What changed on the site and in the product, newest first. Only what you can see for yourself.
On a computer the counting protocol is now the same chip as on a phone, at the top right of the sample's result card, which gives the image grid the height the old row took. Before the first image it sits at the foot of the panel. The label over the sample's number now reads "Average of 2 images · 0.50 µL total", so the volume the concentration rests on is in view.
Removing an image from a sample now takes one tap on its red X, with no question to answer. An Undo button appears for ten seconds, and Ctrl+Z works too, for a single image or a whole cleared sample. A run of deletes can be walked back one at a time. On a phone, Clear sample and the X on the image in view now sit on the counting page itself.
Coming back to the counting page with images from your last sample now asks at once whether to continue it or clear it, then shows a loading screen. Images no longer appear on a blank page seconds later, on top of whatever you had started. They also load faster.
Panning, zooming and switching between the grid and cell tools now stay smooth on images with thousands of cells, where they could stall for seconds. Leaving Fix grid is immediate, and the page uses far less memory.
Fix grid can now count one image whole with Full image, even when the protocol counts squares. Marking grid corners hides the cells so the ruling stays readable, and the corner points are smaller. Cells found outside the counted squares show as a faint outline, and the tips for Fix grid and Fix cells now sit just above their toolbars.
A close-up of one of the four corner squares now reads by itself: the small squares inside it give the scale, and each one can be selected and counted, as in the centre square. A view that holds only part of a large square no longer needs marking by hand. The grid detector also makes fewer false marks. A scale now takes at least three found points, or two when you placed one of them by hand.
The link in your sign-up email now takes you straight into CellCount, to the page you were heading for, rather than the home page. If a team invite brought you here, it returns you to that invite. Open the link on your phone and the window you signed up in signs you in by itself.
Everyone now has a personal account, just them, to count in from the start. A team is your lab's shared account: its people, its images and its counts. When you sign up you are offered your colleagues' teams to join, or you can make your lab's team and invite them. Team owners choose whether colleagues join straight away or ask first.
You can now change your name in the account menu. The top bar shows what you are counting under: your lab's picture in a lab's account, and your initial in your own. You can count on your own from the start, and set up your lab and invite colleagues whenever you are ready.
A lab can now have its own picture, shown beside its name in team settings, in the top bar and in the account menu. Any picture works, in any shape or size, and it is shown whole, never cropped. Where the lab's institution has a logo on Wikidata, team settings offer it with one click.
A new account is now asked for its lab's name and its institution or company, with the institution filled in from a university or company email address or Google account where it can be. Start typing and a list of universities, hospitals and companies appears; anything not in it can be typed in. Both can be set or changed later in team settings.
When you choose which squares an image shows, the drawing of the whole chamber now marks the square in solid green with a ring around it, over a lighter ruling, so even the smallest centre squares are easy to find. In the protocol builder, The whole image is drawn as the plain image, with no ruling.
A protocol can now count each image whole: choose The whole image in the protocol builder and say how many photos to take. Every cell in a photo is counted at the one scale its grid or its file gives, except one cut by the photo's bottom or right edge, so none is counted twice. It is meant for a microscope camera with well set optics, and gives far more volume per photo than one square. Tapping the size on the scale bar opens the square size choice.
On a close-up of the central square, the grid now also uses the large squares' corners it finds to place the small counting squares. A square at the central square's corner used to be drawn a little off the ruling there; it now sits on it. Corners the grid already found stay where they are, and the size of a square is read as before.
An image zoomed in so far that a whole counting square no longer fits used to open a red warning on its own. It now shows a yellow note, labelled Square does not fit, that stays folded until you open it, with what to try: a lower objective, a stronger reducer or a wider capture. Its cells are counted as before. A scale too low to resolve cells still gets the red warning.
On a phone, fixing cells or the grid now folds the sample panel away and puts the tools in a strip under the image, so the image gets most of the screen. Done brings the panel back. Placing grid corners no longer gets in the way of moving around: drag to pan and pinch to zoom without dropping a corner, and tap where one should go. With a mouse, a drag pans and a click places.
An image with no scale used to put a red warning over the whole sample. It now gets a folded yellow note instead: the image still gives its cells and viability, and misses out only on a concentration until it has a scale. Mark one square on it, or capture with the chamber rulings in frame. Its exported figure carries the same note in yellow. A photo too small to count, or a scale outside the operating envelope, still gets the red warning.
Clumps the model found are now drawn on every image by default, in purple, so you can see what was left out of the count at a glance. They are still counted separately, never in the cell count, and the clumps button under Display turns them off.
