Last Updated on September 20, 2026 by Practical Astrophotography Staff
PixInsight has a reputation that arrives before the installer does. People talk about it the way they talk about a difficult instrument: expensive, unfriendly at first, and then hard to put down once it starts making sense. If you are sitting on the trial and trying to decide whether 300 euros is money well spent, that reputation is not very helpful. What you need is a clear picture of what version 1.9.5 actually is, what the trial lets you do, and how to spend 45 days so you can make the call with your own data instead of other people’s finished pictures.
This article is written for that situation. It assumes you already take deep-sky photos, or you are about to, and that you are not looking for a sales pitch.
## What you are actually downloading
PixInsight is an image-processing platform built for astronomical data. It is not Photoshop with a few astronomy filters bolted on. It is closer to a workshop: a large set of tools that operate on linear (unstretched) data, plus a scripting system, plus a file format designed around those tools. Pleiades Astrophoto, the Spanish company behind it, sells a single commercial license and then keeps the 1.x line updated. There is no monthly subscription.
As of mid-September 2026 the current release is **PixInsight 1.9.5 Lockhart**. It runs on Windows, macOS, and Linux. The 1.9.5 release added a MachineLearning module whose first public tool is MLDenoise, a documented format for storing astrometric solutions inside XISF files, and a new distortion model called recursive surface splines. Weighted Batch Preprocessing (WBPP) also moved to version 3.1.0 and now requires 1.9.5.
If a YouTube video from last year shows 1.8.9 or 1.9.3, the bones of the program are the same. The menus have not been reinvented. What has changed is the engine underneath, the official preprocessing script, and a few tools that used to exist only as third-party add-ons.
## The money question, stated plainly
A commercial license costs **300.00 euros** if you live outside the European Union. EU buyers pay that amount plus the VAT rate that applies in their country. The FAQ last quoted an approximate dollar figure around 323 USD; the invoice is in euros, so the dollar amount moves with the exchange rate.
That license is sold to individuals, not clubs or companies, through the official purchase page. One license covers every computer you own and every supported operating system. After the first online activation, the commercial program does not need the internet to launch. Updates for the 1.x series are included. There is no refund once you have downloaded the commercial copy. That last point is why the trial exists.
So the decision is not “rent this for a year.” It is “pay once, keep using the current generation, and live with the fact that you cannot send it back.”
## What the trial really is
The trial is a **45-day, full-featured copy**. No watermarks. No locked tools. No cap on image size. During those 45 days you also get email support and software updates.
The limits that do exist are administrative, not cosmetic:
- You need a working internet connection every time you start the program. The trial checks the license online.
- The trial is tied to one machine and one operating system.
- Proxies and VPNs are not allowed when you request or run the trial. Disposable email addresses are rejected. Microsoft consumer addresses (Outlook, Hotmail, Live, and the rest) have been blocked for years.
- The clock starts when the license is approved, not when you first feel like using it.
Request the trial from the official trial page. Use your real name and a real postal address. Approval is manual. If the form looks fake, nothing happens. After approval you get an email with download instructions and activation data.
If you later buy the commercial license you do not have to throw the installation away. On Windows, people commonly install the commercial build over the trial and swap the license key when prompted. Customizations and repository settings usually survive. On other systems the advice has been more mixed; follow the current install notes that arrive with the commercial download.
## Will your computer run 1.9.5?
Check this before you burn a week of the trial fighting the installer.
**Windows.** 1.9.5 has been tested on Windows 11. Windows 10 sometimes works; it is not guaranteed. Older Windows versions are out. On Windows and Linux the CPU must support AVX2 and FMA3. That knocks out a lot of machines from before about 2013–2015. The installer checks this at startup.
**macOS.** Officially, macOS 15 Sequoia or newer, on Apple Silicon or Intel. The developers say 1.9.5 should run on macOS 14 Sonoma but they have not tested that combination. Native Apple Silicon builds have been available since 1.9.4.
**Linux.** GLIBC 2.35 or newer, X11, no Wayland, no Nouveau driver. Kubuntu 24.04 LTS is the recommended setup.
RAM is the other hard limit. Sixteen gigabytes is listed as a minimum and is not useful for real work. Thirty-two gigabytes is the floor if you intend to calibrate and stack your own frames. Sixty-four is comfortable for many one-shot-color sessions. One hundred twenty-eight is the number the developers keep repeating for modern CMOS cameras and larger mosaics. Fast NVMe storage matters as much as raw core count, because PixInsight writes large intermediate files and uses swap space heavily.
