A guide to Deepmosaic Desktop (Windows). For the browser version, see the Web app guide
Screenshots show the English version of the app. Text typed into the sample project stays in Japanese. Tutorial videos have Japanese subtitles only. Button and menu names in this guide match the English version.
Deepmosaic is a Windows desktop application that automates adding mosaic to videos with AI (deep learning) object detection. Everything from loading a video to AI detection, editing the detection results, and exporting a video with the mosaic burned in happens inside the app. This software is for applying (adding) mosaic. It has no feature for taking off mosaic from footage that already has mosaic applied, or for restoring the original footage.
This page is the guide to the Windows app (Deepmosaic Desktop). For the Web app, which runs in your browser with no installation, see the Web app guide (the differences from the Desktop app are also summarized at the end of this page).
Get it from Microsoft Store. The button below opens the Microsoft Store listing page; select "Get" (or "Install"). You can also find it by searching for "Deepmosaic" in the Microsoft Store app (to open it directly in the Microsoft Store app, click here). Administrator rights aren't required.
project.db in the project folder) when the app starts. During the conversion, the home screen shows "Converting detection results to the new save format (n / N)"Windows 10 (version 2004 or later) / 11 (64-bit). Try it right after installing, with no sign-in required
The installer edition downloaded from our website (shown as the "installer (exe) version" in the app) will stop working after December 31, 2026. You can keep using it until then. Please switch to the Microsoft Store edition soon. Pricing, terms of use, and features stay the same.
If you uninstall an installer edition older than 2.18, do not select the "アプリのデータとプロジェクトフォルダを削除する" (Delete app data and project folders. This item appears in Japanese) check box on the confirmation screen (doing so also deletes the settings, detection results, and project folders shared with the Microsoft Store edition). The AI models and other data are deleted, and the Microsoft Store edition downloads them again the next time it starts. If you first update to 2.18 or later with the in-app update, they aren't deleted.
The available features, accounts, and plans in the Microsoft Store edition are the same as in the installer edition (including subscribing to a plan, changing it, and canceling your subscription from the app). Only the following points are different.
%USERPROFILE%\.deepmosaic) and your project folders are left on your PC. After uninstalling, delete any folders you no longer need in File Explorer (note that deleting a project folder also deletes any exported videos saved in it)On first launch, the "Before you start" dialog appears before the initial setup. Review the End User License Agreement and the Terms of Service (you can read the full text from the links in the dialog), select the "I agree to the End User License Agreement and the Terms of Service" check box, and press "Agree and continue" to start the setup. If you choose "Decline (quit)", the app quits. When the End User License Agreement or the Terms of Service is revised, "The End User License Agreement and Terms of Service have been updated" appears when the app starts, so agree to them again.
The detection engine and the lightweight detection model are included in the app. On first launch, the model data needed for AI detection is downloaded automatically (About 10 GB for AI models and other data during initial setup). During the download, the overall progress (%) and the time remaining are shown, and you can't do anything else until it finishes. Depending on your connection speed, it can take anywhere from a few minutes to more than ten minutes.
Some data, such as the high-accuracy model, is fetched in the background after you reach the home screen. Until that finishes, detection uses the lightweight model, and the detection options screen shows "Preparing the high-accuracy model (about N MB left) — the lightweight model will be used for detection this time."
When the download finishes, a test run of detection and export ("Test run in progress…") starts automatically. If a problem is found, only the affected data is downloaded again automatically. When it is done, the app moves to the home screen on its own and shows "Installation complete" (you don't need to press "Continue").
Until the download and the test run finish, all menu items except "File ▸ Quit" and "Help" (other than "Tutorials") are disabled, and closing the window or reloading the screen is also blocked. If you quit partway through, the app fetches only the missing data and resumes the next time it starts.
On PCs with an NVIDIA GPU, the acceleration components (from NVIDIA; about 1.7 GB to download, about 2.4 GB after extraction) are fetched automatically the first time. They are not fetched on PCs without an NVIDIA GPU. Until they are fetched, the app runs on the CPU (the features are the same, but processing is slower). While they are being fetched, the progress appears in "GPU acceleration" at the bottom right of the screen, and once they are fetched, later detections run on the GPU.
If fetching fails, the indicator at the bottom right shows 'Failed to get the files. Press "Retry", or they will be fetched again the next time the app starts'. Check your connection and press "Retry", or restart the app (the files are fetched again automatically the next time it starts). For how these are handled as third-party software, see Bundled third-party software and licenses.
From top to bottom, the window consists of the menu bar, the screens (Home / Detection options / Detection / Edit), and the footer.
Lets you check and manage the waiting list of detection and export jobs in one place. A badge shows the number of running and waiting jobs (Queue).
"Licenses" and "Documentation" in "Help" are available even during the initial setup. In versions before 2.13 and in 2.14, these two items were also in the profile menu.
Just below the "Theme" row in the profile menu (top right), there are 3 buttons: Japanese, English, and Traditional Chinese, each written in its own language. When you press one, the screen language switches immediately, and the app also uses the selected language the next time it starts.
Click your avatar (or user name) to open it. It shows your sign-in status and plan, and has items to sign in, open your organization (when you belong to an organization on a business plan), manage your plan subscription (the plan selection screen opens on the Free plan, and the payment portal opens on a paid plan), contact us, sign out, switch the theme, and switch the display language. "Licenses" and "Documentation" are in "Help" on the menu bar.
Use "Theme" in the profile menu to switch between "Use OS setting", "Light mode", and "Dark mode". By default, the app follows the OS setting. Your choice is saved, and the app uses the same theme the next time it starts. The video preview and the detection box colors stay the same regardless of the theme.
You can use the app for free for 7 days from the day you first launch it. Right after downloading, you can try it for up to 30 minutes in total without signing in. Creating a free account raises this to up to 2 hours in total. On a single PC, the free allowance is capped at 2 hours in total, even if you sign up again with a different account. The free period ends after 7 days or when you reach the total limit, whichever comes first. No credit card is required. The individual plan (Hobby) is paid by credit card (Enterprise is paid by bank transfer against an invoice; see Organization management).
