leaves its old frame in the picture: shown again after the seek, it was taken for the sought one, and an export could hold a frame from the clip's start. Footage whose next frame is the same source frame (24 fps footage in a 30 fps export) no longer waits seconds for a frame that never comes: such a film exports several times as fast.

• #18 10709ae Thanks @PazChen! - Desktop: a new version, once downloaded, is offered in the top bar beside the projects button (Restart to update) instead of in the chat, where a closed chat hid it. A restart that has to wait (an agent working, an export running) says why beside it.

• #8 6a1d9b8 Thanks @lubobill1990! - Studio: Azure Speech (Azure AI Speech) can be connected under Settings → Providers for voice-over with its neural voices. The key carries the resource's region (<region>:<key>, or AZURE_SPEECH_KEY with AZURE_SPEECH_REGION), and one without a region is refused where it is pasted; Azure times no words, so subtitles spread the text over the line, as with OpenAI's voices.

• #5 e4679d2 Thanks @PazChen! - Desktop: your own key can name several models (Settings → Agents, separated by commas), each checked for reading pictures; the chat's menu switches between them. The menu no longer stays on "Loading" for an agent with nothing to choose.

• #15 12d34cd Thanks @PazChen! - Faster exports that use more of the computer. A video whose key frames are far apart (most camera, phone and stock footage) gets a copy that seeks fast before a render, kept in the project's cache; the pages draw from the copy and the sound still comes from the original. A film with footage renders about three times as fast. Each of a render's pages gets a browser of its own, and a render uses a browser for every two threads with 3 GB of memory each, up to eight (it was up to four, sharing one browser). H.264 and HEVC below master go through the graphics chip's encoder when it has one that works: the Mac's for HEVC too, and on Windows NVIDIA's, Intel's, AMD's or Windows' own. Each one is tried on a moment of video first, and an export it fails partway is made again without it.

• #13 74fd6d1 Thanks @PazChen! - Export: in the desktop app, the export tray (what an export is doing, how far it has got) sits beside the chat instead of under it; pressing Export seemed to do nothing while the film rendered unseen. An ffmpeg that ends before the last frame (not found, an encoder that fails to start) fails the export with ffmpeg's own error; the export used to stay at "preparing" forever. The export dialog finds an ffmpeg that arrives after Studio started (the desktop app downloads it on a first start); every format used to say this computer's ffmpeg could not make it until Studio restarted.

• #11 70602ab Thanks @PazChen! - Studio: a key that Settings → Providers refuses says why under its own field, not at the bottom of the list past every provider.

• #9 acae857 Thanks @PazChen! - Studio: Brave Search, Groq, Pexels, Pixabay and Tavily show their own logos in Settings → Providers, in place of a letter or a glyph of ours.

• #5 e4679d2 Thanks @PazChen! - Studio: the subtitle panel's language and typeface menus are Studio's own, in its colors, grouped and with the chosen one checked, in place of the browser's list, which was light in a dark Studio and read as if every choice were disabled.

• #12 7f21b61 Thanks @PazChen! - Exact frames (render, look, Studio's exports): a video sought as it comes on waits for its own frame on screen. On a slow machine the old frame, shown again as the video turned visible, could be taken for it, and look reported the same t drawing two different pictures.

OpenFilm 0.1.4

Patch Changes

• f486188: Windows: an ffmpeg that is stopped (a cancelled render or export, a mix that runs too long) ends with the real ffmpeg
  behind a Chocolatey or Scoop shim; a name with a : is cleaned on import, rename and new folders as on macOS and
  Linux, not refused; Studio's supervisor opens the null device by its Windows name.
• A render or look frame of a video clip woken ahead of its cut waits for the seek it was woken with before its own:
  on a slow machine the frame showed the clip's first second instead of the one asked for, and look said the same t
  drew two pictures.

OpenFilm 0.1.3

The first release from the open-source repository: the CLI, Studio and the desktop app share this version.

• Media comes only from services you connect with your own key, in Studio's Settings → Providers or from the environment. There are more of them, at least one for every kind of media: ElevenLabs, OpenAI, Google Gemini and fal.ai, now with Groq (fast transcripts with Whisper, voice-over with Orpheus), Pexels and Pixabay (image search), Brave Search and Tavily (web search and reading pages); fal.ai also makes music and sound effects. openfilm get names the services that make each kind you have not connected yet.
• Settings → Providers: every service is Connect or Disconnect. A key pasted in Studio wins over one in the environment, which connects its service unless disconnected.
• Studio's Projects page starts from an existing film: Import from a folder, or from GitHub (a repository or a folder in one, without its git history), and a row of example films. openfilm open <github url> does the same. A project whose folder moved can be located again.
• A Studio started by openfilm open that ends unexpectedly starts again by itself, on the same address; pages reconnect.
• Settings → General shows Studio's version and checks npm for a newer one.
• The Desktop dialog shows what the app does and the agents it works with, and downloads the installer for this computer.
• The desktop app's chat runs your own agents (Codex, Claude Code, Gemini CLI, GitHub Copilot, Cursor, OpenCode and others) or its own agent on your own API key.