Alternative NIND-Denoise Plugin for Darktable

groegerto · March 12, 2026, 12:46am

After I had some problems with the one-click nind-denoise lua plugin by @rengo (LINK) I decided to work on it a bit myself.

The main idea remains the same: Integrate the process presented by @sillyxone in this thread into a darktable plugin. The main change lies in the way the different stages are handled. I introduced a new option called “Only denoise + import & stack image”, which (as the name suggests) doesn’t process the second stage immediately, but instead imports the denoised image with the history of all operations of the second stage back into darktable.
This allows for the following workflow:

Edit the original image in darktable
Export the image through the nind denoise plugin
The denoised image is now imported into darktable and grouped with the original image
Continue working on the denoised image with all operations from the previous edit copied over
When finished, export the image just like one would a regular one

Do note that this workflow does not include the RL algorithm.

However, if the previous process is desired, it is still supported by not selecting the mentioned option. Additionally, some more work is required to get the plugin to work, as the venv has to be setup manually (in comparison to the plugin by rengo).

Repository

Qor · March 12, 2026, 12:54am

github.com/darktable-org/darktable

[AI] AI inference subsystem with ONNX Runtime backend

master ← andriiryzhkov:object_ai_mask

offen 11:08AM - 11 Feb 26 UTC

andriiryzhkov

+6802 -1

This PR introduces an AI subsystem into darktable with two features built on top… of it: 1. **AI Object Mask** — a new mask tool that lets users select objects in the image by clicking on them. It uses the [Light HQ-SAM](https://github.com/SysCV/SAM-HQ) model to segment objects, then automatically vectorizes the result into path masks (using `ras2vect`) that integrate with darktable's existing mask system. 2. **AI Denoise** — a denoising module powered by the [NAFNet](https://github.com/megvii-research/NAFNet) model. This was initially developed as a simpler test case for the AI subsystem and is included here as a bonus feature. Both models are converted to ONNX format for inference. Conversion scripts live in a separate repository: [https://github.com/andriiryzhkov/darktable-ai](https://github.com/andriiryzhkov/darktable-ai). Models are not bundled with darktable — they are downloaded from GitHub Releases after the app is installed, with SHA256 verification. A new dependency on `libarchive` is added to handle extracting the downloaded model archives. ## AI subsystem design The AI subsystem is currently built on top of ONNX Runtime, though the backend is abstracted to allow adding other inference engines in the future. ONNX Runtime is used from pre-built packages distributed on GitHub. On Windows, ONNX Runtime is built with MSVC, so using pre-built binaries is the natural approach for us — I initially expected this to be a problem, but discovered this is common practice among other open-source projects and works well. The system is organized in three layers: 1. **Backend** (`src/ai/`): Wraps ONNX Runtime C API behind opaque handles. Handles session creation, tensor I/O, float16 conversion, and hardware acceleration provider selection (CoreML, CUDA, ROCm, DirectML). Providers are enabled via runtime dynamic symbol lookup rather than compile-time linking, so there are no build dependencies on vendor-specific libraries. A separate `segmentation.c` implements the SAM two-stage encoder/decoder pipeline with embedding caching and iterative mask refinement. 2. **Model management** (`src/common/ai_models.c`): Registry that tracks available models, their download status, and user preferences. Downloads model packages from GitHub Releases with SHA256 verification, path traversal protection, and version-aware tag matching. Uses libarchive for safe extraction with symlink and dotdot protections. Thread-safe — all public getters return struct copies, not pointers into the registry. 3. **UI and modules**: The object mask tool (`src/develop/masks/object.c`) runs SAM encoding in a background thread to keep the UI responsive. The user sees a "working..." overlay during encoding, then clicks to place foreground/background prompts. Right-click finalizes by vectorizing the raster mask into Bézier path forms. AI denoise module (`src/libs/denoise_ai.c`) and preferences tab (`src/gui/preferences_ai.c`) provide the remaining user-facing features. Fixes: #12295, #19078, #19310

groegerto · March 12, 2026, 1:13am

Yes, I’ve seen that and I’m looking forward to using this once it’s fully integrated into darktable. But from what I’ve seen this is not in a state that it can be easily used right now.

nilangiriyar · March 13, 2026, 5:55am

Apologies in advance to discuss about another product.
The RapidRaw is working towards getting this NIND denoise included. I did the initial code changes. It seems to have picked up some speed.