Signing in increases your available detection time even on the Free plan, and on PCs with an NVIDIA GPU, it lets you detect with the high-accuracy model (when you're not signed in, detection uses the lightweight model).
There are no overage charges. If you use up the included hours, your allowance is restored on the next renewal date. If you need more than that, please ask us about Enterprise, our plan for businesses (custom quote).
Starting with 2.13, the sign-in screen has these two entry points. You enter your email address on the authentication screen in your browser. If you have been signing in with a password, you can still sign in to the same account as before by selecting "Sign in" and entering a confirmation code with the same email address (your plan and usage time carry over unchanged).
Your sign-in information is encrypted and stored on your PC, and you're signed in automatically from then on. To sign out, use "Sign out" in the profile menu. To switch to a different account, sign out and then sign in again.
You can run the app on only one device at a time with the same account. If you sign in on another PC, "Your account is currently signed in on another device" appears. If you select "Continue", the previous device is signed out the next time the app starts on it. When you replace your PC, just sign in on the new one to switch over (to move your projects, see Exporting and importing projects).
Your sign-in and plan are checked over the internet when the app starts. Detection, editing, and export work without a connection. If the app can't connect to the authentication server at startup, a banner such as "Can't connect to the network (you can keep using local features)." appears at the top of the screen, but you can keep using the app with your most recent sign-in status and plan. If the banner says "Sign in again.", sign in again from the profile menu.
You can subscribe to a paid plan from the profile menu in the Windows app, or on the Subscribe to a paid plan page with your Web app account. Either way, the subscription applies to both the Desktop app and the Web app for the same account.
To subscribe or change your plan in the app, use the profile menu at the top right of the screen.
To subscribe in the Web app, choose a plan on the Subscribe to a paid plan page, press "Subscribe", and sign in with the same account as the app to go to the credit card payment screen. You can also change your payment method, change your plan, or cancel your subscription from "Plan subscription and management" in the profile menu at the top right of the Web app screen.
You can cancel your subscription at any time, with no cancellation fee or other penalty. After you cancel, you can keep using the app within the free allowance.
Pro is no longer accepting new subscriptions. If you're currently subscribed, you can keep using it. To change your plan or cancel your subscription, follow the paid plan steps above.
The Hobby card shows the monthly price (excluding and including tax), the included hours, and the one-device-at-a-time limit. Each account can have only one subscription to the same plan. Press "Choose this plan" to open the payment screen in your browser; when payment is complete, "Payment complete" appears in the app. For Enterprise, contact us from "Ask about deployment" (pricing is by quote).
When you're signed in, you can delete your account from the icon shown on the right side of the footer on the home screen. This action can't be undone.
To report a problem or ask a question, use "Contact us" in the profile menu (sign-in required). The email address you're signed in with is filled in automatically. Enter your message and press "Send", and "We've received your inquiry. It may take a little while for us to reply." appears. Replies are sent to your registered email address.
When you report a problem, include the version shown in the footer, what you were doing, and the message that appeared. This helps us investigate faster. If you can't sign in, use the contact form.
There are 3 ways to load videos. Whichever you use, the detection options screen opens right after loading.
If you haven't loaded any videos other than the sample yet, an "Add a video to get started" box appears below the list. Drag and drop a video file, or choose one with "Select video" (the same as "Select a video file" in the menu).
Supported formats: mp4 / avi / mkv / mov / wmv / flv / webm / ts / m2ts / m4v / mpg / mpeg (file extensions aren't case-sensitive). When loading from a folder, files that can't be read as video are excluded even if their extension matches, and you are notified of how many were excluded.
When you load several videos at once, the first one appears on the detection options screen. When you press "Start", those settings are applied to all the rest, and the videos are processed in order. However, the detection range applies only to the first video; from the second video on, the whole video is processed. If you load videos while detection is already running, they are also added to the waiting list and start automatically as soon as the current processing finishes (Queue).
Each video you load is managed as a project, and its detection results, edit data, and thumbnails are saved in a project folder for that video. By default, project folders are placed in C:\Users\<user name>\deepmosaic. The folder is created automatically if it doesn't exist. You can change this to another location (such as an external drive) with "File ▸ Select a project folder" in the menu, and videos loaded after the change are saved in the new location.
"All videos" on the home screen lists the projects in the current project folder that were created with the account you're signed in with. If there are projects saved in another project folder, an "Other folders" row with a count appears in the left sidebar (it doesn't appear when the count is 0). Select it to see those projects in the same list.
Videos in "Other folders" belong to a different project folder, so folders and tags aren't available for them (you also can't drag them to folders in the sidebar). To organize them, select "Switch to this project folder" from the video card menu (it also opens with a right-click). The folder where that video is saved then becomes the current project folder.
On the Free plan (including when you're not signed in), if there are no videos on the home screen, a sample video is added automatically and appears as a card with a "Sample" badge ("Sample video added" appears). The sample contains face detection results only, so you can try editing and exporting it right away. The edit page shows "This sample uses face detection. Try body area detection with your own videos." Detecting the sample again doesn't use up your available detection time. Once you delete it, it isn't added again. It isn't added on paid plans.
On the Free plan, a "Getting started (3 min)" card appears at the top of "All videos" (in the "Large icons" view, it takes the first cell of the list, and the sample and other videos follow to its right). Click it to play the video in the app (press Esc to close it). Select "Don't show" on the card to hide it. After that, you can play the video from "Help" ▸ "Tutorials" on the menu bar. The card doesn't appear on paid plans.
Choose from 3 display formats at the top right of the list. The format you choose is kept for next time.
Click a card to open the edit page. The menu at the top right of the card has the following items.
To delete several projects at once, press "Select mode" at the top of the list, click cards to select them (you can also use "Select all"), and press "Delete selected". After a confirmation dialog, the selected projects are deleted along with their folders.
