I have a simple rule about photo editing tools. If the app needs an internet connection to remove a background or resize an image, I am already suspicious. My photos are my photos. They do not need to be uploaded to a remote server for me to crop one or fix the white balance. The “send to the cloud” era of consumer photo editing has been sold to us as a feature, but the cost is real: a monthly subscription, a privacy surface, and a service that can disappear on the vendor’s timeline.
The cleanest replacement I have found for the Adobe / Canva / PhotoWorks class of tools is a self-hosted photo editor running on a NAS (network-attached storage, an always-on box on your home network that holds your files and runs services). Three of them are good enough for everyday use: Darktable, GIMP, and PhotoPrism’s built-in editor. None of them require an account, none of them upload anything, and all of them run on a UGREEN or Synology NAS with at least 8GB of RAM.
Why “always online” became the default
The reason photo editors went cloud-first is the same reason every other category went cloud-first: it makes the subscription model work. Adobe Photoshop, Affinity Photo, and the new wave of AI-first editors (PhotoWorks, Luminar, the Canva built-in editor) all rely on a server round trip for at least one feature. Photoshop needs Creative Cloud for activation and cloud document sync. Luminar’s AI tools run on Skylum’s servers. PhotoWorks’ background remover hits a remote API for the heavy model inference.
Cloud round trips cost more than money. A 2024 EFF (Electronic Frontier Foundation, a digital privacy nonprofit) report on consumer photo apps found that the average “free” photo editor uploads 17 metadata fields per session, including EXIF (Exchangeable Image File Format, the technical metadata embedded in photos by your camera and phone: location, camera model, timestamp, sometimes your name) location data, device identifiers, and the photo’s perceptual hash (a unique fingerprint derived from the image’s content, used to track photos across services even if the file name is changed). The report did not name the worst offenders, but the pattern is clear: if the app is free and the editing is sophisticated, your photos are the product.
A self-hosted editor on your NAS does none of this. The photos never leave your network. The metadata is yours. The model inference, if you use an AI denoise or background removal, runs on the NAS’s CPU (a hardware accelerator like an Intel Core i5 or AMD Ryzen 5 handles modern models fine) or its GPU if you have one. The monthly bill is zero, the privacy surface is zero, and the data export story is “copy a folder.”
The three that work on a NAS
Darktable is the most powerful of the three, and the one I use most often. It is a non-destructive raw photo editor (it never modifies the original file; all edits are stored as instructions and re-rendered on export) in the same category as Adobe Lightroom, and the feature set is comparable for 90% of what most people do. Color grading, lens correction, local adjustments, batch processing, tethered capture, and a real HDR (high dynamic range) merging workflow are all there. The UI is dense and the learning curve is real, but the documentation is good and the community is active. Darktable runs on the NAS itself, and you can drive it from a browser on any device on the network using the headless mode plus a VNC (Virtual Network Computing, a protocol that lets you view and control a remote desktop) viewer.
GIMP (GNU Image Manipulation Program) is the long-standing free alternative to Photoshop, and it is more capable than its reputation suggests. For pixel-level editing, compositing, retouching, and the kind of work that requires layers and masks, GIMP is the right tool. It does not replace Photoshop for high-end print work or for the specific features that only Photoshop has, but for the 80% of edits most people do (cropping, color correction, object removal, text overlays), GIMP handles it well. GIMP runs natively on Linux and is available as a Docker container for NAS deployments.
PhotoPrism is the third option, and the easiest to start with. It is a self-hosted photo library and management platform that ships with a built-in editor. The editor is not as deep as Darktable or GIMP, but the workflow is the cleanest. Browse your library, click a photo, edit it, and the change is saved back to the file. PhotoPrism handles RAW (unprocessed image data straight from the camera sensor, which preserves more detail for editing than compressed JPEGs) files, it does face recognition and object detection on the NAS, and the search is good. For a home user with ten thousand photos, PhotoPrism is the smoothest experience.
The setup that actually works
The install pattern is the same one I described in the Outline article: Docker Compose stack on a UGREEN or Synology NAS, persistent volumes on a directory you control, and a reverse proxy for HTTPS. The three apps do not conflict; you can run all of them at once if you have the RAM.
For a UGREEN DXP-4800 (a four-bay mid-range NAS with 8GB of RAM) running UGOS, the install takes about an hour for all three. The longest part is the initial catalog scan in PhotoPrism, which can take a few hours on a large library. Darktable and GIMP are instant. The total memory footprint is around 3GB, and the CPU usage is low until you actually edit a photo.
Mobile experience is the first gap. Darktable and GIMP both have mobile-friendly web frontends, but the editing experience on a phone is not as smooth as Lightroom Mobile. PhotoPrism has a clean mobile web app, and it is the best of the three for phone-based browsing. For actual editing, you will want a laptop or a tablet with a stylus.
