Freepik will not read a title or a keyword list that sits outside your file. Both fields have to be integrated into the image metadata itself, or supplied in a CSV that travels with the batch. Almost every other rule in the contributor documentation follows from that one requirement, and almost every blank-field rejection traces back to ignoring it.
Freepik, Magnific, and What Actually Changed
Follow the contributor documentation to its current home and you land on a page branded Magnific rather than Freepik. The company folded its AI products under the Magnific name, and the submission requirements moved with them. The contributor rules themselves did not change in the process, which is the part worth holding on to: the field reference below is the same one Freepik contributors were working from before the rebrand, served from a different domain.
Practically, that means two things. Search engines and most contributors still call it Freepik, so you will keep seeing the old name in tutorials and forum threads. And your own workflow does not need rewriting because a logo changed. What does need attention is whether your metadata actually satisfies the requirements, which is a separate question from whether you can find them.
Photo Metadata Requirements: The Field Reference
Photographic submissions have to clear a technical gate before a reviewer considers the content at all. The published requirements are specific:
- File format: JPEG. Not TIFF, not PNG, not raw.
- Colour mode: RGB.
- Colour profile: sRGB, Adobe RGB, ProPhoto RGB or P3. The profile list is wider than most agencies allow, but it is still a list, and an untagged file is not on it.
- File size: over 0.5 MB.
- Dimensions: between 2,000 and 10,000 pixels on any of its sides.
- Title and keywords: integrated in the metadata.
The dimension rule is the one that catches people out, because it has a ceiling as well as a floor. Contributors who upscale aggressively, or who shoot high-resolution and export at native size, can push a long edge past 10,000 pixels and get bounced for being too large. Downsizing fixes it, but only if you notice before you send a thousand files.
Vector Metadata Requirements
Vectors are the other accepted content type, and they carry their own spec sheet alongside the same embedded-metadata rule:
- File format: EPS, saved in CC.
- Colour mode: RGB.
- File size: between 0.5 MB and 80 MB.
- Title and keywords: integrated in the metadata, exactly as with photos.
Vectors also carry a usability requirement that has no photo equivalent. Layers must be arranged properly and named correctly in English, fonts must be free for commercial use and remain editable where editability matters, and packs or element groups have to be presented tidily. None of that lives in the metadata fields, but reviewers treat it as part of the same submission quality judgement, and a messy layer stack undoes an otherwise clean file.
The Title Field: Written for a Buyer, Not for You
The title is the sentence a buyer reads under the thumbnail and the string the on-site search engine weighs most heavily. It has to be in English, it has to describe what is actually in the frame, and it should read as a phrase rather than a keyword dump with spaces. A title that says what the subject is doing, where, and in what mood will outperform a title that lists nouns, because the search index has the keyword field for nouns already.
Two habits reliably hurt. Writing titles in a language other than English is explicitly called out as bad metadata, so translating for your home market is not a shortcut to a local audience. And recycling one title across a whole shoot flattens a set that would otherwise rank on distinct queries, which is a slow leak rather than a rejection but costs more over a year.
The Keyword Field: Six Dimensions, Not One
Keyword guidance for the platform is unusually explicit about what a good tag set covers: the subject, the style, the colour, the mood, the orientation and the use case. That is six different axes, and most under-performing keyword sets only work two of them. A photo of a woman at a laptop is not just woman, laptop, office. It is also minimal, warm, horizontal, copy space, remote work, blog header.
Use case is the axis contributors skip most often, and it is the one closest to how buyers actually search. Designers rarely type what the picture is; they type what they need it for. Landing page, presentation background, social ad, newsletter header. Adding those terms does not dilute the set, because the concrete subject terms are still there carrying the literal match.
The documented bad practices are just as useful read in reverse. Irrelevant tags, tags in a language other than English, negative tags that describe what the image is not, and file-format references used as keywords are all named as problems. That last one is worth repeating because it looks harmless: putting eps or psd in the keyword field describes your delivery, not the picture, and a buyer searching for a file extension is not a buyer at all.
None of this is unique to one agency. The same six-axis discipline is what makes a keyword set travel, and it is covered in more depth in our guide to writing metadata for stock photos, which is worth reading alongside this reference if you are building a set from scratch rather than fixing one.
The CSV Route: One File for the Whole Batch
Embedding is not the only accepted delivery. Contributors can supply metadata through a CSV that accompanies the upload instead, which is how most high-volume workflows actually run. The file is semicolon-separated, with commas reserved for splitting the keyword list inside its own cell, and it carries the filename, the title and the keywords for every asset in the batch.
The filename column is the join key, which makes it the single point of failure. A rename between export and upload, a trailing space, or a case mismatch and the row stops matching its image. The safest habit is to generate the CSV from the same pass that writes the files, so the two can never drift apart, rather than assembling it by hand afterwards from a folder listing.
Whichever route you pick, pick one. Mixing embedded metadata on some files and CSV rows on others inside a single batch is where contributors end up with half a delivery carrying titles and half arriving blank, and the blank half is the one that gets reviewed first.
