Fanar-2-Oryx-IG endpoint handler (Darbi)
A custom handler for a Hugging Face Inference Endpoint that serves QCRI/Fanar-2-Oryx-IG, QCRI's culturally aligned image generation LoRA, on its base model black-forest-labs/FLUX.1-schnell (Apache 2.0; gated, so the endpoint's token must have accepted its terms). No weights are stored here: the handler downloads both at start-up, with the versions the model card was tested with (diffusers 0.37.1, peft 0.18.1).
Request {"inputs": "A falconer at Souq Waqif", "parameters": {"width": 768, "height": 768, "steps": 4, "seed": 1}}, response {"image": "<base64 JPEG>", "seed": 1, "width": 768, "height": 768}.
Darbi (ุฏุฑุจู) uses it for image prompts in its prompt lab: a learner writes a prompt for an image, Darbi scores it, and Oryx draws the improved prompt. Built for the Qai x Hugging Face Hackathon 01 in Doha.