Magnific AI (formerly Freepik AI)

How the platform is organised

The starting point is the required asset: a new image, a library file, an edited composition or an audiovisual sequence. Magnific brings these tasks together, but each tool retains its own inputs, controls and conditions. Selecting an image model does not automatically enable the features of a different video model.

Distinguish four levels: source material, the model transforming it, the workspace organising production and the file eventually delivered. This separation supports comparison without confusing a generation with a finished composition. The offering described here was checked on 8 November 2026. For a team, identifying those levels also clarifies who supplies material, who edits it and who accepts the result.

Reference documentation: Magnific — AI creative suite.

From Freepik to the Magnific name

Freepik began in Málaga in 2010, founded by Alejandro Sánchez, Pablo Sánchez and Joaquín Cuenca. It started as a graphic-resource search engine and developed into content production and distribution. In 2020 EQT agreed to acquire a majority stake; the founders retained an interest.

Expansion brought Videvo, Iconfinder and Original Mockups in 2022, and EyeEm in 2023 following the latter’s insolvency. Meanwhile, the company moved towards AI generation. Javi López and Emilio Nicolás created the Magnific upscaler, launched in late 2023; Freepik acquired it in May 2024.

On 28 April 2026 Freepik adopted the Magnific name. This business continuity explains the coexistence of historical stock documentation, an upscaler with earlier users and a much broader creative suite. These are not three interchangeable names for one function.

Reference documentation: EQT acquisition announcement, 28 May 2020 · From Freepik to Magnific: chronology · Freepik is now Magnific · Videvo acquisition announcement, July 2022 · EyeEm acquisition FAQ, October 2023 · The Decoder: Magnific acquisition, 9 May 2024.

Earlier accounts and stock continuity

Freepik accounts continue on magnific.com. The older magnific.ai service retains legacy access for existing users, with separate subscriptions and tokens; it offers ZIP exports of creations. New registrations go to the current platform. An account on one does not imply a subscription to the other.

EyeEm closed in January 2026. Its migration guide limits transfers to content already distributed on Magnific and contributors accepting the process; a brand redirect does not establish that an entire archive transferred. Before reorganising a library, identify each original’s location and download associated licences. A provider’s name change does not replace the team’s asset inventory or establish which account holds a particular entitlement.

Reference documentation: Freepik is now Magnific · Magnific legacy service guide · EyeEm closure and migration FAQ.

External models, F-Lite and Magnific One

The platform combines other providers’ models with its own developments. In April 2025 Freepik and fal introduced F-Lite, a 10-billion-parameter diffusion model trained, according to the company, on licensed assets. Regular and Textured variants were released. That project does not describe the training of every model available in the suite.

Magnific One introduces a Draft workflow for exploring proposals and Final for higher-resolution output. It accepts references, but moving from draft to final can change composition. Select a model by comparing supported inputs, fidelity to the brief and subsequent work. The workspace provider and model provider may differ; record both when this matters for reproducing the process and interpreting differences between outputs.

Reference documentation: F-Lite launch, 29 April 2025 · Magnific One documentation.

Brand, references and editable elements

Brand Kit can guide Magnific One with colours, typography and visual references in Final mode. The reviewed documentation initially limits it to One; Draft and Brand Kit are unavailable through MCP in that version. Generated wording and logos may need subsequent correction.

For a corporate campaign, separate visual inspiration from elements requiring exact preservation. Palette and style can guide generation; authorised logos and final instructions are better placed as editable elements. A useful review compares several small items, rather than only an attractive cover. This reveals whether hierarchy, contrast and spacing survive changes in message length or publication format. Keep the approved brand assets available independently of the generated background.

Reference documentation: Magnific One documentation · Magnific One product page.

Creative and Precision upscaling

Upscaling distinguishes Creative, associated with Magnific Classic, from Precision, with its own models and controls. Creative adjusts creativity, HDR and resemblance; Precision aims to preserve inputs more closely through different processing options. Available scale depends on the workflow, so one advertised figure should not be applied to every mode.

Enlarging an illustration for a cover differs from enlarging a documentary photograph. New detail may be part of a creative proposal, while in a factual record it changes what can be established from the original. Evaluate a small area containing edges, text and surfaces first. Compare versions at equal size and then at delivery size, ensuring apparent sharpness does not conceal content changes.

Reference documentation: Image upscaler documentation.

