Darkroom
Darkroom shows the pictures and videos in your Workspace as a shelf of thumbnails, and makes a copy of a video for sharing: smaller, in a format every browser plays. It plays the copy beside the original, so you can see what the smaller file costs before you keep it.
- Choose what to show in the list at the left: All Media, Pictures and Videos, then under Folders each folder that holds pictures or videos, indented under its parent, with how many it holds. The newest come first. A folder with folders inside has Hide at its right: it folds the folders inside, and Show brings them back. Its icon opens while they show, and a click on it does the same. A folded folder adds what it holds below to its count, in grey (
7 +12). With the list chosen, Right shows a folder's folders and Left hides them. What you fold stays folded in that browser. Drag the line between the list and the shelf to make the list wider or narrower. On a phone the list is Darkroom's first screen: tap a place to open it, and Back to return.
- Open a picture with a double-click (a tap on a phone). It opens in a picture window, and Left and Right step through the pictures around it. A HEIC or a TIFF opens as a JPEG made for the browser.
- Open a video the same way. It plays in Darkroom's own view, under the QuickTime controller, with what the file is: its size in pixels, frame rate, codec, length, bit rate and sound. A red line says the file records where it was made.
- Tags: each picture carries up to three, shown as small labels under its name, each tag in one of the Finder's seven label colours. TAGS in the list shows every tag with how many pictures carry it, and untagged the pictures with none; choose one to see its pictures. On the chosen picture, × removes a tag and + adds one, offering the tags you already use as you type.
- Tagging by a model: in the TAGS section, Turn On… has Kimi K3.1 Pro write the tags. It says first what this does: exe sends each picture, at most 1024 pixels wide, to the model through exe's Ollama (on ollama.com, for a cloud model), new pictures as they arrive. A dotted label is the model's and a solid one yours; once you change a picture's tags the model leaves them alone. The line under the tags says how many are left, or why tagging stopped (it tries again after a while, or at once with Try Again). Only the node where you turn it on asks the model; the tags reach your other nodes with Node Sync and follow a picture when it moves, because each is kept by the picture's contents.
- Upload… puts pictures and videos in the Workspace: in the folder on show, or else in Pictures and Videos. You can also drop files on the shelf, or on a folder in the list to put them in that folder. Download saves the original to your computer.
- Compress… makes a trial copy of the chosen video. Use picks what it is for: - Web: H.264 and AAC in an MP4, at most 1080p. It plays in every browser and app. - Fit to Size: H.264 that stays under 10, 25 or 100 MB, for chat apps and mail. A long video gets a smaller picture so it fits. - Smaller: HEVC, about half Web's size at the same look. Safari and phones play it; other browsers need a graphics card that decodes it. - Smallest: AV1, the smallest at the same look. New browsers play it; older Safari and phones may not. - Loop: an animation that plays by itself, animated WebP or GIF, ten seconds at most and 640 pixels wide, with no sound.
Picture sets the largest size, Trim the part to keep (as m:ss.s), and Encoder whether the graphics card or the CPU does the work. Include the sound and Remove the location, camera and dates are on until you turn them off.
- Trials under the video lists the copies, each with its progress, its size, how much smaller it is, its SSIM (how close it looks to the original, where 1 is the same) and the encoder that made it. A trial is kept out of the Workspace, so it does not go to your other nodes. Keep saves it beside the original, as
kite (Web).mp4; Discard throws it away, or stops one at work. A trial nobody keeps goes after a week.
- Compare plays the trial and the original together, from one controller. Side by Side shows them in two halves. Wipe shows one picture with a divider you drag: the original on its left, the trial on its right. Actual Size shows both at the video's own pixels; drag to move round the picture. Space plays and pauses, and Left and Right step a frame.
When a browser cannot play the original (an iPhone's HEVC video in Chrome on Linux, or ProRes), Darkroom makes a copy the browser plays and shows that one, and says so.
The work is done by exe's own ffmpeg. The daemon downloads it from exe-assets.v2core.com the first time Darkroom needs it (about 30 MB), and checks it against the checksum in exe. It encodes on the computer's graphics card where one runs: NVIDIA's NVENC on Linux and Windows, Apple's VideoToolbox on a Mac, Intel Quick Sync, AMD AMF or Media Foundation on Windows, VA-API or Vulkan on Linux. Otherwise the CPU does the work. The status line says which encoder is in use. A trial runs at low priority, one at a time; on Linux it also runs under a memory limit of an eighth of the machine's memory (8 GB at most). A trial that takes longer than half a minute sends a notification when it is ready, if notifications are on. Thumbnails, trials and previews are kept in ~/.exe/darkroom, which Node Sync does not copy.