The following notices appear above the list on the home screen, only when needed.
project.db). If conversion fails, "N projects failed to convert" appears. Press "Retry"When you load a video, the detection options screen opens. Here you decide which part of the video to detect and what to detect, then press "Start".
Detection time is roughly proportional to the number of frames processed. If you want to process only part of a long video, narrowing the detection range shortens the wait. Note that your available detection time (allowance) is counted based on the full length of the video even if you narrow the range. Detecting the same video again isn't counted twice.
Detection is also stopped before it starts if your remaining available detection time is shorter than the full length of the video. A message such as "Your remaining detection time (1 h 30 min) is not enough to detect a 120-minute video. Check your plan." appears above the "Start" button, and you can't press "Start". Because the full length of the video is counted even if you narrow the detection range, specifying a range doesn't let you start. You can check your remaining time in the footer (How usage time is measured).
If your license has expired, you've exceeded the usage limit, or your free period has ended, a dialog prompting you to change your plan appears on this screen, and you can't start detection (see Changing your plan).
When the detection options screen opens (and also when you choose a different save location on this screen), the app checks whether it can write to the save location for detection results (the project folder). If it can't, you can't press "Start", and the reason appears: "Can't start because the save location (project folder) can't be written to". Use the buttons shown with it, "Choose the project folder again", "Use the default location" (shown when the current save location isn't the default location), and "Check again", to change the save location, or fix the cause and check again. Where the video itself is stored doesn't matter. If you change the save location with "Select a project folder" in the menu, the check runs the next time you open the detection options screen (Project folder).
When you turn on "Import without detection" with the switch at the top of the right panel, detection isn't run and the video is imported into the edit page as it is. You add the boxes yourself on the edit page (Adding detection boxes). The "Start" button changes to "Import", and "Importing" is shown while it runs and "Import complete" when it finishes. It doesn't use detection time (you can import videos longer than your remaining detection time). However, as with detection, you can't start it when your license has expired, you've exceeded the usage limit, or your free period has ended. While it's on, the detection targets, detection range, and batch run after detection can't be used, and automatic scene detection isn't done (the edit page starts with a single scene). The initial mosaic settings (pixel size) are used for the boxes you add. This switch isn't saved and goes back to off each time you open the screen.
This feature finds the position and extent of block mosaic already in your footage and, on export, applies Deepmosaic mosaic over it again. Use it when the mosaic in your footage varies in block size and coverage, and you want to redo it to your own standard. This feature doesn't take existing mosaic off. It can't restore the original footage.
Automatic detection isn't 100% accurate. Before exporting, review the entire video on the edit page (Checking for missed detections).
Both vary with your PC's performance and the footage. The extra time is also included in "Estimated time" below the detection range.
If you turn on "Export right after detection as a batch" in detection options and press "Start", the app automatically continues after detection finishes and exports a video with the mosaic burned in, without opening the edit page. When you load several videos at once, this applies to all of them.
A batch run skips the step of reviewing the results on the edit page, so always review the entire output video yourself before publishing or distributing it. For the default mosaic settings, the batch run uses what you applied with "Set pixel size" and what you saved with "Make these settings the default" in the mosaic settings panel.
When you press "Start" in detection options, AI automatically detects the target objects in the video. Processing goes through 5 stages, "Prepare", "Preprocess", "Detect", "Analyze", and "Refine mosaic", in that order (4 stages when mosaic refinement isn't used). When "Mosaic" is turned on in detection options, a "Detect existing mosaic" stage, which looks for mosaic already in the video, is added (Finding existing mosaic and reapplying it). Overall progress, processing speed (fps), and time remaining are shown. All the stages that will be used are shown from the moment detection starts. You can pause, resume, and cancel processing. When you load several videos, you can check the waiting list in "Queue" in the menu bar.
If detection stops partway, for example because the app was closed or the PC shut down, the next time the app starts, it appears under "Interrupted detections" above "All videos" on the home screen, and an "Interrupted detections" row with a count also appears in the left sidebar. The card shows the stage where it stopped and how far it got, such as "Interrupted (DETECT 42%)". Press "Resume" to continue detection from near where it stopped, not from the beginning. Press "Discard" and confirm to delete the detection's intermediate progress and proxy video (the original video isn't deleted). If it can't be resumed, for example because the original video is missing or has changed, the project has been edited, or it was created by a newer version of the app, the card shows the reason and you can't press "Resume" (if "Update the engine to resume." appears, apply the update).
During processing, the progress and estimated time for each step are shown.
For a target that is cut off at the edge of the frame, detection extends the detection box beyond the edge of the frame and records it. On export, mosaic is drawn only inside the frame, so the mosaic shape doesn't get squashed at the edge of the frame, even for targets moving out of frame (Player and timeline).
In the Desktop app, detection automatically smooths the motion (trajectory) of the detected boxes. Small frame-to-frame jitter and boxes that jump for just 1 frame are suppressed, so the mosaic flickers less. No settings are needed. Box edges that touch the edge of the frame are kept as they are, so this doesn't affect the extension of boxes beyond the edge of the frame described above. This smoothing applies to videos detected with a version that supports it (videos detected before that don't change).
There are 3 detection targets, "Body areas", "Male faces", and "Female faces", plus "Mosaic" (mosaic already in your footage). You can turn each one ON / OFF in detection options.
Detection and export are managed in a single waiting list (the queue), which you can view and operate from "Queue" in the menu bar. The badge on the button shows the total number of running and queued jobs. Detection processes one video at a time, as does export, and when a job finishes, the next one starts automatically. Jobs for Import without detection have an "Import" badge.
The queue remains even if you close the window. When you quit the app, running detection is interrupted, and you can resume it from "Interrupted detections" on the home screen the next time the app starts (AI detection). Exports left in the queue can't be resumed the next time the app starts (start them again from the edit page).
Click a video card on the home screen to open the edit page. The screen consists of the header at the top, the sidebar running vertically along the edge (and the drawer that opens next to it), the video player (with the built-in edit toolbar) in the center, and the timeline at the bottom.