PS: If this post is not acceptable, please delete. Thanks

martinus · March 24, 2026, 5:27am

From what I have seen in rapid raw is that for most AI related tasks, a connection with comfyui is required or do everything on the CPU.

Personally, I don’t find this a very elegant solution. And it a very different approach from darktable.

I am not sure if that is the case for denoising as well, but if so, then it is not very applicable to dt.

anry · March 24, 2026, 9:26am

AI denoise enabled by NIND model as long as AI upscale in darktable without any plugins and external components is on the way

github.com/darktable-org/darktable

[AI] Add neural restore lighttable module for AI denoise and upscale

master ← andriiryzhkov:neural-restore

opened 08:10PM - 14 Mar 26 UTC

andriiryzhkov

+2887 -0

This is a third and improved part of the original PR #20322. ## Summary - …Add a new lighttable/darkroom utility module (`neural_restore`) that provides AI-based image restoration using ONNX backend models - Supports two operations via a tabbed notebook UI: **denoise** (e.g. NIND UNet) and **upscale** (e.g. BSRGAN at 2x/4x) - Includes an interactive split before/after preview with draggable divider, on-demand generation (click to start), and a detail recovery slider that uses wavelet (DWT) decomposition to recover fine texture lost during denoising - Batch processing runs as a background job with progress reporting, cancellation support, tiled inference with overlap to avoid seam artifacts, and memory-aware tile size selection - Output is written as 32-bit float TIFF and automatically imported into the library, grouped with the source image - Register denoise-nind and upscale-bsrgan models in the AI model registry (`ai_models.json`) ## Details - Built only when `USE_AI=ON` (cmake option, default OFF) - Preview thread uses atomic sequence counter for cancellation and is joined before cleanup to avoid use-after-free - Pixel pipeline: linear Rec.709 → sRGB before inference, sRGB → linear after; planar NCHW layout for ONNX models - Detail recovery: extracts luminance residual (original − denoised), applies per-band wavelet thresholding to separate noise from texture, blends filtered detail back at user-controlled strength <img width="267" height="340" alt="denoise" src="https://github.com/user-attachments/assets/454ada9c-2b94-428f-a2f0-fe8db36b8a06" /> <img width="267" height="340" alt="upscale" src="https://github.com/user-attachments/assets/167afa0d-d408-42c1-a845-713a14664795" /> Fixes: #19310

dtrtuser · March 26, 2026, 7:51am

The noise reduction works great and the upscaling is promising. Thank you Andrii!

helgeschneider · March 26, 2026, 11:15am

Hi Andrii … not sure if this might be a stupid question , LOL, but is the " neural restore " light table window / module already available in the nightly builds ?
At least i cannot see it … on latest build !

Andreas

anry · March 26, 2026, 12:10pm

Not a stupid question. It is not yet merged into a master branch, so not in nightly builds yet.

helgeschneider · March 26, 2026, 2:45pm

Ok thx for your reply

helgeschneider · April 1, 2026, 5:00pm

Cool it´s coming …
Hope your hard work pays off

anry · April 2, 2026, 9:44am

AI denoise and upscale are in master and recent nightly build. Check “neural restore” module, which is available in both lighttable and darkroom.

Masterpiga · April 2, 2026, 10:55am

Brilliantly executed, Andrii and dev team. Models downloaded and modules tested, not a single hiccup or point of friction. Excellent

Masterpiga · April 2, 2026, 12:23pm

@anry I only see the “neural restore” module in the lighttable.

anry · April 2, 2026, 12:25pm

In darkroom it is on the left panel.

Masterpiga · April 2, 2026, 12:28pm

Right, I should remember to unhide it every now and then Thanks!

dtrtuser · April 2, 2026, 4:43pm

It works brilliantly, thank yo Andrii and team! Question: is there any way to see a 1:1 preview? The previous image is very small.

anry · April 2, 2026, 5:49pm

That would be full processing you got when you click “process button”. The idea of preview is to provide you with small and fast way to look into expected result. That’s why it is only small crop of original image.

dtrtuser · April 2, 2026, 8:42pm

Sorry if I wasn’t clear: a small crop is indeed what I would like to have, but at 100% magnification, not downsized.

Same as what, for instance, DXO does: small crop of the full image processed in the preview windows at 100% magnification (what I call 1:1).

nilangiriyar · April 3, 2026, 4:25am

I agree. The preview generated should be default to 100%.
And if the preview is shown as an inlay/PIP in the main window, it would be fantastic.