thesnarkitecht

Speaking as someone who has High-Memory Mac Studios, it is great that there is finally a model that will run at usable speeds with high quality. In my experience, the sweet spot for Mac Clusters is around 200GB of weights, which this is far under. Congratulations to the recursive team for building an open model that will actually run!

I built this project because I did not like paying for Lightroom. Would love input into how to improve.

A cluster of 4 M3 Ultra Mac Studios with 512 GB of unified memory each. Don't recommend. EXO labs is dead and trying to shard models is really hard. On their own, each Mac can generate text at usable speeds, but I assume that you trade memory for speed. I have never used an NVIDIA GPU, so I guess that you trade model capacity for speed. I would like to know what y'all Nvidia people think

Looking into it.

Automatic blemish detection is now included in the Remove menu. I don't think it is perfect yet, only tried it on a couple of photos, but, feel free to open an issue if the performance is not satisfactory, as with the import of Lightroom presets. It would be under Presets/Import Lightroom presets. Thank you!

Will definitely add those ASAP. Thanks for your support!

Most everyday Lightroom editing and library work is there. Missing: print, book, map and web modules, face recognition and tethering. There’s a full list in the README under “How it compares”.

Fair. Your originals are never modified and edits are standard XMP sidecars that Lightroom and darktable can read, so if this project ever stops, nothing is locked in.

No worries, thanks for checking. If anyone wants to verify a download: hashes are in SHA256SUMS on the release, and the installers now have GitHub build attestations (gh attestation verify <file> --repo thesnarkitecht/rembrandt).

Fair question. The scene-referred design is modelled on darktable’s, and darktable deserves the credit for it. My very first prototype ported some of darktable’s algorithms, and I replaced that with an independent implementation before releasing. Both projects are GPL-3.0, and the README now credits darktable directly.

I built Rembrandt because darktable’s learning curve kept me from using it, not because it’s worse; it has far more control. The goal here is approachability: fewer, broader controls, on-device AI for masks and denoise, and Lightroom catalog import.

Stack: the engine is JavaScript with WebGL/WebGPU shaders, in a small Rust (Tauri) desktop shell. LibRaw (C++) decodes RAW files.

Edit: corrected my earlier reply. I wrongly said it was written in C++, and I overstated the darktable answer.

I will admit that all in all darktable has a more comprehensive set of controls over editing. However, it is harder software to use.

Have y'all tried an Ollama Cloud subscription? Their off-hours pricing for V4.1 Flash is extremely competitive.

Apple silicon as well.