Skip to main content
News

UAE's e& unveils Imagine&, a sovereign image AI trained on Emirati iconography

The UAE's e& has launched Imagine&, an image generator trained only on Emirati imagery, as the Gulf's telco giants race to own the visual layer of sovereign AI.

· Updated Apr 18, 2026 6 min read
UAE's e& unveils Imagine&, a sovereign image AI trained on Emirati iconography
## UAE's e& unveils Imagine&, a sovereign image AI trained on Emirati iconography The UAE's telecoms giant **e&**, formerly Etisalat, has launched **Imagine&**, a sovereign image-generation AI trained exclusively on Emirati imagery, as the Gulf's telecom operators race to own the visual layer of sovereign AI. Unveiled this week at a Dubai technology showcase, Imagine& is designed to produce photos, illustrations, and branded assets that actually look Emirati, from Ghaffia scarves to Burj Khalifa skylines, without the persistent stereotypes and mistranslations that foreign image models tend to produce. It is the UAE's most explicit visual-AI answer yet to the question of whose aesthetics the internet will represent next. ## Why sovereign image models matter Foreign image models have a well-documented problem with MENA imagery. Gulf architecture is often mashed together with generic Middle Eastern clichés, traditional dress appears in muddled variants, and national flags are rendered inaccurately on demand. For a region investing heavily in digital national identity, that is not acceptable. Imagine& sets out to fix it by training on curated Emirati visual assets, including heritage imagery, UAE-specific urban photography, traditional clothing references, and approved national symbols, while explicitly excluding content that violates UAE norms. The tool is also tuned for Modern Standard and Gulf Arabic prompts, so users type in Arabic and get recognisably Emirati output. ### By The Numbers - More than 160 internal machine learning systems already deployed at **e&**, one of the largest production AI footprints in the region. - Nearly 60% of UAE residents use generative AI, the highest share per capita globally, per Microsoft's AI Diffusion Index. - 500,000 advanced Nvidia chips a year now cleared for UAE delivery under Microsoft's new export licence. - $15bn Microsoft commitment to UAE AI infrastructure, talent, and research through 2030. - More than 100 partner organisations expected to tap Imagine& via an e&-hosted API in the first 12 months. UAE's e& unveils Imagine&, a sovereign image AI trained on Emirati iconography ## Who Imagine& is built for The first customer group is government. Ministries, tourism authorities, and cultural institutions will use Imagine& for campaigns, signage, and social content, where Emirati accuracy and visual dignity matter most. The second is enterprise. UAE banks, real estate groups, retailers, and airlines need visual content that flatters local audiences, and generic foreign models force them into endless retouching. The third is consumer creators. Imagine& will be exposed via the e&.ai consumer app and selected Snap partnerships, giving UAE residents a tool that produces culturally accurate Eid cards, birthday imagery, and branded memes without friction. > "At a technology fair in Dubai last month, a developer typed two Arabic words into a computer: 'National Day celebration.' Within seconds, the screen filled with images from the Emirates' national story." > — e& Imagine& showcase coverage, April 2026 > "We want Imagine& to produce images that Emirati grandmothers and Emirati teenagers both recognise as their own, without checking twice." > — Hatem Dowidar, Group Chief Executive, e& ## How it fits the UAE's wider AI stack Imagine& plugs into a UAE AI stack that is growing denser by the quarter. **G42**'s compute, Microsoft's **Azure AI Foundry** capacity, Nvidia's licensed chip flows, and **MBZUAI**'s research base now feed a wave of UAE-built products. Our coverage of the [Microsoft export licence for 500K Nvidia chips a year to the UAE](/news/microsoft-nvidia-uae-export-licence-15bn-ai-hub-2026) explains the infrastructure context, while the [Falcon, Jais, and ALLaM Arabic LLM scoreboard](/arabic-ai/arabic-llm-scoreboard-april-2026-falcon-jais-allam) shows how the language side is shaping up. Add in the [HumAIn AI agent marketplace in Saudi Arabia](/news/humain-one-ai-agent-marketplace-saudi-arabia) and the [UAE AI and robotics strategy targeting 30% of the 2040 market](/news/uae-ai-robotics-strategy-30-percent-global-market-2040), and the regional picture becomes coherent rather than scattered.
PlayerRole in the UAE sovereign AI stackImagine& relationship
e&Telco, AI product ownerOperator and publisher
G42Sovereign AI champion, computeInfrastructure partner
MicrosoftHyperscaler, Nvidia licence holderCompute supplier
MBZUAIResearch baseModel design and evaluation
TIIFalcon model familyLanguage side complement
## The pitfalls sovereign image models must avoid Three traps wait for sovereign image models. The first is narrowness. If Imagine& is trained too tightly, it will be excellent at UAE imagery and weak everywhere else, which limits its utility for international campaigns. The second is bias. Curated training data can reproduce exclusion if not carefully audited, and the UAE's social fabric is more diverse than most sovereign models initially reflect. The third is enforcement. Deepfakes, political imagery, and targeted misuse all need guardrails. The e& team has signalled explicit content filters and human review for politically sensitive prompts, plus watermarking on every output. Execution, as always, will determine how well those guardrails hold. - Curated training data focused on UAE imagery, architecture, dress, and symbols. - Arabic-first prompting, including Gulf dialect, and bilingual Arabic-English interfaces. - Explicit content filters and watermarking for all Imagine& outputs. - API access for enterprise and government customers through e&.ai. - Future roadmap items including video generation and bilingual Eid and National Day packs.
The AI in Arabia View: Imagine& is a small tool with an outsized symbolic weight. It is the UAE's visual answer to the problem that foreign AI systems do not see us the way we see ourselves. e& has the compute, the distribution, and the brand recognition to make a sovereign image model useful instead of ornamental. The risk is falling in love with the idea of sovereign visuals and ignoring the actual user experience. To win, Imagine& must be at least as easy to use as the foreign competition, at least as accurate on the UAE's own subjects, and far better at saying no to the prompts it should refuse. If the team nails those three tests, expect similar tools from Saudi Arabia, Qatar, and Egypt within 12 months.
## Frequently Asked Questions ### What is Imagine&? Imagine& is a sovereign image-generation AI built by **e&**, the UAE's largest telecoms group. It is trained on curated Emirati visual assets, supports Arabic-first prompting, and is designed to generate images that accurately reflect UAE architecture, dress, and national symbols while avoiding the cliches typical of foreign models. ### Who can use Imagine&? Imagine& will be available first to UAE government ministries, cultural institutions, and enterprise customers through the e&.ai platform. A consumer release is planned via the e&.ai app and selected partnerships, including Snap lenses designed for Eid, National Day, and UAE-specific cultural moments. ### How does Imagine& relate to Microsoft's chip licence? The US export licence to Microsoft for 500,000 Nvidia chips a year gives the UAE the compute headroom to run and scale tools like Imagine&. Without that infrastructure, sovereign image generation at this quality would be far more expensive and far slower to iterate. ### Will other Gulf telcos follow? Expect similar launches from Saudi Arabia, Qatar, and Egypt within 12 months. **stc**, **Ooredoo**, **Mobily**, and **Vodafone Egypt** all have the relationships and the national data to build similar tools. Competitive pressure and sovereign AI politics both point in the same direction. Is a sovereign image AI a genuine cultural necessity, or a nice-to-have that global models will eventually match? Drop your take in the comments below.