The player overlays detection boxes (BBoxes) in a different color for each class. Drag a box to edit its position and size directly. The player supports stepping frame by frame and seeking. The timeline at the bottom shows the distribution of detections and the scene (cut) strip, and you can click it to jump to any frame.
Detection classes are color-coded: body areas (red) / male face (blue) / female face (pink) / user-created (yellow) / exclusion mask (teal) / mosaic (purple: mosaic already in your footage, only in projects detected with "Mosaic" turned on in the detection options). The label (chip) above each box shows "class name #ID (3-character code) confidence%". Boxes you added manually are labeled "User", exclusion masks are labeled "Exclude", and tracks excluded from mosaic get "[mosaic excluded]". To hide the chips, turn off "Show BBox labels (name, ID, probability)" in "Settings" in the sidebar (the boxes stay visible. Sidebar and drawers).
Mosaic (purple) boxes are polygons (up to 255 points), and as with other polygon boxes, you can move their points to fix the shape ("Range-selecting and bending points" below, Finding existing mosaic and reapplying it).
Selecting, adding, and erasing detection boxes work only when the playhead is in a segment of the original video (the detection source). Segments of imported media aren't subject to AI detection, so they have no boxes.
From top to bottom, the timeline shows the toolbar, the ruler, the scene (cut) strip, the text, clip, and audio lanes (Editing clips), and a row for each detection class. Each class row shows bands for the segments that have detections. Manually added boxes appear in the "User" row, and exclusion masks in the "Exclude" row. Use at the start of a row to expand it into a row for each tracking ID. Right-click an ID row to open the list of IDs that can be merged into that ID ("Merging IDs" above. For merged IDs, "Unmerge" appears at the end of the list). Right-clicking doesn't move the playhead to that position. Tracks excluded from mosaic are shown faded. The timeline is laid out in the time of the exported video, so when you trim or move a clip, the detection bands move with it.
Zoom the timeline up to 16x with Ctrl+wheel or the zoom buttons at the left end of the toolbar. For changing the height and collapsing the timeline, see Editing clips.
To fit boxes precisely around small subjects, you can zoom the player's video up to 8x. While zoomed in, you can still move, resize, add, and delete boxes as usual.
Where nothing was detected, you can add boxes manually with the "1f" (a square with a + and 1f) and "∞" (a square with ∞) buttons in the toolbar. An added box is placed in the center of the current frame as the "User" class (yellow), so drag it to adjust its position and size. Mosaic settings (including per-ID settings) apply to manual boxes as well.
To keep mosaic off just part of the area it covers (for example, to show only the hands next to a face), place an exclusion mask. No mosaic is applied inside an exclusion mask, and that area is shown and exported as the original video. Exclusion masks don't follow the subject, so for a moving subject, drag the mask into position.
You can also change detected boxes and manually added boxes into exclusion masks later. Inside a box changed into an exclusion mask, other overlapping mosaics aren't applied either.
Adding, moving, and deleting boxes are all saved automatically and reflected in the export. Only the detection results and the edge feather value are passed to Premiere Pro integration. Box edits on this screen (moving, changing shapes, deleting, exclusion masks, and so on) aren't passed.
"Tracking" automatically fits an ∞ continuous BBox you placed by hand to the subject's movement as the video plays. Turn it on or off with the button in the control row. "Keyframes" use the frames you corrected by hand as reference points for the movement, and automatically connect the frames between one keyframe and the next.
Use boxes in separate frames as points to create one ∞ continuous BBox that moves between them following the subject's movement. Besides the start and the end, you can place any number of points along the way (waypoints).
Tracking for connected sections goes into the "Tracking queue" and runs one job at a time, in order. On PCs with a GPU, contour refinement (like mosaic refinement, it fits the box to the shape of the object) also runs after tracking. When a job finishes, the interpolated boxes between the points are replaced with the result of tracking (and refinement). Boxes on the frames of the points you placed by hand don't change.
When you correct an AI box by hand, only that frame changes. To fix a range of several seconds to several minutes at once, use the following actions depending on the situation.
Keep the following in mind.
During detection, cuts in the video (changes of scene) are detected automatically and shown as scene strips at the top of the timeline. A scene is the unit within which the block size is kept consistent when "Automatic pixel size" is on, and also the unit for changing Mosaic settings per scene.
Scene changes are saved as edit data, and the detection results themselves don't change. "Reset all edits" returns the scenes to their automatically detected state. Splitting a scene doesn't cut the exported video. To cut the video itself into pieces, split a clip (Ctrl+Shift+K, Editing clips).
The control row below the player contains, from the left, the playback controls (Go to start / Back 1 frame / Play / Forward 1 frame / Go to end / Play blank frames only / Missing frame jump / Mute), the edit toolbar, and the time. The edit toolbar is divided by vertical lines into these groups: Shape picker / Add BBox / Add exclusion mask / Eraser / Reset current frame | Toggle overlay / Switch preview / Tracking | Mosaic settings | Reset all edits | Labels | Playback speed | Keyboard shortcuts. The Tracking button shows only an icon. When there isn't enough width, the toolbar and the time wrap onto a second row. Hover over each button to see its shortcut key.
You can set how the mosaic looks in 5 scopes (tabs). When several settings apply to the same box in the same frame, the setting changed last takes priority ("Setting priority" below). Drag the panel's header to move it anywhere, and drag its bottom-right corner to resize it (you can also select the corner and adjust it with the arrow keys, or double-click it to restore the default size). You can keep editing with the panel open. When you change a value, the change appears immediately in the mosaic preview (M).
Reset each item to its default value with the button at its right end. Press "Make these settings the default" on the "All" tab to make those settings the initial settings for the next video you detect (the settings of this video don't change).
On the "By time" tab, you can change how the mosaic looks for each time range of the video.
When settings from "All", "By class", "By ID", "By scene", and "By time" overlap on the same box in the same frame, the setting changed last takes priority, and all items of that setting are used as they are (items aren't mixed).
