Magnific One launches to give teams an image generator that avoids the 'AI look' and automatically upholds their brand kit

Generative AI image generation has come a long way from the early days of DALL-E and Midjourney in 2021-2022. But even with the most advanced models now supporting generated text and complex infographics, perceptive human beings can still detect and recoil from the "AI look" that many of these generated images exhibit — things like overly detailed textures, synthetic sheen, foods and products that appear far too numerous and voluptuous compared to their real world counterparts.

Here's an example of what I mean:

Now the cloud-based AI image generation aggregator Magnific thinks it can do better by offering a custom, post-trained image generation model shaped to output far more humanistic and naturalistic imagery, guided by its own human graphic designers: Magnific One, launching today, combines an AI-powered creative direction system with a new Brand Kit feature that automatically applies a company's visual guidelines to generated images, checks whether the results follow those guidelines and helps users correct problems before publishing.

With prior image generation models, "it felt like you could have either consistency or imagination, and not both," said Sofía López Martín, Magnific's head of product marketing, in a recent video call interview with VentureBeat. As she continued:

"Our creatives have made this model because they want to have both. I want to be surprised by a model. I want to have this feeling of the texture surprising me. It's giving me something that I didn't expect, but at the same time, if I have a very clear visual direction or art direction, I know this model is also going to perform on that."

The new system is available to paying Magnific subscribers through the company's website, desktop and mobile applications, and Model Context Protocol (MCP) integration, which allows users to access its functionality from AI assistants such as ChatGPT and Claude.

Why it matters for enterprise branding

For enterprise customers, the attraction is straightforward: Rather than requiring every marketer, designer or outside agency to repeatedly explain a company's brand identity through prompts, Magnific aims to make those instructions reusable and enforceable throughout the image creation process.

In an exclusive briefing with VentureBeat, López Martín demonstrated the technology using a Brand Kit she created specifically for VentureBeat, showing how the system could reproduce elements of the publication's visual identity in promotional graphics and identify discrepancies.

The company also disclosed an important technical detail: Although Magnific One is being presented as its own image model, it relies on OpenAI's GPT Image 2 technology, enhanced by Magnific's proprietary creative-direction system.

That distinction illustrates a broader shift in the AI image-generation industry, where companies increasingly compete not just on the underlying models they offer, but on the tools that make those models useful for professional production.

Inside Magnific One: An AI art director between the prompt and the image

According to Magnific, most image-generation systems leave users responsible for describing virtually every aspect of the desired output, from composition and lighting to photographic technique and graphic style.

Magnific One takes a different approach by inserting an additional creative-direction step between the user's original instructions and image generation.

The system interprets the prompt, considers any supplied references and determines an appropriate visual treatment before sending the resulting instructions to the underlying image generator.

For professional users, that means a brief description of a desired advertisement, poster or editorial illustration can become a more elaborate visual composition without requiring the user to specify every design decision manually.

Martin LeBlanc, Magnific's chief experience officer, described the intended difference in the company's launch announcement shared in advance with VentureBeat: "Most image models are very good at making the picture you described. Magnific One makes the picture you would have arrived at with a good art director in the room."

López Martín told VentureBeat that the goal is to combine the unexpected aesthetic results sometimes produced by generative AI with the predictability required by professional designers.

She argued that existing generators can force users to choose between imaginative results that depart from their instructions and faithful interpretations that produce generic-looking images.

"We want to have both," she explained during the briefing.

The company sees this partly as a response to the proliferation of visually similar AI-generated advertising, illustrations and promotional materials.

In the interview, López Martín described how restaurants and other small businesses increasingly use AI-generated posters with recognizable visual characteristics. She said Magnific's creative team wanted a system capable of producing more distinctive results without sacrificing reliability.

Magnific One offers two generation modes.

Draft mode creates eight or sixteen alternative compositions from a single prompt for the credit cost of one image, enabling users to compare styles, moods and visual approaches before selecting a direction.

Final mode produces higher-quality images at 2K or 4K resolution. Users can proceed directly to Final if they already know what they want, or promote a promising Draft composition into a finished image.

The system also accepts up to five reference images to guide the resulting output.

Brand Kit turns corporate design guidelines into reusable AI instructions

The more consequential part of the announcement for enterprise users may be Brand Kit, a companion system that attempts to solve one of the practical difficulties of deploying generative AI across a large organization: maintaining a recognizable visual identity when many people are creating content independently.