Designer, layers and output files

Designer provides a multipage canvas with text, images, vectors and layers. Imports include PSD, PDF and SVG; Auto Layers reconstructs layers from a flat image. Exports include PNG, JPEG, WebP and PDF, with differences between retaining text and vectors and flattening a composition.

Reconstructing layers does not recover the designer’s original file. Inspect masks, fonts, stacking and backgrounds before replacing an existing document. For delivery, retain the editable project and generate the format required by the channel. A reference PDF and a screen image serve different purposes. Test exported pages rather than only the editor view, particularly when layouts contain long passages or different page sizes.

Reference documentation: Designer product documentation · Designer imports and exports.

Video generation and editing

Video generation accepts text, images and references depending on the model. Some models include audio or start and end frames. Video Project Editor assembles clips on a timeline, supports cuts and transitions, and documents a single added audio track accepting MP3 and WAV. Export and credit use depend on configuration.

An explanatory sequence should be designed as checkable shots: initial situation, action and outcome. Generating shots separately makes an action easier to correct without rebuilding the entire piece. During editing, inspect object continuity, orientation and reading time. If the narrative requires simultaneous voices, effects and music, assess whether the available editor is sufficient or whether another program should complete the mix.

Reference documentation: Video generator documentation · Video Project Editor documentation.

Voices, music and sound effects

Audio tools include text-to-speech, voice changing, effects and music. They offer models from different providers and cloning options. In speech synthesis, the script field contains what will be spoken; it should not be confused with a descriptive instruction for generating a scene.

For multilingual narration, prepare complete versions and check acronym, name and number pronunciation in each language. The same wording can need different pauses. Use authorised voices and retain approval when representing an identifiable person. Listen through modest speakers as well: music that works through headphones may obscure words on a reception display. Message comprehension is a delivery criterion independent of how natural the voice sounds.

Reference documentation: Audio tools documentation.

3D objects and scenes

The 3D generator uses models such as Tripo and Trellis. Supported inputs and views vary: some workflows accept several images and others a single image. Documentation describes geometry and texture controls, GLB export and use of the object in 3D Scenes.

A generated object can support composition exploration or illustrate a spatial concept. It is not automatically an engineering model with verified dimensions. For a workstation illustration, first identify the important relationships: relative height, position and access. When these must be precise, begin with technical documentation and check the representation. A decorative resource may tolerate simplifications that would be unacceptable in an explanation of reach, fit or distance.

Reference documentation: 3D generator documentation.

Spaces and collaborative organisation

Spaces organises assets and operations as connected nodes on a canvas. A workflow can combine text, images, generation, editing and upscaling. Collaboration enables work on a shared structure rather than exchanging only finished files.

Its usefulness depends on the diagram being understandable. Name inputs by their purpose, separate exploration from approved versions and retain a branch before changing the workflow. A campaign for several locations could keep one common composition while changing wording or photographs. Before repeating it in a batch, test one location with short wording and another with long wording. That check reveals fitting errors a single sample would miss and makes later handover easier.

Reference documentation: Spaces.

Flows, Agents, API and MCP

Flows packages a predefined sequence for execution through input fields. Agents can plan steps from instructions, tools and knowledge, leaving an editable Space. The API and MCP integrate Magnific capabilities into other systems or assistants. Not every application control is available through every route.

A stable sequence, such as adapting a design to three sizes, fits an explicit workflow. An open brief may need exploration through an agent. In either case, specify the expected file and an acceptance condition. Automating fifty versions should not prevent reviewing one, stopping the process or recovering its original input. Record which changes require renewed editorial approval as part of the production instructions.

Reference documentation: Agents and Flows: when to use each · Magnific API · Magnific MCP.

Metadata and creation lineage

Pro and Enterprise documentation describes CSV metadata export and PDF creation records. These can include model, instructions, references, author and date, alongside a lineage view. A bulk export from a shared space does not automatically download every member’s creations.

This information helps explain how an asset was produced but does not certify the truth of its subject. For a communication library, add an internal identifier, purpose and approval status. Keep metadata with the delivered file and authorised sources. A later review can then locate the asset and understand what changed without relying on somebody remembering a conversation or reconstructing a generation chain. Production history and factual evidence serve different purposes.

Reference documentation: Metadata export and lineage.

Stock, generated content and licensing

Stock and AI conditions are not identical. The Stock guide permits commercial use of free assets with attribution, subject to restrictions; the generated-content guide reserves commercial use for paid plans and describes Free as personal use with attribution. Identify the category before reusing an asset.