Open "Person gallery" (P) in the header to see the detected faces grouped by person (detections with low face likelihood are not included in any person and go to "Others"). You can turn mosaic on or off and correct the gender for each person, and if the same person was detected as separate people, you can merge them into one. You can drag the panel to move it.
For projects detected with an older version, "This video was detected with an older version. To use the people list, run detection again." appears. In that case, load the video again and run detection.
This is a list of the shortcuts you can use on the edit page. You can also hover over at the right end of the toolbar to show the same list. Shortcuts don't work while you're typing text.
| Category | Key | Action |
|---|---|---|
| Playback | Space | Play / pause |
| J / K / L | Stop / Stop / Play (compatible with Premiere Pro) | |
| V | Mute / unmute | |
| , | Decrease playback speed | |
| . | Increase playback speed | |
| B | Play blank frames only (skips frames that have a BBox) | |
| Frame navigation | → | Forward 1 frame (if you hold Shift while a BBox is selected, it moves forward 1 frame and carries over the shape of that box. While carry-over is locked, the shape carries over even without Shift) |
| ← | Back 1 frame | |
| Shift twice | Lock / unlock shape carry-over (press Shift twice quickly. While locked, → and the wheel also carry over the shape without Shift. Esc also unlocks) | |
| PageDown | Forward 10 frames | |
| PageUp | Back 10 frames | |
| Home | Go to start | |
| End | Go to end | |
| [ | Previous missing frame | |
| ] | Next missing frame | |
| Zoom and view | + / = | Zoom in on frame |
| - | Zoom out on frame | |
| 0 | Reset zoom | |
| H | Show / hide detection overlay | |
| M | Switch BBox / mosaic preview | |
| C | Open / close the "Labels" menu in the toolbar (labels to show and filter) | |
| BBox editing | 1 | Add a BBox to this frame only |
| 2 | Add a continuous BBox to the following frames | |
| 3 | Add an exclusion mask to this frame only | |
| 4 | Add a continuous exclusion mask to the following frames | |
| W | Set the shape of BBoxes you add / draw to ellipse (press again to switch back to rectangle) | |
| O | Set the shape of BBoxes you add / draw to polygon (press again to switch back to rectangle) | |
| Delete | Delete the selected BBox (when you last worked in the player) | |
| Ctrl+C | Copy the selected BBox (when you last worked in the player) | |
| Ctrl+V | Paste the copied BBox into the current frame (when you last worked in the player) | |
| Y | Set the selected BBox as the connection source / connect. Select a BBox in another frame and press it again to create a continuous BBox that connects the points by following the subject's movement (when you last worked in the player. "Connect from here" and "Connect to here" in the right-click menu do the same. A single Ctrl+Z undoes it) | |
| Shift+Y | Add the selected BBox as a waypoint (while a connection source is set. When you last worked in the player. "Add as waypoint" in the right-click menu does the same) | |
| Esc | Clear the connection source (stop without connecting after placing points) | |
| Ctrl+Z | Unfreeze continuous BBox → reset current frame | |
| Q | Rotate selected BBox by 15° | |
| Shift+Q | Rotate selected BBox by -15° (counterclockwise) | |
| F | Make the current position of the selected BBox a keyframe (Tracking and keyframes) | |
| X | Turn tracking mode on / off (while it's on, the ∞ continuous BBox follows the target during playback, with →, and when you scroll the wheel up. See Tracking) | |
| Tools | G | Open mosaic settings panel |
| Ctrl+K | Split at the playhead (splits the clip and the scene at the same time. Same as the scissors in the toolbar; a single Ctrl+Shift+Z undoes both) | |
| Timeline | Ctrl+Shift+K | Split at the playhead (when the timeline has focus. Same as Ctrl+K) |
| Backspace | Delete the selected clip / text / audio track | |
| Ctrl+Shift+Z | Undo timeline edit | |
| Ctrl+Shift+Y | Redo timeline edit | |
| Ctrl+← | Move the selected element 1 frame earlier | |
| Ctrl+→ | Move the selected element 1 frame later | |
| Ctrl+D | Duplicate selected element | |
| Ctrl+C | Copy the selected element (when you last worked on the timeline) | |
| Ctrl+V | Paste the copied element (text and audio at the playhead, clips after the selected clip. When you last worked on the timeline) | |
| Delete | Delete the selected element (when you last worked on the timeline) | |
| Page | P | Open / close person gallery |
| A | Open / close media (sidebar) | |
| T | Open / close the text / watermark / audio / fade / transition drawer (the item you opened last) | |
| Ctrl+Shift+T | Collapse / expand timeline | |
| Ctrl+Shift+B | Minimize / expand sidebar | |
| Ctrl+Enter | Open export dialog | |
| Ctrl+Shift+O | Open project folder | |
| Ctrl+Shift+P | Import into Premiere Pro | |
| Mouse | Wheel | Step forward / back a frame (10 frames with Shift. While a box is selected, scrolling up with Shift held moves forward 1 frame and carries over the shape, like Shift+→, and scrolling down moves back 1 frame. While carry-over is locked, scrolling up carries over the shape even without Shift) |
| Ctrl+wheel | Zoom the frame in or out around the cursor position | |
| Ctrl+left-drag | Pan the frame (works only while zoomed in. Dragging with the middle button also works) | |
| Left-drag | Left-drag on an empty area to draw a BBox in the shape chosen in the shape picker (rectangle / ellipse / polygon). For polygons, click to place points, and close the shape with the start point / Enter / double-click. A short tap clears the selection | |
| Double-click | Delete the BBox | |
| Drag | Drag the rotation handle of the selected BBox to rotate it (in 15-degree steps with Shift) | |
| Drag | Drag a polygon point to reshape it | |
| Double-click | Double-clicking a polygon edge also adds a point (1 point per double-click) | |
| Click+Delete | Click a point to select it, then press Delete to delete it (a polygon can't have fewer than 3 points) | |
| Drag | Drag a handle of the selected polygon to reshape the curve (Alt moves one side only) | |
| Alt+drag | Alt+drag a polygon point to pull out handles (turns a corner into a curve) | |
| Alt+click | Alt+click a polygon point to turn it back into a corner | |
| Ctrl+click | Ctrl+click a polygon point to delete it (the polygon won't go below 3 points) | |
| Click | Click an edge of the selected polygon to add a point | |
| Shift+left-drag | Drag around points of the selected polygon to select them together (they're grouped into one connected segment. If it's 1 point, that single point is selected; tap to clear) | |
| Drag | Drag inside the points you selected together to bend the segment while both of its ends stay fixed (points nearer the middle move more) | |
| Right-click | Track menu for the detection box (switch gender / exclude from mosaic / invert / exclusion mask ("Make exclusion mask" / "Change back to mosaic") / merge IDs / connect ("Connect from here" / "Add as waypoint" / "Connect to here") / change shape / delete). On an ID row of the timeline, the list for merging IDs. On a bar of the timeline, a menu to duplicate, copy, paste, and delete | |
| E+right-drag | Delete continuously (eraser). With the eraser button in the toolbar, you can choose "Delete touched BBoxes / Delete only touched parts", "Circle / Rounded rectangle", and "Size". It keeps working if you play or step frames while holding it down |
The "Clip lane" of the timeline shows the sequence of the exported video as bars (clips). At first there is a single clip of the original video (the detection source). You can trim, split, and delete to cut unwanted parts, or add imported media before or after it. The text lane is above the clip lane and the audio lane is below it, and all lanes are laid out on the exported video's time.