After you install 1.9.5, run **Process > Thread Performance Analysis** once. Old thread data from a previous version can make a new install feel slower than it should.
On Windows, add exclusions in Windows Security for `.xisf`, `.xpsm`, `.xdrz`, and `.xnml`. Real-time scanning of those files can stall saves and project loads.
## First hour in the program
The first launch looks unfinished. That is normal. You get a dark workspace, a process console that prints text like a compiler, explorer panels that can be hidden, and toolbars that do not resemble Lightroom.
A few habits save a lot of time:
Open **Process Explorer** and leave it parked where you can search. Almost every tool lives there. Typing part of a name is faster than hunting menus.
Learn the difference between a **process** and a **script**. A process is a compiled tool with its own window. A script is JavaScript that calls those tools in a sequence. WBPP is a script. HistogramTransformation is a process.
Every process window has the same small row of icons at the bottom. The blue square applies the process to the active view. The circle applies it to a preview. The document icon creates a process icon on the workspace so you can reuse the settings. New users ignore those icons and then wonder why nothing happened.
**ScreenTransferFunction (STF)** is the stretch you see, not the stretch you save. It is a display stretch. Your file stays linear until you run a real stretch such as HistogramTransformation or MaskedStretch. If you skip this idea, every later tutorial will sound like nonsense.
**Previews** are small working copies of a region. Draw one, apply a process to the preview, and only then apply it to the whole image. That is how you avoid waiting two minutes to discover a setting was wrong.
Save often in **XISF**, PixInsight’s own format. It keeps 32-bit data, astrometric solutions, and metadata that FITS will mangle or drop. Export TIFF or JPEG only when you need a file for the rest of the world.
Projects (`.xosm` plus the project folder) store the workspace: images, process icons, previews. Use them. Closing PixInsight without a project and then trying to reconstruct a session from memory is how people lose evenings.
## Use the official starter pack first
Starting with 1.9.4, PixInsight ships a starter pack under **RESOURCES > Learning**. It is also on the public start page. The pack is short on purpose: a welcome note, a quick-start guide with sample images, a companion to the official YouTube introduction series, and a document on the ideas behind calibration and stretching.
Do that before you install a pile of third-party repositories. The temptation is to skip straight to a “ten-step workflow” video that assumes BlurXTerminator, NoiseXTerminator, StarXTerminator, and a custom process-icon set. Those videos are useful later. In week one they hide the program you are trying to evaluate.
The official YouTube channel is [@PixInsight](https://www.youtube.com/@PixInsight). Watch the interface videos with the program open. Pause and click the same buttons. That sounds obvious. Most people watch them like entertainment and then sit down cold.
## Set up folders before you process anything
PixInsight does not care how you name files, but you will. A simple layout is enough:
- `raw/lights`
- `raw/darks`
- `raw/flats`
- `raw/bias` or `raw/flatdarks`
- `masters`
- `process`
- `export`
Keep Calibration frames matched to the lights: same gain/offset, same temperature if you have cooling, same optical train for flats. If you mix sessions, put them in separate light groups. WBPP 3.1.0 can handle mixed geometry in one run, but that is a convenience, not a reason to be sloppy.
Give the program a fast scratch disk. In Global Preferences you can point swap storage at an NVMe drive. If swap lands on a nearly full system disk, Stacking a few hundred frames will feel like the machine has died.
## Preprocessing: this is where the trial starts to tell the truth
Most of the “is this worth it?” answer shows up in calibration and stacking, not in the final color tweak.
**WBPP 3.1.0** is the current official batch preprocessor. It calibrates, weighs, registers, and integrates lights. It can reuse reference frames, recover an interrupted run, and keep mixed-geometry groups in one session. FastBatchPreprocessing (FBPP) now shares the same version number. Both need 1.9.5.