AI-Generated Content: The Prompt and Model Are Metadata Too
If the asset was generated with AI, the metadata set grows. Alongside filename, title and keywords, the CSV carries the prompt used to create the image and the model it came from, with the model written exactly as it appears in the dropdown on the contributor panel rather than however you happen to abbreviate it.
This is the field most likely to be filled in casually and most likely to cause trouble later, because disclosure requirements are one of the areas agencies have been tightening across the board. Treating the prompt as a throwaway note rather than a record you may need to stand behind is a habit worth breaking early. We cover how the major agencies differ on this in metadata for generative AI images.
Rejection Reasons That Sit Next to the Metadata
Clean metadata will not rescue a file that fails on content, and the published rejection reasons are worth knowing because several of them are avoidable at capture rather than at upload. Technical problems are listed first: artefacts, noise, over-sharpening, resizing damage, chromatic aberration, wrong white balance, depth-of-field errors and soft focus. Then taste: heavy vignetting, obvious filters, tilted angles, weak composition. Then sensitive content. Then personal information visible in frame, which includes full names, identification numbers, addresses, phone numbers and licence plates.
Releases are not optional. An identifiable person requires a model release, and identifiable property, including a trademark, a privately owned building, a pet or an object, requires a property release. Submitting without them is listed among the serious mistakes that can end the collaboration agreement, which puts it in a different category from a file that simply gets declined.
Two more entries in that serious column deserve attention because contributors trip them while trying to be efficient. Resubmitting images already sent in, whether or not they were accepted, counts against you. So does rotating, flipping or reframing an image you have already submitted and presenting it as new work, which is explicitly not treated as original. A batch tool that cheerfully re-uploads your archive can walk you into both.
There is also a timing rule that surprises people: content tied to holidays and important dates is not accepted once the date has passed, and a design carrying a year is only accepted while that year is current. Seasonal work has a submission window, and metadata will not widen it.
Keeping Freepik Metadata Consistent With Everywhere Else
Almost nobody submits to one agency. The moment you send the same shoot to several, the metadata question stops being what does this platform want and becomes how do I satisfy several sets of rules from one source of truth. English-only titles are safe everywhere. Six-axis keywords travel well. But keyword count limits, banned terms, mandatory terms and category names differ by agency, and a set tuned for one can quietly break another.
That is a separate decision from whether the platform is worth your files in the first place. If you are still weighing that, our breakdown of Freepik contributor requirements and earnings covers the levels and the payout side, and whether Freepik is safe for contributors deals with the licensing questions that come up most often.
This is the problem Rastock.ai was built around. It writes titles, descriptions and keywords from the image itself, applies each agency's own rules rather than one generic set, and delivers by FTP to 10+ agencies while tracking what landed where. Banned and mandatory keyword handling and batch rollback are part of the feature set, because speed without control is how one bad batch reaches ten agencies at once. There is no revenue share, no claim on your files, and IPTC and CSV export stay free, so you can always walk away with your own metadata. The 14-day free trial does not ask for a card.
Whatever you use to produce it, the requirement itself is simple enough to hold in your head. The title and the keywords have to be inside the file or inside the CSV, in English, describing the picture across subject, style, colour, mood, orientation and use. Everything above is detail hanging off that sentence.
Frequently asked questions
Where do Freepik titles and keywords have to be stored?
They have to be integrated into the file's own metadata, or supplied in a CSV that accompanies the batch. The contributor documentation states this for both photos and vectors. A title typed only into a web form is not the same as an embedded title, and relying on the form is the most common reason a batch arrives with blank fields.
What file specifications do Freepik photos need to meet?
Photos must be JPEG in RGB colour mode, using an sRGB, Adobe RGB, ProPhoto RGB or P3 colour profile. File size must be over 0.5 MB, and the image must measure between 2,000 and 10,000 pixels on any of its sides. Files outside that dimension range are rejected before the metadata is assessed at all.
Is Freepik the same thing as Magnific now?
The company folded its AI products under the Magnific name, and the contributor documentation now lives on the Magnific domain. The submission requirements themselves did not change with the rebrand. Most contributors, and most search traffic, still use the Freepik name, so you will see both in circulation for some time.
What does the Freepik metadata CSV contain?
It is a semicolon-separated file carrying the filename, the title and the keyword list for each asset, with commas separating individual keywords inside their cell. AI-generated content adds two more values: the prompt used to create the image, and the model name written exactly as it appears in the contributor panel dropdown.
Which keyword habits are treated as bad metadata?
Irrelevant tags, tags written in a language other than English, negative tags describing what the image is not, and file-format references such as eps or psd used as keywords are all named as problems. Good sets instead cover six axes: subject, style, colour, mood, orientation and use case.
Can I resubmit an image after cropping or flipping it?
No. Rotating, flipping or reframing an image you have already submitted is explicitly not treated as original work, and resubmitting files sent in previously, whether or not they were accepted, is listed among the serious mistakes that can end the collaboration agreement. Batch tools that re-upload an archive can trigger this without you noticing.