Your edits show up in the preview as is and are included when you export. Mosaic is applied to the sections from the original video.
Right-click a clip, text, or audio track bar to show a menu with "Duplicate", "Copy", "Paste", and "Delete". If you right-click an empty area of a lane, only "Paste" appears. The menu items and the shortcuts do the same thing.
Timeline edits (clips, text, audio, watermark, fade, transition, and adding media) are recorded in a history separate from detection box edits. Use "Undo" (Ctrl+Shift+Z) / "Redo" (Ctrl+Shift+Y) to the right of the timeline zoom buttons to go one step at a time, or choose an entry in the list from the history button to undo everything back to that operation at once. The list shows operation names such as "Split", "Trim", "Move", and "Add text" with their times, newest first, and keeps up to 100 entries. The history is cleared when you reopen the video.
The height and the collapsed state carry over the next time you open the edit page.
In the media panel, which you open with "Media" (A) in the sidebar, import the video, audio, and image media you want to use on the timeline. Imported media is copied into the project folder, so you can keep using it even if you move the original file (it's also included in the zip when you export the project).
Imported media isn't covered by AI detection. Mosaic isn't applied to media sections, and you can't add detection boxes or use the eraser there. For footage that needs mosaic, first load it as a video, run detection, export it, and then import the result.
Use "Text" in the sidebar to burn text into the video. You can add as many text items as you like and set the display range, position, and format of each. The display range is specified with frame numbers of the exported video.
Above the clip lane of the timeline, a bar showing the display range appears for each text item, one row per item (text lane). Drag a bar to move it together with its display range, or drag either end to change the start or end. Selecting a bar also selects the same text in the text panel. Right-click to duplicate, copy, paste, or delete.
Use "Watermark" in the sidebar to overlay one image, such as a logo, on the entire exported video. Import the image in the media panel first (you can also set it with "Set as watermark" on an image in the media panel).
To add a text watermark, use text shown over the entire video.
Use "Audio" in the sidebar to adjust the audio of the original video and audio tracks such as background music. On export, the audio of the original video and all audio tracks are layered (mixed) into the output.
Below the clip lane of the timeline, a bar appears for each audio track, one row per track (audio lane). The bar shows the waveform of the background music, so you can line things up while seeing where the music builds up or goes silent. Drag a bar to move it, or drag either end to trim the part that is used (the start and end of the song). Background music keeps playing across gaps between clips. For audio whose length is unknown, "length unknown" is shown instead of a bar.
Fade the exported video in from black and out to black. There are two kinds: "Overall fade", applied to the start and end of the whole video, and "Clip fade", applied to the start and end of each clip. For the overall fade, enter the seconds in "Fade" in the sidebar. For clip fades, enter them in the section for the selected join, which opens when you choose "Fade" from the scissors at a join, or in the "Selected clip" section, which appears when you select a clip. You enter the seconds only in the sidebar (not on the timeline).
Fades appear in the preview just as they do on export. Frames where the overall fade and a clip fade overlap get both. Set fades for each audio track in Audio and background music, and the switching at joins between clips in Transition.
Add a switching effect (transition) at the join between adjacent clips. Press the scissors at a join and choose "Transition" to open "Transition" in the sidebar (the drawer heading is also "Transition") with that join in focus, then choose the type and length. You enter the type and length only in the sidebar (not on the timeline). When there is only one clip, there is no join, so you can't add one. You can't add a clip fade to a clip end on the side of a join that has a transition.
Use the "Export" button in the edit page header (Ctrl+Enter) to set up the export. Press "Start" to add the job to the queue, and follow its progress in a dedicated window. Any unsaved edits are saved automatically when the export starts.
The mosaic block size is adjusted automatically so that it doesn't fall below the self-regulation standard (the long side divided by 100, minimum 4px) for the export resolution. Even when you export 4K media at Full HD, the output video meets this standard.
Areas covered by an exclusion mask get no mosaic and are exported as the original video, and their edges blend in with the edge feather. For a detection box that runs off the edge of the frame, mosaic is applied to the part inside the frame, based on the shape extended beyond the edge of the frame during detection (the shape isn't squashed at the edge).
"MOV (ProRes 422)" exports in ProRes 422. Its image quality (bitrate) is lower than "MOV (ProRes 422 HQ)", so the files are smaller (roughly 2/3 the size). In earlier versions, choosing "MOV (ProRes 422)" still exported in 422 HQ. If you need the same quality as before, choose "MOV (ProRes 422 HQ)".
