Magnific launched One and Brand Kit. The PR does not name GPT Image 2

What did the 8 October release say?

The Cision PR Newswire article we opened, headline “Magnific One: an AI image model with art direction built in,” is dated 8 October 2026, 09:34 ET, and lists Magnific as the source. The lead says Magnific announced One to bring “built-in art direction and brand consistency” to generated images. A following sentence says the model “reads a simple prompt and decides the art direction before generating the image.”

Chief experience officer Martin LeBlanc is quoted: “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.” Brand Kit, “launching alongside Magnific One,” is described as an agent that extracts logo, palette, typography, imagery, and tone from a website, a guidelines PDF, a few images, or a written description. The team reviews that first kit. “Magnific One then applies it to every image it generates.”

Availability on the release is a single sentence: One is available that day on all Magnific paid plans worldwide, on the web, in desktop and mobile apps, and through the Magnific MCP, at magnific.com/one. The about-block says the company was founded as Freepik in Málaga in 2010, relaunched as Magnific in 2026, and “serves more than one million paid subscribers worldwide” from Málaga and San Francisco. Those subscriber and relaunch lines are Magnific’s.

What did VentureBeat add from the briefing?

Franzen’s 8 October VentureBeat story, which we read in full, is reporting from an exclusive briefing with Sofía López Martín, Magnific’s head of product marketing. It says One inserts a creative-direction step between the user’s prompt and generation, accepts up to five reference images, and offers Draft mode — eight or sixteen alternatives for the credit cost of one image — and Final mode at 2K or 4K. Brand Kit, in that write-up, influences Final-mode composition, then checks style, color, typography, and logo use, plus artifacts, framing, and unreadable text, and offers a correction path.

The technical paragraph is the one the PR omits. López Martín, as quoted by VentureBeat, said “It’s powered by GPT-Image-2,” and at one point described “some post training that we’ve done on GPT two,” then described the distinguishing work as prompt-based. VentureBeat says a written FAQ later answered whether Magnific trained its own model with “No. Nothing was trained,” and did not name GPT Image 2. We did not receive that FAQ. We are reporting VentureBeat’s account of the interview and of the document.

Vendor quality talk and an unnamed base model are the usual announcement pattern. For keeping those columns apart, see how to read an AI company announcement. For another October image model we actually sampled once, see Nano Banana 2.1; that test is not a Magnific One comparison.

How do you get it, and what does VentureBeat say it costs?

The PR’s access line is paid plans, worldwide, four surfaces. VentureBeat reprints Magnific credit prices: 180 credits for High 2K, 250 for High 4K, 600 for Max 2K, and 1,000 for Max 4K. It derives about $0.12 to $0.73 per image from annual plans if the full credit allotment is used, and lists individual annual list prices starting at $14.50 a month for Premium, $33.75 for Premium+, and $82.50 for Pro Starter, before tax. Those dollar figures are VentureBeat’s arithmetic on Magnific’s published rates, not a price we were quoted.

VentureBeat also reports an 8–15 October desktop promo: unlimited High-quality Draft and Final at 2K and 4K on eligible higher-tier plans, with Max still burning credits. The same story says VentureBeat’s own older-account desktop login did not receive that window. That is their login, not ours.

Brand Kit is optional on VentureBeat’s account and “currently works only with Magnific One rather than the other image-generation models available through the Magnific platform.” Runway already ships a product also called Brand Kit; VentureBeat notes Magnific is not first to that name. Magnific’s unpublished private arena figure, about 68.5% win rate against Midjourney, GPT Image 2, Nano Banana 2, and Seedream 5 Pro, is Magnific’s, with no public votes or protocol.

What did the connected Magnific catalog show?

On 9 October we listed the signed-in Magnific MCP image-model catalog and read the Brand Kit guide. The catalog row for slug magnific-one names the model Magnific One, family magnific, tool text-to-image. It lists aspect ratios including 16:9, resolutions 1k, 2k and 4k, qualities medium and high, supportsReferences true, extraSettings including draft, requiresPremium true, featured true, and private true. The server note on that list says private means “internal or feature-flagged, not generally available.” That flag sits next to, and does not cancel, the PR’s paid-plan sentence. The same catalog text says Brand Kit only works with magnific-one. For what an MCP server is, see MCP explained.

The Brand Kit guide we opened describes a draft kit that becomes ready after the owner reviews extracted atoms, and says website and document extraction have a daily per-user limit. It says PDF and text documents “cannot be uploaded from outside the app yet,” which is narrower than the PR’s PDF sentence. We did not create a kit, scrape a URL, or call images_generate. The catalog’s agentRecommendation describes Draft as 4, 8, or 16 images; VentureBeat’s briefing said eight or sixteen. We are leaving both numbers in their columns.

A GET of https://magnific.com/one and https://www.magnific.com/one on 9 October each returned HTTP 403, so we have no product-page copy to reconcile with the PR. The connected catalog is an authenticated tool listing, not a substitute for that page.

What should teams take from the disagreement?

If you need a ship date and a packaging claim, the PR is enough: paid plans, Brand Kit beside One, MCP as a listed surface. If you need a base-model name or a training claim, you are in VentureBeat’s briefing, where the interview and the FAQ do not line up. “Post-training” and “nothing was trained” cannot both be taken as a settled technical description from Magnific’s own materials as we have them.

The 68.5% arena figure and the per-image dollar range are not reasons to retire Midjourney, Firefly Custom Models, or Runway Brand Kits. VentureBeat says Magnific has not published the arena method, and it reports that its own desktop login did not receive the unlimited-week promo. We generated no Magnific One samples in this review, so we have no quality, brand-check, or latency observation of our own.

What did we not verify?

This is an evidence review of the 8 October PR Newswire HTML, VentureBeat’s same-day briefing story, and a 9 October listing of the connected Magnific MCP catalog and Brand Kit guide. We did not open magnific.com/one (403), did not receive Magnific’s FAQ, did not buy a plan, and did not generate Draft or Final images. Treat art-direction language as Magnific’s, GPT Image 2 and the training contradiction as VentureBeat’s report, and the catalog flags as what the signed-in tool returned.