The sample's concentration, viability and cell count are now big enough to read from across the bench. The concentration shares a frame with its dilution, and viability with the Total count and Viability switch, so it is plain which setting moves which number. The Sample result list is gone, since the report carries it; an image is left out from its This image card. With a cell picked, Shift flips it between live and dead. On a phone, the camera button is larger.
Edit squares and Select squares on the image are now one Fix grid button. Its bar switches between Squares, to pick the squares you counted, and Corners, to place or remove a grid corner. A tool's bar takes the buttons' place and ends in Done, so more of the image stays in view. On a computer, every correction on the image is listed in the Tools panel, and pointing at one lights up its cell.
With a mouse, the choices for a cell now open in a small menu on the cell itself, so the pointer has almost no distance to travel, and the arrow keys move to the next cell that way. A right-click on a cell, on empty image or on a grid corner does its one job and leaves everything else as it was, with no mode left switched on.
Fix grid on the image now has its own bar: undo and redo, put the detected grid back or use it for the first time, clear the grid, change the square size, or mark a square when no grid was found, with the reason it was not found said right there. On a phone, an image that needs a square marked makes the Fix grid button pulse. Whether the detected grid is applied by itself is now part of a counting protocol, so a lab sets it once for everyone.
The default sensitivity in the protocol builder now lists Low, Balanced and High in that order, the same as the counting page, so the choice reads the same wherever you make it.
A Fix cells button on the image lets you change or delete a cell, or add one the count missed, on a phone as on a computer; on a computer a right-click does the same. Tap a missed cell and, where the model saw it below the sensitivity cut-off, its own outline is used. Undo, Redo and a list of every correction on the image sit in one bar, and a deleted cell can now be brought back.
Display on the counting page now has a Layout choice: Auto, Mobile or Desktop. Mobile gives a touchscreen kiosk or a tablet the phone layout at full width, with the controls sized for a finger. The choice is saved on that device.
While you draw cells with a mouse, in Fix cells or Annotate, thin horizontal and vertical lines now follow the pointer across the whole image, so a box edge lines up with a cell edge or a grid line at any zoom. On a phone, Fix cells and Fix grid are round icon buttons in the lower left corner, away from Fit, and the scale bar is gone, leaving more of the image in view. Turn one on and its bar takes their place, ending in Done.
On a phone, the sample’s concentration now sits above the controls, labelled as the average of the images in the sample, with the dilution it was calculated at under it. Viability and the cell count sit in the next column, with the switch to + Viability under them. The counting protocol is in the row below, there before your first photo too. Each one opens a short explanation and the choices, including how the average is weighted by the squares each image counted.
The help button on the counting page now walks through a count in five steps, each with a picture. The counting step shows a real chamber frame outlined the way the app draws it, with the Total count and + Viability switch to try. On a phone, the steps name the phone’s buttons.
A count now reports the total count by default: live and dead cells together, with clumps counted separately, never in the cell count, and every counted cell boxed in white (Display can add a gap around each box). For trypan-blue viability, switch the result to + Viability. A protocol can make either one your lab’s default. With no dilution stated, a total count is read as undiluted, and + Viability as 1:1 (×2 dilution), which doubles the concentration; a dilution you state always stands.
Finishing a sample now offers two buttons, Next sample and See report. The full export is on the report page. On a touchscreen, the sample name field stays in view when the on-screen keyboard opens.
On a phone, the Tools button now says so, the menu lists Protocols so you can set up how your lab counts before your first photo, and Undo all corrections is there whenever an image has any. The help pages now describe the screen you are on.
The two handles on the Cell size slider are larger, and each catches a touch that lands just beside it, so the size range can be set with a gloved finger at the bench.
When a sample is outside the operating envelope, each image that put it there now carries a warning triangle in the sample, red for a capture that is out and amber for one in the caution band, so you can retake or remove it straight away. The warning itself folds to one line, so it no longer crowds out the images.
Vitr, the company behind CellCount, is part of Microsoft for Startups. Its partner badge now sits with our other supporters on the front page and the About page.
Buttons, links and highlights across the site now take a deeper teal drawn from the Vitr identity, running from the logo toward the navy, and dark enough that white text on a button reads clearly. The ViTR logo keeps its own cyan, which nothing else on the site uses.
Sensitivity is now a row of rising bars from Low through Balanced to High, in place of three buttons, and the presets are listed in that order everywhere. The size gate has its own Cell size tool, open whenever the image has a measured scale, and pointing at a bar in its chart shows the size range and how many live and dead cells fall in it. The tools panel on the counting page opens every time you arrive, and a chevron on its button shows that it folds away.