"MOV (DV)" can export only 720×480 (29.97 fps) and 720×576 (25 fps). "MOV (DNxHR …)" and "MOV (CineForm)" can export only videos whose width and height are even numbers, and "MOV (CineForm)" also requires a width that is a multiple of 16 and a height of 32 or more (otherwise, the reason is shown when you start). All of these formats use a MOV container, and the files are dozens of times larger than H.264, so keep an eye on the free space at the save location.
To open these settings, turn on "Burn in timecode" in the export settings and press "Settings" next to it. Burned-in timecode helps reviewers refer to positions in the video when you export a review copy (for example, to share with others for checking).
When the setting is on, the timecode also appears with the same look in the edit page preview and in the progress window preview.
When you start an export, a progress window opens as a separate window for each job. The job continues even if you close it, and you can reopen it at any time with "Open progress in a separate window" in the queue.
You can package a project (detection results and edit data) into a single zip file and hand it over to another PC or another person. This lets you split the work, for example by running detection on your own PC and asking someone else to review and correct the results.
Choose "Export project" from the video card menu on the home screen, and specify where to save the file. The zip contains the source video (if it's still in the project folder), detection results, edit data, scenes, thumbnails, representative face images, and the media you added in timeline editing (video, audio, and images). If the source video can't be found, a message tells you so, and everything except the video is exported.
Choose "File ▸ Import project" from the menu and select the zip. The imported project is extracted into the current project folder and added to the home screen.
If a project with the same project ID is already on your PC, "Matching project found" appears, and you choose one of the following.
While the project is being imported, the progress (extracting, then registering) is shown. On the PC that imported the exported project, you can edit it, export it, and send it to Premiere Pro right away.
From the video card menu or the button in the edit page header (Ctrl+Shift+P), you can automatically import the detected mosaic information into an Adobe Premiere Pro project. Before sending, the app automatically checks Premiere Pro itself, the plug-in, and whether Premiere Pro is running, and if anything isn't ready, it guides you through the setup steps. This feature is available only in the Windows app.
If the Creative Cloud component can't be found, "The Creative Cloud component needed for automatic installation was not found" appears. Install or update the Creative Cloud desktop app, or install the plug-in manually from "Open plug-in folder".
When you send, the "Premiere Pro integration" widget appears at the bottom right of the screen and moves from "Processing in Premiere Pro…" to "Import into Premiere Pro complete". Premiere Pro then does the following to the sequence in the open project.
If any box in a project has "Invert (apply outside)" turned on in the mosaic settings, the project isn't sent (Premiere Pro can't reproduce inversion, so the outside would be output without being hidden). In that case, "Sending was stopped because N settings use Invert (apply outside). Premiere Pro can't reproduce inversion, so the outside would be exported without being covered. Turn off Invert, or export from this app." appears. Turn off inversion, or export with this app.
After the import, you can adjust the position and block size with normal Premiere Pro editing. If there's no response for a while after sending, "Timed out" appears. Check that the plug-in panel is open in Premiere Pro, then try again.
AI detection isn't perfect. Before you export, check and correct the results with the following tools. Responsibility for the final check and for how the content is expressed rests with the creator.
The cases AI struggles with (extreme backlighting, heavy motion blur, partial occlusion, compositions that are rare in the training data, and animation / 3DCG) are listed on the home page. If your media falls into any of these cases, review it more carefully.
In the Microsoft Store edition, new versions are delivered by Microsoft Store and installed automatically. AI models are downloaded and replaced automatically at startup when new ones are available. The following describes the installer edition (until December 31, 2026; see Switching).
The installer edition automatically checks for a new version when it starts. If an update is available, a dialog appears, and you choose either "Update now" or "Later".
Select "File ▸ Check for updates" in the menu to see "Current", "Latest", and "Status" for three items: the app, the engine, and models and assets. When "Updates are available" is shown, you can apply them with "Update now" (the app) or "Update models and assets". "Check again" checks once more. If the app can't connect to the network, "Check your network connection, then try again." appears.
Deepmosaic does not send your videos or their detection results to a server. Everything is saved in the following locations on your PC.
C:\Users\<user name>\deepmosaic. Each video has its own subfolder containing a database that holds the detection results and edit data (project.db), thumbnails, representative face images, and more. Loaded videos themselves are not copied; they are referenced from their original location. Media added in timeline editing (video, audio, and images) is copied to the assets subfolder and is also included when you export the project (Project folder)%USERPROFILE%\.deepmosaiccatalog.db), encrypted sign-in information, the device ID, and logs (logs\app.log; logs for each detection are in logs\runs). When you contact us, we may ask you to attach the logs%LOCALAPPDATA%\deepmosaicIf you put the project folder inside a cloud-synced folder (such as OneDrive), autosave during editing can conflict with syncing and result in "Save failed". We recommend a local folder.
project.db and catalog.db are databases managed by the app. While the app is running, do not open them with other tools or modify their contents (doing so can corrupt your data). Make backups or copies after you quit the app, and if there are files whose names start with project.db (-wal / -shm), copy them as well.
You can uninstall the Microsoft Store edition in the same way. There is no check box on the confirmation screen, and your settings, sign-in information, and project folders are left on your PC (Differences between the Microsoft Store edition and the installer edition).
When you uninstall the installer edition, the AI models and other data (%LOCALAPPDATA%\deepmosaic, about 10GB) are deleted automatically (if you reinstall, they are downloaded again during the initial setup; from version 2.18, the installer edition doesn't delete them on a PC where the Microsoft Store edition is installed).If you select the "アプリのデータとプロジェクトフォルダを削除する (中の書き出し動画も削除されます)" (Delete app data and project folders; exported videos inside are also deleted. This item appears in Japanese) check box on the uninstall confirmation screen, your settings and sign-in information (%USERPROFILE%\.deepmosaic) and the project folders that Deepmosaic created (detection results and edit data, including any exported videos saved inside those folders) are also deleted. If you leave it unselected, these remain. Original videos and exported videos outside the project folders are not deleted. Nothing is deleted when you update.