What I gave up
AI features are the second gap, and the deepest. Adobe’s Generative Fill, Photoshop’s Neural Filters, and Luminar’s sky replacement all run on powerful remote GPUs and the results are genuinely good. The self-hosted AI photo tools are catching up (Stable Diffusion XL with a ControlNet (an auxiliary model that gives the AI additional guidance, like depth maps or edge outlines, so you can control the output more precisely) model can do plausible sky replacement, and Real-ESRGAN (an open source AI model that upscales and restores low-resolution images with surprisingly good results) is a free AI upscaler that runs on the NAS), but the experience is not as polished. You trade convenience for control, and the gap is visible.
Community presets and LUTs are the third gap. Lightroom’s preset marketplace, the cloud library, and the cross-device sync are genuinely useful, and the self-hosted alternatives do not have an equivalent. Darktable’s preset support is good, but the community is smaller. The presets, LUTs (Look-Up Tables, small files that remap colors to give a photo a specific look, like a film emulation), and scripts are not as deep as Adobe’s marketplace.
Collaboration is the fourth gap. If you share your library with family members, the multi-user features in Lightroom are a clear win. Darktable’s library model is single-user, and PhotoPrism’s multi-user support is basic. For a single photographer, this is fine. For a family, you will feel the gap.
What I would tell past me
Three things, in order of how much time they saved me.
- Start with PhotoPrism, not Darktable. PhotoPrism is the easiest to install, the easiest to use, and the easiest to migrate from. Once you are comfortable with the self-hosted workflow, add Darktable for the heavy editing. Trying to learn Darktable and the self-hosted pattern at the same time is too much.
- Pick a metadata standard before you import. EXIF, IPTC (a more general metadata standard used by photo agencies, supporting copyright and licensing info), and XMP (Adobe’s metadata format that Darktable and Lightroom both use to store edit history) are the three you will encounter. Pick one (XMP is the most portable), and make sure your editing tool writes to it. Mixing tools without a shared metadata standard is how you lose edit history.
- Back up the database, not just the photos. Darktable’s library database is a SQLite file. PhotoPrism’s index is a Postgres database. The photos are the easy part; the databases are what get lost. Schedule a weekly export of the database to a second device. The 5 minutes of setup is the difference between a recoverable failure and a lost year of edits.
What to check before you commit
A few things worth verifying on your setup before you start the install:
- Your NAS has at least 8GB of RAM. The combined footprint of Darktable, GIMP, and PhotoPrism is around 3GB, plus the OS and other containers. A 4GB NAS will run one of the three comfortably, but not all three. The UGREEN DXP-2800 with 4GB is enough for a single app; the DXP-4800 with 8GB is enough for all three.
- Your photo library is on the NAS, or you have a clear import path. Editing photos that live on your laptop and only sync to the NAS when you import them is workable, but the workflow is smoother if the library is local to the NAS. Plan a one-time import before you start.
- You have a RAW workflow, or you are willing to start one. Darktable and PhotoPrism both handle RAW files well. If you shoot JPEG only, you can use any tool, and the self-hosted options are overkill. RAW shooters are the audience that gets the most from this setup.
- You are okay with a learning curve. Darktable’s UI is dense, and the documentation is good but not great. GIMP’s UI is dated, and the keyboard shortcuts are non-standard. Plan a weekend of setup and a month of tuning before you decide whether the workflow works for you.
Trade-offs
Self-hosted photo editing is not free in setup time. The install took about an hour for all three apps, with another two hours of tuning over the first month. The AI feature gap is real, and the mobile experience is a step down from Lightroom. The community presets are thinner than Adobe’s marketplace.
In our case, the $240 per year I was paying for Adobe Photography Plan bought a feature set I used about 60% of. Self-hosted covers the 60% I cared about, and the missing 40% is the AI generative features, which I rarely used. Your math will be different if you depend on Generative Fill, the cross-device sync, or the polished mobile app. The closer your workflow is to “edit on a laptop, keep the library local,” the cleaner the win is.
The NAS hardware cost is the same as in the Outline article: about $400 for a two-bay DXP-2800 with no drives, plus $200 for a pair of 4TB NAS drives. If you already have a NAS for the wiki, the photo tools add about 3GB of RAM usage and zero additional hardware cost. The monthly bill stays at zero.
If you are a single photographer who edits on a laptop and is uncomfortable with the cloud upload model, this is a clear win. If you are a family of five with cross-device sync needs, the collaboration gap is real, and you should plan a month of testing before committing. If you depend on Photoshop’s Generative Fill, the AI gap is not closing fast enough, and you should keep the Adobe subscription.