A first WBPP run should be boring and correct:
1. Add lights, darks, flats, and bias or flat-darks.
2. Confirm the script grouped them the way you think it did.
3. Set an output directory that is not inside the raw folder.
4. Leave drizzle off until you have a clean integration and a reason to use it.
5. Let it finish. Read the logs. The log is part of the product.
Then open the master light. Apply an auto STF. If the stack is a grey mush with no stars, you have a calibration or alignment problem. Fix that before you touch denoise or saturation. A pretty stretch on a bad integration wastes the rest of the trial.
SubframeSelector still exists if you want to inspect FWHM, eccentricity, and SNR by hand. WBPP already weighs frames. Manual rejection is for nights you do not trust.
If WBPP fails, the cause is usually one of these: mismatched flats, files with no usable stars, a path with odd characters, or a disk that filled up halfway through. The console text looks hostile. It is usually specific once you scroll to the first red error.
## After the stack: linear work first
A master light from WBPP is linear. The Histogram is crushed to the left. That is correct.
The usual order, in plain language:
**Crop.** DynamicCrop or a simple crop to remove stacking edges.
**Gradient removal.** DynamicBackgroundExtraction is the older interactive tool. GradientCorrection and MultiscaleGradientCorrection are the newer path, and they can use the MARS sky model if you download that database from the distribution system. You do not need MARS on day one. You do need a flat background before color calibration.
**Color calibration.** SpectrophotometricColorCalibration (SPCC) is the current built-in method if the image has an astrometric solution and you have the Gaia data PixInsight expects. Plate-solve with ImageSolver first if SPCC complains. PhotometricColorCalibration is the older cousin. Manual color sliders are a last resort, not a style.
**Deconvolution / sharpening.** The built-in deconvolution tools are still there. They need a good PSF and a good mask. This is the step that makes beginners feel the program is hostile, because a bad deconvolution looks worse than no deconvolution.
**Noise reduction.** 1.9.5 gives you MLDenoise as a first-party machine-learning tool, with optional GPU paths (CUDA on NVIDIA, CoreML on Apple Silicon) and a local-support mask. MultiscaleLinearTransform is the older, fully manual method. You can evaluate the new tool without buying anything extra.
**Stretch.** HistogramTransformation with a preview, or MaskedStretch if the stars blow out. After this the image is nonlinear. Some tools want to stay on the linear side. That is why people duplicate the view before stretching.
**Star handling.** StarNet and StarXTerminator are add-ons. PixInsight can still work on stars and nebulosity together. Starless workflows are popular; they are not mandatory for a trial verdict.
**Curves, HDR, saturation.** CurvesTransformation, HDRMultiscaleTransform, and a light SCNR pass for neon green. This is the part that looks like ordinary photo editing. It goes better if the linear work was clean.
You will not learn all of that in 45 days. You can run that sequence on two or three of your own targets and know whether the program fits your head.
## Built-in tools worth forcing yourself to try
Ignore the full process list. These are the ones that decide whether you keep paying:
- WeightedBatchPreprocessing
- ImageSolver / SPCC
- DynamicCrop
- GradientCorrection or DynamicBackgroundExtraction
- ScreenTransferFunction and HistogramTransformation
- CurvesTransformation
- RangeSelection and RangeMask, plus a basic luminance mask
- MultiscaleLinearTransform
- MLDenoise
- PixelMath at the “copy this expression from a tutorial” level, not the “write a pipeline” level
If you can take a messy OSC stack to a finished image with only those, you understand enough of PixInsight to judge it. Everything else is refinement.
## Third-party tools, and why they confuse the purchase decision
A lot of modern PixInsight videos are not demonstrations of PixInsight. They are demonstrations of RC Astro’s XTerminator tools plus a few scripts. BlurXTerminator, NoiseXTerminator, and StarXTerminator are paid modules. They are widely used because they are fast and they look good. They are not included in the 300-euro license.
Seti Astro, StarNet, and various repository scripts are a separate ecosystem. After a major release such as 1.9.5, some of those scripts break until their authors post updates. That happened on 1.9.5 with several popular repositories; it is normal, and it is temporary.
For the trial, install extra repositories only if you already know you want to test a specific paid module. Otherwise you will spend day four fixing a script that failed to register instead of processing the Rosette. The core program is what you are buying from Pleiades. The add-ons are a second bill, and they should be a second decision.
