v1.3.0

Cubric Vision 1.3.0

Windows, macOS (Apple Silicon) and Linux portable builds below. Extract and run — no installer.

Updating from an earlier version? On macOS and Linux your app checks for updates on startup and will offer this version in-app. On Windows, download the full zip below — the in-app updater in 1.2.0 and earlier does not work, so this release ships no Windows update bundle. Your projects live in Documents and are untouched either way.

Important

• WINDOWS NOW STARTS FROM CubricVision.exe — start.vbs and start-with-terminal.bat are gone; double-click CubricVision.exe in the folder instead. On a clean Windows 11 machine, Smart App Control blocks .vbs and .bat files outright with no way to allow them, so every way we shipped to start the app was blocked and double-clicking did nothing at all. An executable does not get that treatment. Windows may still warn you the first time with a blue "Windows protected your PC" box because the app is not code-signed yet — click More info, then Run anyway. You only do that once.

• ALREADY ON WINDOWS? If the app currently does not start for you at all, updating cannot fix it — the updater is a .bat and is blocked by the same rule. Download the full build from the release page and unpack it fresh; your projects live in Documents and are untouched. If you update in place, an old app folder is left behind in your Cubric Vision folder — nothing uses it any more and it is safe to delete, about a gigabyte back. Updating from inside the app becomes the recommended way on Windows from this version onward — it no longer runs a .bat, so Smart App Control cannot block it. Getting TO 1.3.0 is a full download either way: the in-app updater in 1.2.0 and earlier is broken on Windows — it downloads the patch and then silently applies nothing — which is why this release ships no Windows update bundle. Windows patches resume at 1.4.0.

What is new

• NO GPU? SKIP THE INSTALL AND RUN IN THE CLOUD — setting up Cubric used to mean installing the ComfyUI engine first, even on a machine that could never use it. The install screen now offers a way straight past it, with a link to the setup video; Cubric then runs entirely on a cloud GPU. Turn it back off in Settings → RunPod whenever you want the local engine after all.

• NEW MODEL: FLUX.2 KLEIN — the fastest image model in Cubric Vision, and the only one that can take things out of a picture: paint over an object, hit Remove, and it is gone in about four seconds. It covers the full set in one model — generate from text, reshape an existing image, follow a depth reference, edit with up to three reference images, clean up detail and upscale — with eight style looks available on every one of those. Quality is modest next to Krea 2: reach for Klein when you want an answer now or something removed, and for Krea 2 when you want the finished piece. Tip: Depth on this model takes two images if you give it two — the first supplies the pose, the second supplies who is in it.

• KREA 2 EDITS FOLLOW YOUR INSTRUCTIONS MUCH MORE CLOSELY — the edit path has been rebuilt around the newer Krea 2 edit engine. Your reference image is now fitted to the shape of the picture you are making, so asking for a new setting actually gives you a new setting instead of quietly keeping pieces of the original background. Restaging someone into a new pose, outfit or scene lands closer to what you asked for, likeness is better, and references whose shape does not match your output no longer come out stretched or soft. Tip: the more of your reference frame the character fills, the better the likeness — crop your reference to the person before using it.

• KREA 2 QUALITY SETTING PRODUCES FAR MORE NATURAL IMAGES — everything made at the quality speed, not just edits. The sampler was pushing the model too hard, which flattened lighting, over-saturated colour and gave skin and surfaces a waxy, plastic look. It is now dialled back and a second detail pass finishes the image: lighting reads like a real photograph, skin and fabric hold real texture, and backgrounds resolve properly instead of going soft. It also sticks to your prompt more literally instead of quietly adding flattering details you never asked for. The fast setting is unchanged.

• BLEND TWO VERSIONS OF AN IMAGE WITH A MASK — when a re-run fixes one part of a picture and ruins another, you no longer have to choose. Paint a mask over the area you care about, select that entry and the one you want to borrow from, right-click and pick Mask composite. Add drops the other entry's pixels into the masked area; Subtract keeps the other entry and takes only the masked area from the one you painted. The blend lands as a new entry in the strip so nothing is overwritten, the edge is softened automatically, and a mask drawn as an outline fills in.

• HOLD ALT TO DRAG A GALLERY CARD STRAIGHT INTO ANOTHER APP — dropping a card into Discord, Photoshop or a browser upload box used to do nothing, because a plain drag only hands over a promise of a file that other apps ignore. Hold Alt while you drag and the card leaves as a real file, so any app that accepts a dropped file now takes it; you can let go of Alt once the drag is moving, and with several cards selected, Alt-dragging one drags them all. A plain drag is unchanged: onto the prompt to reuse a card, into a folder to save it.

