Saudi Arabia is getting an AI image model designed around its culture

Built around a culture, not adapted to it Adobe's Firefly Foundry and HUMAIN are building an AI image model designed specifically around Arabic-language prompting and Saudi and Arab cultural context, according to my.gov.sa's official summary of LEAP 2026's announcements . That framing distinguishes the project from the far more common approach of taking a model trained primarily on English-language, Western-centric data and adding Arabic-language support on top of it afterward — a retrofit that often leaves cultural context and visual accuracy noticeably weaker than the same model's performance in its original training language. What Firefly Foundry actually is Firefly Foundry is Adobe's platform for building and customising generative AI models tailored to a specific partner's needs, rather than relying solely on Adobe's own general-purpose Firefly models. Using that platform in partnership with HUMAIN means the resulting Saudi-focused image model is tailored to this market's language and cultural references; the materials TechScoop reviewed did not set out how it is trained or how much it departs from Adobe's existing Firefly models. Why prompting in Arabic is harder than it sounds Generative image models generally perform best in the language and cultural context they were most heavily trained on, and Arabic — with its own script, grammar and regional cultural references — poses a genuinely different set of challenges than simply translating English prompts word for word. A model designed around Arabic-language prompting from the outset can interpret nuance, regional terminology and cultural references far more accurately than one that first translates an Arabic prompt into English internally before generating an image, a common workaround that tends to lose exactly the contextual detail that makes an image feel authentic rather than generic. Why cultural context matters as much as language Language is only part of the challenge; visual cultural context is the other half. A model trained predominantly on Western imagery data can struggle to accurately render regionally specific clothing, architecture, landscapes and everyday scenes without introducing subtle inaccuracies or generic substitutions. Building a model with Saudi and Arab cultural context embedded from the start is aimed directly at closing that gap, giving local creators, marketers and businesses a tool that produces visually authentic results rather than requiring extensive manual correction after the fact. Part of a much larger Adobe-HUMAIN partnership This culturally contextualised model is one specific piece of Adobe's broader Saudi commitment, which TechScoop covered in full in its piece on Adobe's more than $4 billion expanded Saudi partnership . Seen alongside that wider deal, the localised image model functions as a concrete demonstration of what "locally contextualised generative AI" actually means in practice, rather than remaining an abstract phrase attached to a large investment figure. What this could mean for local creators For Saudi and wider Arab-region creators, marketers and small businesses, a genuinely localised image model lowers a real practical barrier: the need to either accept visually inauthentic AI-generated content or spend significant additional time and budget correcting it manually. If the model performs as designed, it could meaningfully expand who is able to use generative AI image tools productively — extending the benefit beyond creators fluent enough in English-language prompting techniques and toward a much broader Arabic-speaking user base. Part of a wider race to build Arabic AI This localised image model is also one entry in a broader, increasingly competitive push to build genuinely capable Arabic-language AI systems, spanning text, voice and image models alike. TechScoop tracks that wider competitive landscape, including efforts like the Arabic-focused ALLAM language model and dedicated Arabic AI companies, in a separate piece on how the race to build Arabic AI is entering a new phase . Adobe and HUMAIN's culturally contextualised image model adds a visual-generation dimension to that race, alongside the text- and language-focused efforts already under way elsewhere in the market. The test will be in daily use, not the demo As with any AI model claiming improved localisation, the real test will come once ordinary users start prompting it for everyday tasks rather than curated demonstrations. Whether the model reliably renders accurate regional clothing, architecture and everyday scenes across a wide range of prompts, rather than only performing well on the specific examples chosen to showcase it at launch, is the detail that will ultimately determine whether Saudi and Arab creators adopt it as a genuine working tool rather than treating it as a one-time novelty. Sources my.gov.sa (National Platform)