Stock restrictions include redistributing standalone assets or incorporating them into trademarks. A subscription also does not guarantee that every generated output receives copyright protection or respects third-party rights. Retain the downloaded asset’s licence and permissions for supplied references. If a project requires exclusivity, do not infer that requirement from the ability to download or generate an image. Record the applicable conditions at selection rather than reconstructing them after publication.

Reference documentation: Stock licenses and attribution · AI content and copyright.

Plans, credits and enterprise conditions

The current guide distinguishes Free, Premium, Plus, Pro, Business and Enterprise; Premium+ was renamed Plus. Consumption depends on tool, model and configuration. Unlimited access advertised for certain images does not imply unlimited video, audio and editing.

Enterprise documents private organisational content and no training on its inputs and outputs. Contractual coverage for intellectual-property claims has conditions and exclusions; it should not automatically be extended to other accounts. To select an arrangement, estimate a complete job: exploration, correction, exports and collaborators. Check who administers the account and what happens when team membership changes. The useful cost is that of an accepted asset, not only its first generation.

Reference documentation: Plans and pricing documentation · Enterprise content and legal.

Audiovisual production and adoption initiatives

Magnific Originals supports audiovisual productions with creators, including the fantasy series The Chronicles of Bone. These works demonstrate narrative uses of the environment and combine direction, generation and editing decisions. They are production examples rather than documentation of every model’s internal operation.

The Magnific Fund campaign in May and June 2026 offered discounts and training to eligible European companies. Its conditions specified that it was neither a grant nor a financing instrument. Historical announcements should distinguish a completed commercial program from a permanent product entitlement. To assess adoption, examine the described production process and required resources as well as the audiovisual result on display. A finished film alone does not reveal the revision effort behind it.

Reference documentation: Magnific Originals · Magnific Fund campaign terms, May2026.

Practical applications in OSH

In occupational safety and health, it can support campaign materials and illustrative resources for training sessions. Combining formats allows a visual concept to be adapted for a guide, screen or video using technical content that has already been reviewed. The different outputs still need to be checked individually for meaning and accuracy.

It can support campaigns needing consistent assets across formats, prepare illustrations or adapt images for a layout. In OSH, distinguish creative resources from records of facts. An illustrative header can be modified; a photograph documenting a facility’s condition needs its original preserved.

It can also help organise a workflow from concept through editing to adaptation. Decide which version is approved and which changes require another review, especially when a still image becomes video. Maintain clear responsibility for the message across those transitions.

Worked example and result checks

Example: prepare an incident reporting campaign with an illustration, a header and a brief animated piece. The responsible team checks that every format conveys the same action and that scenes neither reveal personal information nor portray unsafe conditions as acceptable.

1. Define one action the campaign should explain and prepare approved wording. Identify the destination formats and audiences.
2. Create or select an authorised illustration. Check that details do not imply blame or identify real people.
3. Adapt the asset for the header and audiovisual item. Preserve the same instruction and review what changes between formats.
4. If an image is enlarged, compare the original and result in relevant areas. Reject added details that change its meaning.
5. Retain source material, alternatives and the approved item. Test whether the message is understood in the intended distribution channel.

This is a proposed editorial campaign workflow, not a completed publication or evidence of preventive improvement.

Limitations and conditions of use

AI upscaling may add details that were never present. Do not use enhanced images to interpret damage, read identification plates or substantiate investigation findings. Preserve originals and distinguish illustrations from photographs documenting facts. Also review the usage conditions for any selected materials.

Access to several models does not imply identical conditions. Check permitted inputs, output use, visibility and account options. Do not build a workflow around a feature described in old Freepik material without checking current documentation. Keep the original available so another reviewer can assess the transformation.

Reference documentation: Image upscaler documentation · AI content and copyright.

When to choose it and how to complement it

Magnific fits workflows combining creative assets and evaluating tools within one environment. Adobe Express can support final communication layout. Where editable vector families are required, compare that need with Recraft.

A useful trial goes beyond enhancing an attractive photograph: include an image containing text and fine detail to see what is preserved and what is reinterpreted. Record the model used so differences between outputs can be explained and reviewed.

Frequently asked questions

Is Magnific still only an upscaler?

Not according to its current presentation: the website describes a broader creative suite and the rebranding from Freepik.

Does enlargement recover hidden evidence?

No. Processing may generate plausible detail; it does not establish that the detail existed in the original scene.

What should be kept after transforming an image?

The original, transformed version and purpose of editing, without confusing an illustration with a documentary record.

Reference documentation: Freepik is now Magnific · Image upscaler documentation.