• IMAGE DESCRIPTIONS ARE SHARPER AND GET STRAIGHT TO THE POINT — Describe an image no longer starts with "The image shows…"; the description opens on the subject itself so you can paste it straight into a prompt, and it is considerably more thorough, covering subject, clothing, materials, lighting, composition and background in enough detail to recreate the shot.

• PICK WHAT YOU'RE MAKING FROM A STRIP, NOT A DROPDOWN — every operation your model can do now sits above the prompt box as a row of chips with the current one lit. An operation you cannot run right now stays visible but dimmed, and hovering it says exactly what is missing — needs 1 image, paint a mask first, takes at most 2 images — instead of quietly disappearing from a list. The same strip appears inside the parameters popup, clearing the last image drops you back to text-to-image on its own, and the operation you actually chose comes back the moment you add an image again.

• EVERY OPERATION NOW EXPLAINS HOW TO PROMPT IT — operations want completely different prompts and nothing in the app said so. Open the parameters popup and there is now a ? above the operation strip: a short guide for whichever operation you are on — what the prompt is for, a couple of examples, and the mistake people usually make. It follows your model too: the SDXL family is pointed at comma-separated tags while the newer models are pointed at plain sentences.

• CHOOSING A MODEL IS A FULL-SCREEN CONTACT SHEET — the model dropdown is now a grid of preview tiles like the Model Library, so you pick by looking at the work rather than reading names — click the model button above the prompt and it opens the same way every time. LoRA & Upscale settings open from the model's own tile.

• TAB JUMPS BETWEEN THE GALLERY AND THE CARD YOU WERE WORKING ON — one key, both directions: from the gallery it opens the last card you had open in that project, and from inside a card it drops you back to the gallery. It remembers per project and after you close the app. If a project has no cards yet, Tab does nothing at all — and if you delete the card it was holding on to, it forgets rather than sending you somewhere you did not ask for.

• CROP OUTSIDE THE PICTURE — the crop box no longer stops at the edge of your image: drag it past any side and the new area fills with a colour you pick. That flat colour is what you hand to a model next — switch to an edit-capable model, do not mask anything, and just describe what should be there; it paints into the coloured area. Good for turning a portrait into a landscape. The box now also snaps to your image edges when it gets close, and there is a third Resolution mode beside Ratio and Free where you type an exact width and height and the result comes out at exactly those pixels. Images only; video crop is unchanged.

• MASKING IS REBUILT: FOUR TOOLS, AND IT CAN NOW FIND ANYTHING YOU CAN POINT AT OR NAME — Brush is painting, plain and simple. Points is new: click anywhere on the image and the mask snaps to whatever you clicked — a face, a pair of headphones, the coffee cup on the table; right-click to push the selection back off something it grabbed by mistake. Name lets you type what you want instead — hair, or cup and earring — say how many of each to find, and every one comes back as its own thumbnail to pick from. Detect is the face / hand / person finder you already had, unchanged. Everything adds on top of whatever you have painted, so all four mix freely. Anything a detector found now draws green, a new black-and-white view in the strip shows the mask exactly as it is handed to the model, the eraser's ring is visible against dark images, and the prompt box now stays open while you mask. Points and Name run on a new model that installs alongside the engine — expect a one-off download of about 1.7GB after you update. Tip: dots are cumulative, but they describe one region per run — click, hit Add, then start on the next part.

• CTRL+Z ON THE CANVAS — every mask edit can now be taken back — a brush stroke, a Clear, committing a detection — with Ctrl+Shift+Z to put it back again. The history follows you between mask tools, and starts fresh when you open a different image.

Fixes

• The app is no longer slow while connected to a cloud Pod. Opening your model library could leave the grid blank for ten to fifteen seconds, and Settings would sit there before showing how full your volume was or letting you press Disconnect. The cause was on the Pod, not in the app: measuring the volume's space and checking which models are installed were done in a way that made the Pod stop answering anything else until they finished, and each new request started a fresh measurement on top of the last. Both jobs now run out of the way of everything else, the volume measurement is taken once and shared, and it refreshes immediately whenever you install or remove a model. The library and Settings now open straight away — it lands the next time your Pod starts, no action needed.

• If the app fails to start, it now says so instead of vanishing. A crash during startup used to close the window with nothing left behind — no message and nothing in the log. Startup failures now show what went wrong and write the same detail to the log file, so a report has something in it. The update download also names itself honestly: unzipping it gives you a folder marked update-only, which is a patch to apply over your existing copy, not a second copy of the app.