Setting up a Brand Kit begins with familiar corporate materials, including a brand guidelines PDF, website, existing images or a written description.

An AI agent analyzes those materials and produces an initial representation of the brand's visual identity, including its logos, typography, color palette, imagery and overall aesthetic.

Users can review that output, correct mistakes, add missing details and save it as a persistent set of instructions.

The resulting kit can be shared across teams and assigned to projects. An agency working with multiple customers can maintain separate kits for different brands.

When employees generate images using Magnific One's Final mode, the selected Brand Kit influences the initial composition and design.

The system subsequently checks each image against its understanding of the brand, looking for mismatches involving visual style, color selection, typography or logo usage. It also conducts broader quality checks for problems such as visual artifacts, framing and unreadable text.

If an image fails a check, the interface identifies the problem and allows users to request a correction.

The distinction matters because merely including brand instructions in a prompt does not guarantee that an image generator will follow them. Magnific is attempting to add a layer of verification and correction, although the company does not claim every initial result will comply perfectly.

Brand Kit is optional, and it currently works only with Magnific One rather than the other image-generation models available through the Magnific platform.

Magnific demonstrated VentureBeat's own brand inside the product

To demonstrate how the system works, López Martín prepared a sample VentureBeat Brand Kit ahead of her interview.

She collected publicly available information about VentureBeat's visual identity and provided it to Magnific's brand-building agent, which generated an initial kit describing the publication's design characteristics.

The interface assembled information about the logo, likely typography, color choices and preferred imagery, alongside examples and visual guidance.

López Martín explained that a company with formal brand guidelines could supply those documents instead, improving the precision of the initial results.

She then selected the VentureBeat Brand Kit while generating a mock promotional graphic for a publication event.

The system applied the stored visual guidance to the image, but the initial result did not satisfy all of its brand checks.

The interface flagged issues involving the use of an accent color and an excessively theatrical visual treatment behind a speaker, which it considered inconsistent with the publication's established identity.

López Martín showed how an employee could select the identified problem and ask Magnific to correct it without manually rewriting the entire generation prompt.

She also demonstrated how the same workflow could support further changes to text and other visual elements.

The example was particularly relevant to organizations in which marketing staff, editors, event teams and external contractors all generate promotional materials while working under the same corporate identity.

A publication could establish a single Brand Kit and make it available to multiple employees, reducing the need for each person to learn complex prompting techniques or maintain their own collection of visual references.

Similarly, a retailer could use separate kits for individual product lines, while an advertising agency could maintain distinct brand instructions for each client.

Magnific also provides Auto Layers and Designer tools for precise changes to typography, colors and copy without requiring users to regenerate the entire image. Additional editing can take place through supported Photoshop and Figma workflows.

The VentureBeat demonstration was conducted by Magnific rather than through an independent hands-on evaluation. It nevertheless provided a concrete example of how the company expects enterprises to incorporate the product into existing creative workflows.

Magnific One uses OpenAI's GPT Image 2 — but what was actually trained?

Perhaps the most revealing part of VentureBeat's briefing came when López Martín was asked to explain the technology underlying Magnific One.

Unlike companies that develop foundation models from scratch, Magnific relies on an existing image-generation system and adds its own creative technology.

"It's powered by GPT-Image-2," López Martín said.

She characterized Magnific's modifications as "some post training that we've done on GPT two" but subsequently described the system's distinguishing technology as "prompt-based" and consisting of textual instructions shaped by the company's creative specialists.

That explanation creates an unresolved technical distinction.

In machine learning, post-training generally refers to additional training or refinement of an existing model after its initial development. Prompt engineering and orchestration, by contrast, can change how a model behaves without modifying its underlying parameters.

Magnific's subsequently supplied written FAQ takes an unequivocal position on the distinction.

Asked whether the company trained its own model, the document states: "No. Nothing was trained."

The FAQ instead attributes Magnific One's behavior to prompt improvement, references, creative-direction instructions and the application of brand rules before generation.

It does not identify GPT Image 2 as the underlying model, describing the foundation more generally as current image-generation technology.

The interview and written materials therefore support different interpretations of Magnific's technical contribution.

López Martín's comments suggest some degree of additional training or model modification. The FAQ describes a system operating around an existing model without any additional training.