To delete the account itself, see Deleting your account.
Measured processing speeds by GPU, the list of supported formats, and the status of a macOS version are on the system requirements page. For the requirements of the browser version, see Supported browsers and requirements for the Web app.
If this doesn't solve the problem, contact us through Contact us in the app or the contact form. Attaching the version shown in the footer and %USERPROFILE%\.deepmosaic\logs\app.log (for detection problems, also the file updated around the time of that detection in %USERPROFILE%\.deepmosaic\logs\runs, where the log of each detection is kept) helps us investigate faster.
There is also a Web app that requires no installation and that you can use with the same account from a browser such as Chrome or Edge. Detection, editing, and export are all processed inside the browser, and your video is not sent to a server. The detection targets and the features of the edit page are the same as in the Desktop app, but there are some differences, such as these: export formats are limited to MP4 (H.264) / WebM (VP9 / AV1), it handles videos up to about 1.9GB and in practice only mp4 / mov, and it has no Premiere Pro integration, project import, resuming of interrupted detections, or regrouping of people.
How to use it, the supported browsers, and a list of differences from the Desktop app are covered in Web app guide. You can open the Web app at app.deepmosaic.co.jp.
The Desktop app includes the following third-party software. In the app, you can see the same list and the full text of each license from "Help ▸ Licenses" in the menu bar. The list shows only the software names; press a row to open its details, such as the version, license, provider, and where to get the source.
--enable-gpl --enable-version3, consisting of ffmpeg.exe / ffprobe.exe and the FFmpeg DLLs that they load at run time (avcodec-62.dll, avdevice-62.dll, avfilter-11.dll, avformat-62.dll, avutil-60.dll, swresample-6.dll, and swscale-9.dll). Libraries including libx264, libx265, and libxvid (all GPL v2 or later) are statically linked into these DLLs. The list of linked libraries and their licenses can be found in "Help ▸ Licenses" in the app (ffmpeg-NOTICE.txt).The following components are not included in the app. They are downloaded automatically from our distribution source the first time, and only on PCs with an NVIDIA GPU (about 1.7 GB to download, about 2.4 GB after extraction). They are not downloaded on PCs without an NVIDIA GPU. Until the download finishes, the app runs on the CPU, with the same features.
On the Enterprise plan, multiple accounts in your organization share one contract. Invite and remove members, check the organization code, and check usage for the whole organization from "Organization management" in the app. We send invoices by email (to change the number of accounts or cancel, contact us through the Enterprise inquiry form).
"Organization" in the profile menu appears only for accounts that have joined an Enterprise organization. In the profile menu (top right), "Organization" appears in the same place where "Plan subscription and management" appears on the Free plan and paid plans. For pricing and included hours, see the pricing page (minimum number of accounts: 1).
Select "Organization" to open the "Organization management" dialog, with a list of items on the left. Admins see 4 items: Permissions / Billing / Invoices / Usage, and members see 2: Permissions / Usage. The content of the selected item appears on the right. The top left shows "Joined as Admin" or "Joined as Member". If "Organization management isn't available yet" appears, wait a while and open it again.
Organization members have one of two permissions: "Admin" or "Member".
The member list shows each member's name (or email address), permission, and last sign-in date (if the last sign-in can't be retrieved, the date they joined is shown instead). Above the list, "N / M seats in use" and the number of accounts still available are shown. Change permissions and remove members from each row of the list.
Only admins can invite members and resend invitation emails. With the Member permission, "Permissions" doesn't show the invite section, and a message saying so appears instead. To invite someone new, ask your organization's admin.
Instead of an invitation email, you can also join with an "Organization code" shared across the organization. Use this when many people need to join, or when you want to distribute the code within your company and have each person register on their own. We send the organization code in the welcome email when your contract becomes active (see "How to sign up (quote and payment)" below).
ABCD-1234-EFGH, and can be copied with the button next to it). Only admins can see the organization codeThe only way to join is the "Organization code" field on the sign-in screen. You can't join from the menu after signing in. If you're already signed in, sign out first, then enter the code in the "Organization code" field on the sign-in screen and sign in again.
Even if the organization code is wrong or there aren't enough accounts, the sign-in itself succeeds; you just aren't added to the organization. If "Organization" doesn't appear in the profile menu, check the code and the number of available accounts with your admin, then sign in again using the steps above. Anyone who knows the organization code can join. Don't share it with anyone who isn't involved. An account can belong to only one organization. If you already belong to another organization, "You already belong to another organization." appears and you can't join.
Starting with 2.13, you can open the app from the link in the welcome email and join. You don't need to type the organization code.
Admins don't enter an organization code. When you open the link for admins, no organization code is entered, so just sign in with the email address that received the welcome email. If the app doesn't open, copy the organization code from the page, select "Have an organization code? (business)" on the app's sign-in screen, and paste it into the "Organization code" field.
Talk to us about getting started
Enterprise is a contract based on a quote (minimum number of accounts: 1). You can't apply directly from within the app.
You can check the contract status (status, payment method, next renewal), the number of accounts and the detection time per account, and the organization code in "Organization management" > "Billing" in the app. Only admins can open "Billing" and "Invoices".
Organizations that previously subscribed by credit card can change the number of accounts and use the payment portal (change the payment method, view receipts, cancel) from "Billing", and check past charges from "Invoices".
For contracts paid by bank transfer, we send invoices as PDFs by email. They don't appear in "Invoices" in the app ("No invoices yet." appears there). The following applies to organizations that previously subscribed by credit card.
"Organization management" > "Invoices" lists all past charges, newest first. Only admins can open "Invoices".
After payment, it may take a little while for the status to change to "Paid". Select "Refresh" to get the latest status. Bank transfers are reflected after the payment is confirmed.
Enterprise included hours are not allotted to each member individually; they are a pool shared by the whole organization.
Both admins and members can check usage.