• The log file you send with a bug report is now complete, and it keeps its history. Errors raised by the app itself were being written to a second log file that nothing ever collects, so the one you could actually send was missing exactly the messages a report needs — a failed startup could look like nothing had happened at all. Everything now goes to one file. That file also used to keep only a single old copy, and setting up the engine alone could fill it twice over and push the whole setup out of reach; it now keeps the last twenty, so what happened yesterday is still there today.

• Editing an image in the strip now always uses the image you are looking at. A picture dropped or reused into the prompt box while working inside a history strip used to quietly become the input for everything you ran afterwards, no matter which entry you had selected — so an upscale or an edit could come back looking like a completely different picture, and a lower Denoise would not help. The strip now always works from the entry you have open; choosing a multi-image operation inside the strip simply runs it on the selected image. Video is unchanged.

• Stopping a generation no longer says it finished. Pressing Stop could still pop a "Generation finished" notification with the completion chime, because the engine often completes the step it was already working on and hands back a finished image. That result is still kept — it just no longer announces itself as a successful run you did not ask for.

• Krea 2 upscaling is sharp again. Upscales came out noisy — bad enough to be unusable at full quality, and mediocre on the fast setting. The upscaler now finishes with a short refining pass, and full quality no longer borrows the fast setting's accelerator, so it actually renders at the quality you picked. Both speeds improved; full quality is the bigger jump.

• Your images no longer carry hidden data. ComfyUI stamps the entire generation graph — your prompt, negative prompt, seed, model names and the full node workflow — invisibly inside every PNG it produces. Cubric Vision now strips that out as each image is saved, so you can share or publish your work without leaking the prompts behind it. Pixels are untouched: the file is byte-for-byte the same image, just without the hidden payload. Applies to newly generated images.

• Mac and Linux: video generation with LTX works again. Cubric Vision was installing whatever version of the ComfyUI engine happened to be newest that day instead of the version it was built and tested against, and a recent engine release removed something the video component depends on — so LTX video quietly stopped working, and the app could not even tell the two versions apart to offer you a repair. It now installs exactly the tested version and records what actually landed. Windows was never affected.

• Mac: pictures come out as pictures again, not grey mush. Engine setup on Apple Silicon was installing a development snapshot of PyTorch — an unreleased daily build — instead of a finished release, so the engine you ended up with depended on the day you happened to install it. One of those daily builds rendered every SDXL image as flat grey noise, and it did so silently: no error, a normal generation time, and a normal-looking result in your gallery until you opened it. Mac now installs the same tested, stable PyTorch that Windows and Linux have always used. If you set up the engine on a Mac and your images came out grey, reinstall the engine from Settings and they will be fine.

• macOS: installing a model no longer fails partway through. Setting up any model that uses depth stopped with "Installation Failed" and a Retry that could never succeed — one of the components it downloads asked for a piece of software that is only published for Windows and Linux graphics cards. Cubric Vision now skips that piece on Mac, where it was never usable in the first place. Windows and Linux are untouched.

• Cloud Pods can no longer report a model as installed when nothing downloaded. While a Pod was still starting up, it could answer the "what is on this volume?" question with only part of the picture — and anything missing from that answer was taken as installed, so models you never touched ticked over to installed with nothing on the volume, and the next generation failed on files that were never there. An incomplete answer is now treated as "don't know yet": those models keep whatever state they already had and settle on the next check a few seconds later.

• Gallery cards now match the picture inside them. A card was laid out using the size you asked for rather than the size that came back, so whenever a model produced something a different shape, the card was cut wrong and the image sat inside it with padding. Cards are now measured from the finished image, so the grid lays out cleanly.

• Windows: setting up on a fresh machine now finishes. On a clean Windows 11 laptop the setup ran for an hour, then stopped with "Installation Failed", and Retry re-downloaded everything and failed the same way. Three things were behind it: the download folder ended up with the app's long name written twice, pushing some files past the length Windows allows a path to be; one component asked for a developer tool most people do not have; and the error message blamed the wrong step. All three are fixed — the Windows download unpacks into a single short folder, the unnecessary component is gone, and if anything does go wrong the message names the exact part that failed. Setup also checks up front whether the folder you unpacked into is too deep for Windows and tells you to move it before downloading several gigabytes, instead of after.

• The startup window no longer flashes white. On slower machines the loading window appeared as a blank white box for a few seconds before the logo showed up. It now waits until it has something to show.

• Connecting to a cloud Pod no longer downloads models you never asked for. When Cubric reconnects and finds a Pod component sitting at an out-of-date version, it repairs it — and that repair used to re-run the entire install for every model the component belonged to, quietly pulling down gigabytes that were never on your Pod. On a fresh Pod that could fill a 150GB volume in about a minute, and it took back any space you tried to free. The repair now updates only the component that actually drifted, which is a few hundred kilobytes, and it says what it is doing in the log instead of running silently.