The hemocytometer guide now draws the improved Neubauer ruling square by square, with the volume over each square and its multiplier, shows how to seat the cover glass so Newton’s rings appear (pressed and rubbed, since moisture can dilute the sample), and says how to clean the chamber. The counting app page draws what one image holds under each protocol choice, with the crossings the grid detector reads in it.
The front page and the About page now describe the benchmark for what it is: sixteen images counted side by side by CellCount and the most popular automated cell counter. The figures are unchanged: how closely the two agree on cell concentration, and the limits of that agreement.
If CellCount is updated while a page is open, the next page you open now reloads into the new version instead of stopping with an error. If a page ever does stop working, it shows a Reload button and reloads by itself after a short wait, so a screen in fullscreen or kiosk mode recovers without a keyboard.
The Vitr wordmark in the footer is now in its own cyan, and the small card that holds the scale bar beside the zoom buttons carries it too, in place of the old cell icon. The browser tab and home screen icons show the V of the wordmark, cyan on navy, so it reads at that size. The scale bar and everything else on your image are unchanged.
A sample counted from TIFF images, such as an instrument or microscope export, now shows each counted frame as a figure in its report, and the full export writes the overlay images for it. Before, those figures were missing and the export had no overlay folder.
The concordance, the limits of agreement and the live-or-dead agreement quoted across the site now say which model they were measured on: our released Huginn 2.0 trypan blue exclusion model. A future model will be published with its own figures.
A new page sets out cell counting as a whole: the methods labs use and what each one reads, whether it tells live from dead, the cells per mL formula, trypan blue viability, how many cells to count and how to choose. It links down to the hemocytometer guide, the calculator and the phone guide for the details.
The phone guide now shows where a phone held to the eyepiece lands at 10x, 20x and 40x, against the photo sizes common phones take, and what it means to have the whole square in the frame. The app guide shows the same for a microscope camera with each objective and adapter, and the specifications now carry the operating envelope in full, with all three figures. Every figure comes from the operating envelope itself.
An image counted on one of the smaller squares now only has to hold that square. A crop that a 0.25 mm corner square nearly fills, taken at high magnification, is inside the operating envelope instead of flagged as zoomed in too far. When a 1 mm square does not fit the frame, the result suggests counting the corner squares instead.
The figures on the front page are now how closely CellCount agrees with a leading automated cell counter on the same images: Lin’s concordance on cell concentration and the Bland-Altman limits of agreement. The plots, the data behind them and the method are on the About page.
The header and the footer now read CellCount by Vitr, set in the typeface of vitrbio.com, and the browser tab and home screen icons show the Vitr wordmark. A link to the site now shares a new card. The counting view is unchanged: the colours on your image still mark live and dead cells and the region being counted.
The grid detector now finds the crossings of the 0.2 mm medium squares inside the centre large square as well as the 1 mm ones, and tells them apart. A photo inside the centre square, with no 1 mm line in view, now takes its scale from the medium squares instead of reading them as large ones, and a frame showing both checks one against the other. Each kind of crossing has its own colour. If the rulings cannot be found, you place the crossings by hand.
The protocol builder now asks what each image has: one large square, the whole ruling, one 0.2 mm medium square of the centre, or a 2 x 2 block of them. On the medium squares you pick any of the 25, or a quick pick such as Four corners and centre. Under such a protocol the grid applies by itself when the detector finds the medium squares' crossings, and an image that shows the large squares instead is counted at 1 mm and says so.
Under a protocol of the centre's medium squares, the sample sheet and the camera's guide draw the same squares, one an image or a 2 x 2 block at a time, and the guide says how many squares along the row or the column the next one is. Where no grid is found, Mark one block places the lines inside a block from its four outer corners. An upload stops at the number of images the protocol calls for.
Each image now shows the size of the squares it counts: the size the grid detector read, the protocol's, or 1 mm until you say otherwise. A guide puts one square from your photo beside drawings of the four Neubauer improved squares so you can pick the one you counted. The scale, the volume and the concentration follow your choice, on other images taken the same way too. Hand-marked squares give a concentration once their size is known: the protocol's where it states one, or the one you pick.
After you mark one square by its corners, the same squares are now drawn out to the edges of the image. Tap a corner further out and every square between it and the one you marked is counted, and the drawn squares line up with the ruling there as well, which corrects them where the camera sees the chamber at a slight angle. This also works where a few crossings were already found on the image, such as the corners of the larger squares around the ones you count.
