Useful evidence, with its limits visible

Private file tools and public archive search are separate

The private similarity tool, favicon generator and SVG inspector process the selected artwork in your browser. Files are not uploaded, saved to an account or added to the public archive. Normal page scripts and fonts still load over the network.

Archive matching is a different workflow: choosing file-based matching sends the query to the configured service. Available results depend on the active index and its recorded method. A metadata fallback is not a visual match.

How archive look-alikes are selected

The visual service compares the whole artwork and, when it can separate one reliably, a left-hand emblem hypothesis. DINOv2 and SigLIP2 features are considered alongside foreground shape and local geometric correspondence. Supporting signals must refer to the same pair of artwork views. Long wordmarks need stronger shape agreement because unrelated lettering can share a horizontal layout.

Identify mode looks for the depicted brand. Look-alike mode compares other brands and can return no strict results when evidence is weak. Exploratory results relax that filter and need closer inspection. Ranks are retrieval order, not match probabilities; the current rules have not been calibrated on an independent, human-labelled cross-brand benchmark. The service does not retain uploaded query images.

Editorial comparison pages are separate from automatic rankings. They identify the archived versions being discussed and explain limited resemblance; a famous historical comparison may no longer describe either brand’s newest identity.

What private comparison measures

SHA-256 identifies identical source bytes. A normalized preview hash identifies equal rendered samples. Two 64-bit grayscale difference hashes use light and dark backgrounds; an alpha-weighted 64-bin colour histogram and aspect ratio supply additional signals. Non-identical files are ordered with fixed weights: 70% hash distance, 20% colour distance and 10% aspect distance.

Those weights are heuristics, not confidence percentages. Rotation, cropping, thin strokes, compression and similar silhouettes can produce misleading neighbours. This tool neither uses a trademark register nor estimates legal risk. Inspect the original artwork and the relevant context before making a decision.

Next step: a formal trademark search

Neither archive matching nor private comparison searches a trademark register, and neither can tell you whether a mark is registered, available or likely to conflict. Before adopting, filing or challenging a mark, search the official registers for the countries and classes that matter, and take legal advice where the decision carries risk.

• WIPO Global Brand Database: international registrations and many national registers, with an image search.
• USPTO trademark search: United States.
• TMview: European Union and participating national offices.
• Department of Patents, Designs and Trademarks (DPDT): Bangladesh.

Colours are sampled approximations

Most colour swatches were sampled from the archived artwork during import and rounded to a coarse palette of 16 levels per channel, so they can differ noticeably from the values in a brand’s own guidelines. The importer now stores the average of the sampled pixels instead of the rounded value for new imports; existing swatches have not been re-sampled, and both are still samples. Pixels whose red, green and blue values are all above 235 are skipped as background, so pure-white parts of a mark are not sampled, and the most common remaining colour (often black lettering) becomes the primary swatch. Very light greys just below that cut-off are kept; older records rounded them to #EEEEEE, which can appear as a white-looking swatch. Records imported from the earlier archive also stored #FFFFFF as the second colour whenever no second colour was measured: that value is a placeholder, not a sample. The CMYK figures on the colours page are simple arithmetic conversions of the sampled value, not print specifications, and no Pantone match is given. Use the brand’s published guidelines for production colour.

Source review is separate from availability

A file’s presence, an import timestamp or an account claim is not an editorial review. Public source-review labels require a recorded source and a qualifying reviewer record. Publisher originals and generated renditions are labelled separately. Original formats are prioritized as AI, EPS, PDF, SVG, then raster formats when the actual files exist.

In a file list, “original” means the exact file the archive registered, kept unchanged. It does not mean the brand owner’s own master artwork. Many Adobe Illustrator (.ai) files came from SimilarLogo’s earlier archive, imported from the owner’s WordPress collection. No publisher or press-kit source is recorded for them, and who drew them is not recorded, so file lists label them “SimilarLogo archive file” and they are not confirmed as official artwork. Some may be redraws rather than the owner’s artwork and can differ from it in detail. Files collected from a publisher’s own press or media page are labelled with that source on the brand page.

Approved community previews require recorded staff display approval. Download approval is separate. Display withdrawal removes public preview eligibility as well as its image-sitemap entry. A permitted preview does not imply permission to redistribute the original.

Entity and history evidence can be incomplete

A foundation’s project listing establishes that listing, not the legal owner or country of incorporation. Country associations, company relationships and project affiliations should be read with their citations. Unverified archive year labels are not adoption dates. Missing history remains unknown rather than being reconstructed from filenames. New research separates logo changes, company events, product launches and broader brand-system updates. Brand histories are compiled with AI assistance from cited sources and rewritten in the archive’s own words. Each researched brand page lists its sources publicly, with numbered references on each milestone, and says when the research was compiled. An editor has not yet independently checked this research; the page says so until one does. Required artwork attribution and licence notices remain public. A researched milestone does not imply that its historical artwork is available to download.

Technical checks do not grant a licence

Decoding, sanitization and checksums help validate file handling. They do not certify artwork identity, malware safety, originality, endorsement or legal permission. Read the source conditions and usage notice, and use the correction workflow when the evidence is wrong.

Saved collections, legacy compatibility and reports

Private comparison can open folders and bounded ZIP archives. Images stay in memory. Saving a collection is optional and persists only filenames, dimensions and fingerprints in the current browser profile, with a maximum of 200 entries and no automatic eviction. JSON collection backups can be restored locally; JSON, CSV and print reports include potentially sensitive filenames and should be reviewed before sharing.

Existing v1 collections are read without rewriting or deleting them. Their average hash, difference hash, edge histogram, colour histogram and alpha coverage are compared separately using the original pixel-sampling recipe. Legacy comparisons use 70% shape and 30% colour weighting. They lack exact-file hashes and image bytes and are never described as exact duplicates or automatic merge instructions. V1 and v2 distances are not comparable.