Both accounts identify Magnific's creative-direction technology as a principal differentiator, but they do not establish whether the underlying model's weights were changed.

The distinction is important for enterprise technical teams evaluating whether they are adopting a distinct model or a specialized application built on top of another provider's technology.

It also affects questions about system portability, dependencies on OpenAI, long-term reproducibility and whether the creative-direction system could eventually operate across multiple foundational models.

During the interview, Magnific representatives said they were exploring ways to extend the technology to additional image generators.

For now, however, Magnific One and its Brand Kit integration remain a combined offering.

The company has not provided a technical explanation reconciling the interview's post-training characterization with the FAQ's statement that no training took place.

How Magnific One compares with Runway, Adobe Firefly, Midjourney and FLUX

Sources: Magnific pricing, Runway API pricing, Adobe Custom Models, Midjourney plans and Fal.ai FLUX.2 Pro pricing.

Magnific faces significant competition, particularly because several established platforms already provide ways to maintain visual consistency across AI-generated images.

The most directly comparable offering may be Runway, which has developed Brand Kits of its own.

Runway's kits let organizations save and share images, logos, colors, descriptions and other visual references across projects. Its AI agent can use those materials during generation, and it can create a kit from written instructions or a PDF containing brand guidelines.

Runway's September 2026 updates added agent-based Brand Kit creation and support for accessing kits through its MCP integration.

Like Magnific, Runway is positioning reusable brand context as a way to reduce the effort required to produce consistent marketing assets.

Adobe Firefly takes another approach. Its Custom Models capability allows companies to train specialized models on their own approved images, producing content that reflects particular visual styles, characters or subjects.

That creates an important technical distinction: Unlike Magnific's written description of Brand Kit as a prompting and guidance system, Adobe explicitly describes training customized models on customer-provided assets.

Midjourney, meanwhile, offers style and image references that can guide generations toward a desired aesthetic. It emphasizes creative control and visual quality, although its reference-based process is different from the structured enterprise brand management and automated compliance checks Magnific is emphasizing.

Black Forest Labs' FLUX models are another major competitor on the image-generation side. Through providers such as Fal.ai, developers can access FLUX generation and editing models directly, build workflows around reference images and, with supported variants, customize models for particular visual requirements.

Fal.ai is primarily an infrastructure and model-access provider rather than a direct equivalent of Magnific's completed creative application. For enterprises with engineering teams, however, it offers an alternative to buying a packaged interface: building their own brand-aware production systems on top of image-generation APIs.

The distinction is therefore less about whether competing image generators can maintain a visual style at all, and more about how much work the customer must do to establish, share, verify and enforce that consistency.

One point is especially important: Magnific cannot claim to be the first company offering AI-powered Brand Kits.

Runway provides functionality with the same name and overlapping capabilities, including the use of AI agents to interpret brand information.

Magnific's potentially meaningful distinction is the combination of automatic creative direction, brand-aware image generation, explicit post-generation compliance checks and precise editing within a single workflow.

Whether these features produce more consistent results than Runway's or Adobe's systems remains unproven by an independent comparison.

Magnific claims a substantial advantage in private blind image comparisons

Magnific is also attempting to establish its image generator as a competitor on visual quality, not just enterprise workflow features.

During the briefing, López Martín described an internal blind-comparison arena in which participants selected preferred images without being told which system produced them.

The tested models included Magnific One, Midjourney, GPT Image 2, Google's Nano Banana 2 and Seedream 5 Pro.

According to López Martín, Magnific One recorded an approximately 68.5% win rate during the evaluation, putting it ahead of the other participating models.

She said the comparisons were intended to emphasize aesthetic preferences, with some tests presenting images without their original prompts.

However, Magnific has not published the full results, voting data or methodology.

The reported win rate therefore cannot be treated as independent evidence that Magnific One outperforms rival models.

It is also unclear how the results would change with newer releases, including GPT Image 2.5, which was not included in the described evaluation.

The company's own written FAQ acknowledges that comparative evaluations have not been publicly released.

For enterprise customers, an equally relevant benchmark would be the percentage of generated images that satisfy brand guidelines without correction, the number of revisions needed to reach approval and the total cost of producing usable campaign assets.

Magnific has not supplied comparable measurements for those tasks.

Pricing, subscriptions and availability

Magnific One is available worldwide on the company's paid subscriptions, with pricing based on image resolution and quality.