An image is now held to what it counts. A frame only has to fit one whole square, or the whole block it counts, with a little margin, so a capture of the medium squares no longer has to be as large as one of a 1 mm square, and is no longer marked out of range when it is not. The detail needed across each cell is unchanged.
Under a counting protocol, tap the small map in the corner of the camera view to see the whole ruling of the chamber, drawn to scale, with the squares already taken and the next one marked. Beside it the same drawing zooms in on that square until it fills the view, which is how the camera should show it once the stage is there.
The protocol builder no longer has a magnification field, and the camera's start checklist no longer lists one: every image takes its scale from its own grid, so the objective never changed a number. An image that fails to process no longer takes up one of the images a protocol calls for, so its replacement can be uploaded straight away.
A protocol no longer needs a name: one left empty is named from what it counts, such as "4 corners, both sides". The field that labelled the dilution now holds notes for whoever counts, one a line, shown on the checklist before counting starts and on the sample report. The protocol builder in the demo is now the same as the one in the app.
A sample report now says where each image's scale came from: the grid detector, a grid corrected by hand, or one marked by hand. It says the same of the square size and the dilution, and marks any that nobody stated as assumed. A square size stated once in a sample now carries to the images marked by hand after it.
When an image gives no concentration, because nothing gave it a scale or nobody has said which squares its hand-marked grid shows, the sample's concentration is pooled from the other images and names the ones it left out. The list of images under the concentration then has a switch on each one: untick an image and it is taken out of every number the sample reports, its cells included. Tick it again to count it.
An open count now takes up to 150 uploaded images in one sample, and a protocol takes as many as it calls for. Past the latest 18, an older image's square shows its picture, and tapping it opens a smaller copy made on this device, marked as one, with its counts in the sample as before. An image whose picture is not on this device still counts, and says so.
An image that fails to process, such as a file that cannot be read, now gives back the image it took from your team's monthly allowance, within about a minute. The same holds for jobs sent through the API. A failed image also no longer stops the counting page, and an image past the first 18 no longer drops out of the sample's totals while its counts arrive.
The hemocytometer calculator now opens with the cell count formula written out, then the calculator, then three worked examples: corner squares, a dense sample in the centre grid, and a two-step dilution. It also says how to calculate the dilution factor.
A new answer in the FAQ, in English and in Norwegian: there is an API, with a Custom plan. It says what you send, what comes back and why the counts are not a concentration, and links the API documentation.
Moving from one image to the next on the counting page could open the "This image" card in red for a moment, saying "No scale", even when the photo was in range. The card now shows the image's own scale from the start, and so does an image whose count has just come in.
Under a sample's concentration you can now open a list of the images it is pooled from, with the scale each was counted at and the volume it counted. An image with no scale says so. One image read at the wrong scale stands out from the rest at a glance.
Every page that says CellCount finds the hemocytometer grid now also says what happens when it cannot: you place the crossings by hand. That covers the phone guide, the manual method, the specifications and the FAQ, in English and in Norwegian.
The front page, the first FAQ answer in English and Norwegian, and the summary for language models now open with the same plain description: a web app with a free plan that counts cells in hemocytometer images, finds the grid by itself or lets you place the crossings by hand, and gives viability and cells per mL with the dilution you set.
The closing paragraph of the hemocytometer calculator now says what the guides already do: CellCount finds the grid in your photo by itself, and if the rulings cannot be found you place the crossings by hand.
Whether CellCount is free used to be answered near the end of the FAQ. It now comes straight after what CellCount does and whether you have to draw the grid, in English and in Norwegian. The answer itself is unchanged.
Counting by hand? The new calculator turns a hemocytometer count into live and total cells per mL, viability, the cells in your suspension and the volume to seed, for any dilution and for large or centre squares, with the arithmetic shown. No account needed.
Three pages have plainer headings: "Count cells from a phone photo of a hemocytometer", "CellCount specifications" and "Cell counting terms, defined". The phone guide now also says that a phone photo counts on the Free plan, and what to do when the rulings in a photo cannot be found: place the grid’s crossings by hand.
Two new answers in the FAQ: what the Free plan includes, and how CellCount finds the hemocytometer grid in your image by itself, with the crossings placed by hand only when the rulings cannot be found.
Under its opening paragraph the front page now links to how to count cells with a hemocytometer, to the free hemocytometer calculator and to counting from a phone photo of the chamber. The footer on every page links the two guides.
