ShotSieve is an AI photo culling software I built because I could not find a free tool that uses learned models to spot low quality photos in large collections. When you have a large collection, it becomes painfully slow to review every shot manually. By using image-quality models such as TOPIQ, this AI photo culling software helps me surface blurry, out-of-focus, underexposed, overexposed, and otherwise weak photos much faster.
The tool is intentionally simple. You point it at a folder, let it analyze the images, and then make the final keep/reject decisions yourself in a visual desktop workflow. In other words, the models help with the first pass, but the final decision stays in your hands. If you are looking for a local-first photo culling app, or trying to fit AI-assisted culling before your editing workflow, this post explains what ShotSieve does today, where it fits, and where its current limits are.
Why this AI photo culling software exists
Most photo culling tools fall into one of two camps: fully manual review, which is painfully slow on large collections, or heavy automation, where the software tries to make too many decisions for you. I wanted something in the middle instead: a tool that speeds up the boring part of the job without but still allows me to make the final decision.
In practice, the workflow I wanted was simple:
- Scan a large folder.
- Run an AI-assisted analysis.
- Narrow the set to likely keepers and likely rejects.
- Delete or move the rejects after review.
That approach saves time immediately. However, it still leaves the final review to the photographer, which is exactly how I think it should work.
What this AI photo culling software does
ShotSieve is a GUI-first desktop app for first-pass photo culling. Right now, the workflow supports:
- Folder scanning with incremental rescans and missing-file cleanup.
- A local SQLite cache for scan, scoring, comparison, and review state.
- JPEG preview generation for faster browser-based review.
- Learned image-quality scoring through a curated in-app model set.
- A local review UI organized around Library, Compare, Review, and Settings.
- Keep/reject actions, filtering, batch operations, fullscreen preview, and export or move flows.
- Portable runtime-pack launchers for Windows, Linux, and macOS.
The goal is intentionally narrow: help sort high-volume folders faster while keeping the final decision in your hands.
How the AI photo culling workflow works
A typical first-pass workflow looks like this:
- Launch ShotSieve.
- Choose a photo folder in Library.
- Run Analyze to scan and score supported images.
- Use Compare if you want to test model behavior on the same library.
- Move into Review to mark keepers, rejects, and selections.
- Move or delete the rejected files and clean up your collection.
As a result, ShotSieve works well as a front-end culling pass before Lightroom or any other editing and catalog workflow.
AI photo culling software models and runtime options
The in-app model catalog is currently centered on three learned image-quality assessment options:
| Model | Practical role |
|---|---|
topiq_nr | Default model and main general-purpose scoring option. |
clipiqa | Fast secondary option for quick comparison passes. |
qalign | Heavier model available only on supported accelerator-backed paths. |
If you are choosing a download, the simple rule is:
- Choose NVIDIA/CUDA if you have a supported NVIDIA GPU.
- Choose MPS on Apple Silicon Macs.
- Choose DirectML on Windows when you have a modern AMD or Intel GPU and CUDA is not available.
- Choose CPU when you want the safest fallback or are unsure, although performance will be lower overall.
One important compatibility detail is that qalign is intentionally unavailable on CPU and DirectML paths in the current app. On those runtimes, topiq_nr and clipiqa are the practical choices. In addition, qalign was far too slow in my testing outside supported accelerator paths.
From my own use, I usually run TOPIQ first, clean up the obvious low quality photos, and then do another pass with the other models. Because I am on NVIDIA hardware, I can also run Q-Align. That sequence has given me the best results so far, since TOPIQ occasionally misses a weak photo that another model catches.
Limitations
Any credible photo culling AI review should be clear about what AI can and cannot do.
ShotSieve can rank images by learned quality signals, but it cannot understand the full context of a photo. For example, a bokeh-heavy image may receive a lower score because the background is blurred, even when the shot is artistically correct. Likewise, if you shoot bracketed HDR exposures, the intentionally overexposed and underexposed frames may also rank as low quality. Therefore, human review is still necessary.
- Learned IQA scoring depends on subject matter, shooting style, and model behavior.
- Final human review is still necessary for story, emotion, and client-specific choices.
- Model weights are not bundled and may download on first use.
- Third-party AI model licenses may limit commercial use.
- Runtime support is still evolving as the project matures.
How to try ShotSieve
- GitHub repository: https://github.com/cfelicio/ShotSieve
- Quick start and package selection: https://github.com/cfelicio/ShotSieve#readme
- Build and developer guide: https://github.com/cfelicio/ShotSieve/blob/main/docs/building.md
For source installs, launch the app with shotsieve-desktop. For downloaded bundles, open the ShotSieve-* launcher that matches your platform and runtime target.
FAQ
Is ShotSieve free?
Yes. ShotSieve is a free and open-source project. The application is licensed under AGPLv3 or later. However, learned IQA libraries and model weights can have separate non-commercial or research-use restrictions, so review third-party licenses before commercial deployment.
Is ShotSieve cloud-based?
No. ShotSieve is local and self-contained. You scan, score, review, and manage the culling workflow on your own machine through a local desktop and review UI.
Does ShotSieve replace Lightroom?
No. ShotSieve works better as a first-pass culling companion. Use it to reduce a large folder to stronger candidates, and then continue editing and delivery in Lightroom or your preferred editor.
Does AI photo culling replace human review?
No. ShotSieve is intentionally built around AI-assisted narrowing followed by manual review. The models can help prioritize, but the photographer still chooses the final keepers.
Which ShotSieve package should I download?
Use CUDA for supported NVIDIA GPUs, MPS for Apple Silicon Macs, DirectML for Windows machines with non-NVIDIA GPUs, and CPU when you want maximum compatibility.
Final take
ShotSieve is for people who need help with volume and cannot afford to manually review an entire collection looking for weak photos.
If you want AI photo culling software that scans folders, scores images locally, lets you compare models, and keeps the final keep/reject pass in your hands, ShotSieve is built for that job.
Try it out, compare the first-pass review time against your usual culling workflow, and decide whether it earns a permanent spot before your editing app. If you run into issues or have feedback, feel free to contact me.