Why generated images look plastic
A camera records a scene through a lens, a sensor and a compression pipeline, and every one of those stages leaves a fingerprint. Photons arrive unevenly, so a sensor adds noise that is strongest in the shadows and almost invisible in bright areas. Film and digital sensors both add grain that varies from frame to frame. Lenses soften fine detail at the edges. File formats then quantise all of it in slightly different ways each time you save.
Image generators work differently. They predict pixels, so they tend to produce surfaces that are statistically too tidy: shadows with no noise at all, skin and sky that stay smooth where a sensor would have produced texture, and detail that repeats instead of degrading. That uniformity is what people describe as the plastic look — the image is clean, but it is clean in a way cameras never are.
What this humanizer actually does
Three things, in order. First, identifying metadata is removed from the file: EXIF blocks, GPS coordinates, XMP packets and editor records such as Photoshop segments. Second, grain and sensor noise are added back, weighted toward the shadows the way a real sensor behaves, at an intensity you control. Third, the image is re-compressed, which removes some of the over-smooth residue the generator left behind and gives the file the texture of a normal camera export.
Your pixel dimensions are preserved, and the output format follows the input — JPEG stays JPEG and PNG stays PNG. Setting the noise slider to zero still strips metadata completely, so you can use it purely as a metadata cleaner. Everything runs on the Canvas API inside the tab, and the download is assembled locally from the bytes already in memory.
What it is not for
This is a texture and metadata tool, not a way to make a generated image pass as a documentary photograph. Do not use it to present synthetic images as evidence, news photography or proof of anything that did not happen — see our terms of use for the uses we prohibit. We remove metadata rather than invent it: the tool never writes fake camera models, fake timestamps or fake GPS coordinates into a file. Forging provenance data would be a different product, and we are not building it.
We also make no promise about what any detector will report afterwards. Detectors weigh frequency patterns, watermark records, provenance boxes and metadata, and they are retrained constantly. What we can say is narrower and honest: the file loses its identifying metadata, and its texture stops looking synthetic. Use it on images you have the right to edit, and treat it as part of a workflow rather than a switch.
Use it responsibly. Editing an image is fine; misrepresenting where it came from is not. If you publish an edited or generated image, say so — a caption does more for your credibility than any file-level change.
Your images stay on your device
There is no upload step, no queue and no server holding your file. The image is read from your disk into browser memory, processed there, and written back out as a download. We cannot see your photos, we do not store them, and we do not use them to train anything. That is also why the tool keeps working when you go offline, and why you can process client work or unreleased material without asking anyone's permission.