The list of your corrections beside the counting screen now names each one, such as a cell changed from live to dead or a box you drew, instead of printing its pixel coordinates, and pointing at a row lights its cell up on the image. Change back on a changed cell returns it to what the model called it. Until now that button took the cell out of the count altogether.
A new page says what CellCount does with a photo of a hemocytometer in three steps, then shows one real counted frame, what comes back, how it was tested and which chamber and stain it counts on. It ends with short answers: whether an app can count the chamber automatically, whether you draw the squares, whether a phone is enough, and whether it is free.
With a USB microscope camera plugged in, the camera window now opens on it, even on a tablet or laptop with cameras of its own. Without one, it opens on the camera you last chose from its list on that computer.
Each guide, the hemocytometer calculator and the FAQ now carries an Updated line under its opening paragraph: the day its words last changed on this site, not the day the site was last rebuilt. Pages whose text has not changed keep the date they last did.
With the camera open on a laptop or tablet, the space bar now takes the picture wherever you last clicked, so one hand can stay on the focus knob while the other snaps. A small label under the shutter says so. A space typed into a tag is still just a space.
A stage moves on one knob per direction, so the guide now takes a protocol’s squares in an S: along one row, down, and back along the next. Four corners is three straight moves instead of a diagonal in the middle. Squares are no longer called by a number that jumped around as you went. The report names each one by where it is, such as the top right of Side 1, and the counting screen numbers a protocol’s squares in the order you take them.
A sample whose captures only reach the caution band of the operating envelope now shows its warning closed to one line, one tap from the detail. A rejected capture still opens it all the way. With grid detection switched off, a sample with no scale says so and offers to switch it on, which gives chamber images their scale where they stand.
New accounts used to land on a page that asked them to book a call and then wait for us to switch the account on. That is gone. You sign up, you are on the Free plan, and you can count a sample straight away. The offer of a call is still there because it is genuinely useful for setting a lab up, and so are the docs, but neither is something you have to get past first. The app says what it needs as you go.
The annotated frames in a report and in an export are the pictures people pass around, and until now nothing on them said where they came from. They carry a small CellCount mark on the scale bar, which is the one piece of chrome that cannot be cropped off without taking the calibration with it. Frames with no measured scale get the mark too. Images exported for annotation are untouched: a mark baked into a training image is a defect, not a signature.
The operating envelope sets what an image has to be before we will count it, and a capture that missed it used to produce a result that looked like any other. The result and the report now say so plainly: which images, what is wrong with them, and what it costs the number. An image with no scale counts as outside, because without one nothing can place the capture inside the envelope. Nothing is blocked; the call is still yours.
The image used to stop dead when its edge met the edge of the viewport, which meant a cell right at the border of a frame could only ever be drawn in the strip where the zoom buttons and the scale bar sit. Any point of the image can now be dragged to the middle of the screen, so nothing on the frame is permanently behind a control. The image still cannot be lost: the middle of the screen always stays over it.
On a phone, Back now closes whatever is open on top of the counting screen instead of leaving the page: one press per layer, the way every other app on the device behaves. Back out of the camera and you land on the sample view you opened it from, not on a blank frame. It refuses to leave while a photo is still being processed, so a sample can no longer end up quietly short of a frame.
On a phone the counting screen is now the image, edge to edge, with the controls gathered into one bar pinned along the bottom where a thumb reaches. The frame is wider by the borders that used to sit around it, one finger pans it and two pinch it, and a row of chips steps between the images in the sample without leaving the frame. Finishing the sample is on that bar too, and it lights up as soon as every image has been counted.
Five new pages answer the questions labs bring to a call: how to count cells with a hemocytometer by hand, how to count from a phone photo of the chamber, how to get a second reading of the images an automated counter exports, how to count yeast, and how PBMCs, clumps and messy samples are handled. The FAQ is now also in Norwegian.
A specifications page states what goes in (formats, resolution, chamber and stain, the three ways in), what comes back, what the record holds and what leaves as documents. A glossary defines the terms around cell counting, each with its published source. The front page is now also in Norwegian.
A FAQ page answers the questions labs ask before they try CellCount, in short: how accurate it is, which samples and instruments it works with, what you need, how a count is checked, protocols, the record, exports, and how a lab gets started.
The company behind CellCount is now called Vitr; the legal entity stays Lowkey Vision AS, and nothing changes in the terms or the privacy notice. The company has its own site at vitrbio.com, with notes of its own.
cellcount.ai is the address of the site and the app. lowkey-vision.com still redirects here, so bookmarks and links keep working. Email addresses stay on lowkey-vision.com.