## A 45-day plan that does not waste the clock
**Days 1–2.** Install. Run Thread Performance Analysis. Work through the official starter pack and the first official interface videos. Open a single good image that is already stacked, even if it came from another program, and learn STF, previews, and process icons.
**Days 3–7.** Put one complete session through WBPP. Bias/darks/flats/lights that belong together. Get a master you trust. Write down what failed.
**Days 8–20.** Process that master with built-in tools only. Crop, gradients, color, a cautious denoise, a stretch, curves. Save a project every night. Export a JPEG you would actually post. Then do a second target in a different category: a galaxy if the first was a nebula, or the other way around.
**Days 21–30.** Repeat the same targets with one extra technique each time: a better mask, SPCC instead of a guess, MLDenoise instead of a heavy wavelet denoise, a star reduction that is not an add-on. Compare versions. The comparison is the education.
**Days 31–45.** Process a night that is not easy: thin cloud, gradients, a target low in the south, a mosaic if you already shoot those. This is the data that makes people keep the license. Pretty textbook files make every program look good.
Keep a text file of settings that worked. PixInsight will not remember why you chose a value. You have to.
## Things that make beginners quit, and what they usually mean
**“It looks like nothing happened.”** You applied a display stretch only, or you applied a process to the wrong view, or you applied it to a preview and expected the main image to change.
**“Everything is green.”** Linear OSC data often is. SCNR after a proper color calibration, not before.
**“The stars turned into donuts.”** Overdone deconvolution or a bad PSF. Undo. Mask the stars. Try a weaker setting.
**“SPCC failed.”** No valid plate solve, incomplete Gaia files, or a tiny crop with too few stars. Solve the image first. Download the Gaia data sets the tool asks for.
**“WBPP ran for hours and produced a blank file.”** Read the first error in the log. Check disk space. Check that lights are really lights.
**“My favorite script vanished after 1.9.5.”** The author has not posted a build for this core yet, or the repository URL changed. The official tools are still there.
**“The program wants the internet.”** You are on the trial. That requirement goes away after commercial activation.
None of these are signs that you are bad at this. They are signs that PixInsight tells you almost nothing unless you ask it in its own dialect.
## How to decide if the license is worth it
Price only makes sense against the alternatives you would actually use.
Siril is free and much better than it used to be. AstroPixelProcessor is paid and is built around calibration and stacking. Photoshop and Affinity can finish a stretch if something else does the linear work. PixInsight’s case is the middle of the pipeline: controlled calibration, statistically heavy integration, gradient tools, spectrophotometric color, and a non-destructive workspace that still makes sense six months later when you reopen the project.
Buy it if, by day 30, you can point to a specific result you could not reach as cleanly in the tools you already own, and you can see yourself doing that work again. Do not buy it because a forum thread said serious imagers use it. Do not buy it on day two because MLDenoise looks magical on someone else’s 20-hour integration.
Skip it, or wait for another trial later, if your bottleneck is still acquisition: tracking, focus, flats, or too few hours on target. Software will not invent signal. Also skip it if your computer is under the RAM and CPU line and you were hoping the trial would feel snappy anyway. It will not.
Remember the no-refund rule. If you are 80 percent sure, finish the trial on your worst data, not your best. Worst data is the honest test.
## If you do buy it
Purchase from the official commercial license page, as an individual, with the same identity you intend to keep. You get download instructions and a permanent account on the distribution servers. Install 1.9.5 from there. Activate once while online. After that you can work offline.
Keep the installer. Major versions are full installs, not tiny patches. When 1.9.6 or 1.10 appears you will download a new package, not click “update” and walk away.
Re-add third-party repositories only after the core is running and WBPP has produced one clean master on the commercial install. Then update scripts. Then, if you still want them, look at paid modules as a separate budget line.
## A last practical note
PixInsight is large because the problem is large. Raw astrophotography files are dim, noisy, and full of artifacts that ordinary photo editors were never asked to model. The program’s manners come from that. It shows you numbers. It keeps data linear until you say otherwise. It lets you ruin an image with one slider and expects you to use previews.
The 45-day trial is long enough to get past the manners and short enough that you should not wander. Install 1.9.5 Lockhart, use the starter pack, stack one real night with WBPP 3.1.0, finish two images with the built-in tools, and only then decide whether 300 euros buys you a workshop you will keep open. That is the whole test. Everything after that is practice.