The service charges 180 credits for High-quality 2K images, 250 credits for High-quality 4K, 600 credits for Max-quality 2K and 1,000 credits for Max-quality 4K.

Based on Magnific's published annual subscription rates and credit allocations, those charges translate to approximately $0.12 to $0.73 per image, assuming customers consume their full annual credit allowances. The actual effective cost varies according to plan, usage and generation settings.

Magnific's individual annual subscriptions begin at $14.50 per month for Premium, followed by Premium+ at $33.75 per month and Pro Starter at $82.50 per month, before taxes.

For the first week of availability, October 8 through October 15, Magnific is also offering unlimited generations through its desktop application on eligible higher-tier subscriptions. The promotion covers both Draft and Final modes at High quality in 2K and 4K, while Max-quality images continue to consume credits.

However, when VentureBeat downloaded the desktop application for the first time and logged in with an older account, we were not given the unlimited generations promotion, and instead directed to subscribe to be able to use the new Magnific One model.

From Freepik's stock-image library to Magnific's AI platform

Magnific One represents the latest stage in a substantial transformation for the Málaga, Spain-based company, which operated as Freepik for more than fifteen years.

Freepik was founded in 2010 by Alejandro Blanes, Pablo Blanes and Joaquín Cuenca as a search engine and marketplace for design resources, including photographs, vectors, icons and templates.

Its business grew around a freemium model that let customers access a limited selection of graphics without payment while charging subscribers for expanded libraries and features.

In May 2020, private equity firm EQT agreed to acquire a majority stake in the business, as TechCrunch reported at the time. Freepik reported approximately 32 million monthly visitors and 20 million registered users at that point.

The company's transition toward generative AI accelerated as new image models began challenging the traditional business of licensing preexisting stock imagery.

A key acquisition came in May 2024, when Freepik purchased Magnific, a Spanish startup known for an AI-powered image upscaler that could enhance and reinterpret existing images.

As Tech.eu reported, the acquisition was Freepik's largest to that point and brought specialized generative-image technology into a company historically associated with stock content.

The two brands initially continued operating separately, with Freepik serving as the larger creative platform and Magnific retaining its identity as a specialized AI image enhancement product.

That changed on April 28, 2026, when Freepik adopted the Magnific name across its broader business.

According to The Next Web's coverage of the rebrand, the consolidation reflected the company's move away from its identity as a stock-asset marketplace and toward an integrated production environment for AI-generated images, video and other creative content.

Magnific reported approximately $230 million in annual recurring revenue and more than one million paying subscribers around the rebrand. Those figures are company-reported.

Today, the platform offers third-party AI image and video models, editing tools, collaborative workflows and an extensive licensed asset library.

Magnific One adds a new dimension to that strategy. Rather than simply giving customers access to competing image models through one interface, the company is attempting to build its own differentiated creation experience on top of foundational technology.

The approach could allow Magnific to preserve some product differentiation even as the underlying image generators become more widely available through competing services.

What Magnific One means for enterprise AI adoption

Magnific One illustrates how the competitive landscape around generative AI is changing as powerful foundation models become available through multiple products and interfaces.

For marketing teams, the challenge is increasingly not whether an AI system can produce an attractive image from a prompt. It is whether that image can be generated repeatedly, under established corporate guidelines, by employees who may not have specialized design or prompting expertise.

A system that centralizes those guidelines, identifies deviations and allows employees to correct the results could make AI generation more practical for campaigns requiring large numbers of creative assets.

The potential advantages extend beyond reduced manual design work.

Shared brand instructions could help organizations maintain consistency across departments, outside agencies and regional offices, while automated checks could reduce the number of assets requiring extensive human revision.

However, Magnific's materials do not establish how often its brand checks successfully detect deviations, how reliably the underlying generator follows brand guidelines or what the full workflow costs compared with competing enterprise design software.

Its private aesthetic benchmark also does not resolve those operational questions.

Magnific's strategy ultimately rests on a broader proposition: The most valuable AI creative product may not be the company that trains the most capable underlying image generator, but the one that makes powerful generators behave consistently enough for everyday professional use.

With Magnific One, the company is betting that its experience in creative software, combined with OpenAI's image-generation technology and persistent brand guidance, can deliver that experience.

The remaining question is whether its integrated creative direction and verification systems offer enough measurable value to stand apart from increasingly capable competitors offering similar workflows.