We ran a comparative test of Al Rassam Al Arabi V3.1 R1 37 against DALL-E 3 and Midjourney V6 using standard Arabic prompts. The results are striking:
| Feature / Prompt | DALL-E 3 | Midjourney V6 | Al Rassam V3.1 R1 37 | | :--- | :--- | :--- | :--- | | Accuracy of Arabic text | 10% (gibberish) | 5% (random shapes) | 95% (legible) | | Understanding "Mashrabiya" | Generic lattice | Partial | Correct geometry | | Bedouin tent realism | Generic brown tents | Detailed but Western bias | Culturally accurate fabric & layout | | Speed (per 4 images) | 15 seconds | 25 seconds | 12 seconds | Al Rassam Al Arabi V3.1 R1 37
Note: The "37" suffix model outperforms its predecessors (V2.0) by a 40% margin in cultural relevance scores, according to internal beta testing. We ran a comparative test of Al Rassam Al Arabi V3
Indie game studios are utilizing the "37" style modes to generate concept art for characters like Jinn and Ghouls that adhere to authentic folklore rather than Hollywood stereotypes. Al Rassam Al Arabi V3.1 R1 37
Earlier versions of Al Rassam relied purely on latent diffusion. V3.1 R1 37 integrates a transformer-based language encoder specifically optimized for Modern Standard Arabic (MSA) and dialectal Arabic (including Egyptian, Levantine, and Khaleeji). This allows users to input prompts such as "ارسم لي خيلًا عربيًا أصيلًا في صحراء عند الغروب" (Draw me an authentic Arabian horse in the desert at sunset) without the prompt translation errors seen in Western models.
You do not need to speak perfect classical Arabic. V3.1 R1 37 understands regional dialects. A user can say:
The model will adjust the visual aesthetic accordingly—Egyptian brass trays versus Gulf dallah coffee pots.