Kovert, home

Rename photos by what is in them

A camera calls a photo DSC_4801.jpg, which tells you nothing and tells Google less. Kovert reads each image with a vision model running on your own machine and renames it for what it shows. Nothing is uploaded, and there are no credits to buy.

Download for WindowsDownload for macOS

v0.1.0 · Windows 10 and later · macOS on Apple Silicon · free for personal use

The Kovert drop zone, with quick actions for common jobs underneath it.

Folder in, real names out

  1. Drop the photos in

    A folder or a zip, as many images as the disk holds. Mixed batches are fine, videos in the same drop keep normal pattern names.

  2. Pick a prompt style

    SEO is the default, with photoshoot, carspotter and product next to it. The prompt is plain text and fully editable, and a good custom prompt can be saved and shared as a key.

  3. Press convert

    Each image is downscaled to 640 px, sent to your model, and named from the answer, lowercase and hyphenated. If a call fails, that file keeps a normal name and the run carries on.

What the names look like

DSC_4801.jpg becomes golden-retriever-puppy-park.webp. IMG_0231.heic becomes red-brick-townhouse-at-dusk.jpg. The model describes what it sees, Kovert turns that into a filename.

The AI name is one token in a naming pattern, next to index, date and EXIF tokens like camera, lens, ISO, shutter, aperture and focal length. A product shoot can come out as 2026-03-white-ceramic-mug-handle-left-014.webp without a keyboard being touched.

For a shop with 300 product photos, that is the difference between a media library search engines can read and one they cannot.

You bring the model

Kovert ships no model and calls no cloud. It talks to a vision model you run yourself in Ollama, on the same machine or on a server you own, and asks only for its address.

Setup is on you and takes a few minutes: install Ollama, pull a vision model such as LLaVA, Moondream, Qwen-VL or Gemma, and point Kovert at it. Because every image is downscaled to 640 px before it is sent, even modest hardware gets through a folder quickly.

For client shoots this is the whole point. The images go to your own hardware and nowhere else.

No per-image pricing

Web renamers price by the image, monthly credits with caps around 100 photos on the small plans. Kovert has no per-image anything.

A 3,000 photo archive costs the same to rename as ten photos, which is nothing.

Questions

Which models work?
Any vision model Ollama serves. LLaVA, Moondream, Qwen-VL and Gemma all work, and small models stay quick because Kovert sends a 640 px version of each image.
Do my photos leave my machine?
No. They go to the Ollama address you configured, your own PC or your own server, and nowhere else. The renamer has no cloud side.
What happens when the model fails on an image?
That file gets a normal pattern name and the run keeps going. No stops, no half-finished batches.
Does it rename videos too?
No, the vision step is images only. Videos in the same batch are named by the pattern, with tokens like index and date.
Can I change how the names read?
Yes, the prompt is an ordinary text field. Edit it, or start from one of the four styles, and save the result as a key you can share with a colleague.
Is it free?
At home, yes, all of it, no account. Work use needs a business licence, €49 a seat, one time. The models themselves are free downloads through Ollama.

Related

Point it at a folder and see what comes back.

v0.1.0 · Windows 10 and later · macOS on Apple Silicon

Download for WindowsDownload for macOS