• Installing a model after connecting to a cloud Pod no longer does nothing. The first model you installed after connecting could sit on "Queued…" indefinitely — the app was waiting on a job that had already finished before it started listening, which happened on almost every connect once your models were already on the volume. Installs now start immediately, and if the queue is ever held up for any other reason it says so in the log. Connecting is also quieter: it no longer pops an "engine:assets installed" message every single time.

• The Model Library no longer goes blank when an install or uninstall finishes. The library was rebuilding the whole grid from scratch, throwing away every preview it had already loaded and asking for them all again. Previews are now kept and reused, so the grid updates in place and never blanks — the same when you filter, search, or change a model's options. Video previews show their first frame straight away, and an uninstall now says "uninstalled" rather than "updated".

• An uninstalled model no longer shows a full progress bar where its Install button belongs. The app was still holding the finished install job for a model it had just deleted, so the model looked like it was still there. It now lets go of the job the moment you uninstall, on both your own machine and a cloud Pod.

• Settings no longer forgets where your models are. If your models folder had "temp" or "tmp" anywhere in its path — even a Windows account name that happens to contain those letters — then simply opening Settings quietly pointed Cubric back at its own default folder, and the whole library read as if nothing was installed. That check is gone. Cubric also writes every change of models folder to its log now, so if the folder ever does move unexpectedly, there is a trail.

• The app no longer opens to a black window on a slow machine. Cubric waits for its own background service to come up before showing the window, and after five seconds it stopped waiting and opened anyway — on older hardware, or on a first run from a cold disk, the service was often still starting, so the window asked for a page that was not being served yet and simply never asked again. You were left with a black window and no way out of it, since the app has no reload shortcut. It now keeps asking until the service answers, so the app appears as soon as it is ready however long that takes.

• Linux and Mac: Retry now recovers an interrupted engine setup. The first engine install is several gigabytes, so stopping partway through is easy to do — a dropped connection, a power cut, a full disk, or simply quitting and coming back later. Retry failed to recover from that in two different ways. It used to fail instantly, every time, because the setup tool refuses to reuse either leftover from the first attempt: the Python environment it had created, or the engine folder it had already downloaded. Then it did the opposite — it finished in seconds, announced success, and left an engine that could not start, because it judged whether the install was already done from a file that exists from the very first moment of setup, so a half-finished install looked complete and only the extras were reinstalled. Either way there was no way forward from inside the app; the only fix was deleting folders by hand that nobody would think to look for. Setup now replaces the Python environment cleanly, picks up the existing engine folder where it left off, and checks whether the install genuinely finished before deciding what to repair — so Retry does what it says, and keeps what you already downloaded. Windows was never affected.

Engine

• The local engine now installs ComfyUI 0.29.2, pinned. Earlier builds installed whatever version happened to be newest that day, which is what broke LTX video on Mac and Linux. Cloud RunPod Pods stay on 0.28.0 for this release; nothing changes for you between the two.

Installing and first launch

Both desktop builds are unsigned — there is no certificate and no installer. Signing would start SmartScreen's reputation clock rather than skip it.

Windows

Extract the zip anywhere and run CubricVision.exe. Windows may show "Windows protected your PC" — click More info, then Run anyway. From 1.3.0 the zip extracts to a plain folder with the exe at its root; there is no start script in the launch chain any more.

macOS (Apple Silicon / arm64)

Any downloaded build is quarantined. In Terminal:

xattr -dr com.apple.quarantine "<extracted folder>"

then double-click start.command.

Linux

Extract the tarball and run ./start.sh. Engine setup needs git — the installer will offer to install it if it's missing.

Platform status

• Windows — fully validated end to end on the zip published here: clean extract, first launch, local engine setup, model install and image generation all verified on the maintainer's Windows 11 machine. Launch was separately verified on a clean Windows 11 host with Smart App Control enforced; that second machine has no GPU, so it only covers starting the app.
• Linux — install and launch validation only, on a weak Ubuntu laptop. Generation is unvalidated until a stronger Linux host or a contributor tests it.
• macOS (Apple Silicon / arm64) — fully validated end to end on the zip published here, on rented Apple Silicon: fresh extract, engine setup, model install and image generation, plus interrupted-install recovery. Broader community validation is still welcome.

Reporting a problem? Please include your platform and OS version, CPU architecture, GPU and driver stack, the artifact name, whether you did a clean install or an update, what the launcher did, what engine setup did, whether you tested generation, and the tail of your app log.

Cubric Vision is a local image and video generation app.