What photos make a good likeness (and what to skip)
Likenor asks you to choose a folder of photographs of one person. It examines each file on this computer. Nothing from that folder is uploaded. The app can tell a sharp, readable face from a blur, a crowd, or a near-copy. It does not publish a required photo count or a quality score, and this article will not invent either.
What the folder is for
The folder is the training set. The likeness can only learn the face that is actually in those files. A set of one person, seen clearly, in more than one moment, gives the trainer something to work with. A set of the same screenshot, saved six times, does not become more informative by being longer.
The product has not published an official list of how many pictures you must supply. Useful advice is still possible without that number. Prefer a set you can stand in front of and recognise as the same person, not a dump of every image on the disk.
Photographs that help
Use ordinary photographs, not advertisements for photography. Daylight or plain indoor light is enough. The face should be the reason the picture exists.
- One person in the frame, with a face that is easy to read.
- The original file from the camera or phone, not a screenshot of that file and not a tiny export from a chat app.
- A mix of closer shots and a few wider ones, so the face is not only one crop.
- More than one angle. A set that only ever faces the same way teaches that one view.
- More than one expression. A set that only ever smiles the same smile bakes that look into the likeness.
That last pair matches what the app already says in its own quiet voice when a kept set is almost uniform: add other angles, and add variety. It is guidance about the set you already have, not a score out of 100.
Clean files help. JPG and PNG originals are the formats the product expects to read. A picture with no caption, no watermark, and no sticker over the cheek is easier to use than the same face under a date stamp.
What to skip
Skip the files that do not actually show the person you mean to train.
- Blur. Soft focus cannot be sharpened back into a real eyelash. The app's own note is that blur cannot be recovered.
- Group photographs where a second face is in the frame. Crop to the one subject, or leave the file out.
- Pictures where the face is a small patch in a crowd, a poster, or a reflection.
- Heavy sunglasses, masks, or hands that hide the face you want the model to learn.
- Near-duplicates: burst shots that are the same pose a fraction of a second apart. The app will set those aside as nearly the same shot.
- Resized copies, screenshots, and chat exports when you still have the original.
- Heavy filters that rewrite skin and eye shape. The trainer will learn the filter.
A photograph with no readable face cannot contribute expression or framing. If the app cannot find a face, that file has nothing to place. That is a fact about the picture, not a judgement of the person.
How the app uses the folder
You point Likenor at the folder on this PC. The examination runs here. Training photos are not uploaded. See Privacy. If a file is unreadable, replace it with a valid image rather than trying to repair a broken container.
The app may keep some photographs and set others aside. Typical reasons already written in the product are: no face was measured, the file does not clear the minimum gates, it is nearly the same shot as a stronger one, or that framing already had enough better photographs. Those are local decisions. They are not a public league table.
Do not send the folder to support@likenor.com. Support will ask for a specific log if one is needed, not a photo set.
Before you start a run
Photography does not remove the hardware bar. You still need Windows or Linux and an NVIDIA GPU. There is no AMD or Intel path. VRAM needs are still being confirmed. After install, a model of about 60 GB downloads, the owner finishes setup in the app, and training runs later from the folder. The trained likeness can be exported and reused in other tools, including ComfyUI. That sequence is in Train a personal AI likeness on your PC: photos never leave the machine.
Neither public installer is published yet. The Windows file will be a signed Likenor-Setup.exe. The Linux file will be an unsigned Likenor.AppImage with no public URL today. This article does not link a setup executable. When you want the current download state, go to Download. When you want the buying facts, go to Pricing. Price is TBD. Checkout is not open. No photograph you read about here needs to leave the machine in order to decide.
If the other option on your list is a site that trains from an upload, read Local face model vs uploading photos to the cloud before you send the same folder anywhere else.