Image-to-Text Endpoints

Curated picks for OCR, captioning/VQA, and detection/segmentation. Moondream 3 is the dominant pick across all three; Florence-2 and SAM-3 complete the toolset. Verify with genmedia models --endpoint_id <id> --json before running.

OCR, extract text from image

fal-ai/got-ocr/v2: GOT OCR 2.0
fal-ai/florence-2-large/ocr: Florence-2 Large (OCR head)
fal-ai/moondream3-preview/segment: Moondream 3 Preview (segment also reads text regions)
fal-ai/moondream3-preview/query: Moondream 3 Preview (query for text content)

Caption / VQA

Image description and visual question-answering.

fal-ai/moondream3-preview/caption: Moondream 3 · Caption
fal-ai/moondream3-preview/query: Moondream 3 · Query (VQA)
fal-ai/florence-2-large/caption: Florence-2 Large
fal-ai/florence-2-large/detailed-caption: Florence-2 Large · Detailed
fal-ai/florence-2-large/more-detailed-caption: Florence-2 Large · More Detailed
fal-ai/video-understanding: Video Understanding
fal-ai/auto-caption: Auto-Captioner
perceptron/isaac-01: Perceptron · Isaac 0.1
perceptron/isaac-01/openai/v1/chat/completions: Perceptron · Isaac 0.1 (OpenAI-compatible)

Detection / Segmentation

Nesne tespit ve maskeleme.

fal-ai/moondream3-preview/detect: Moondream 3 · Detect (open-vocabulary detection)
fal-ai/moondream3-preview/point: Moondream 3 · Point
fal-ai/moondream2/object-detection: Moondream 2 · Object Detection
fal-ai/moondream2/point-object-detection: Moondream 2 · Point Object Detection
fal-ai/sam-3/image/embed: SAM 3 · Image Embed (segmentation backbone)
fal-ai/florence-2-large/region-to-category: Florence-2 · Region-to-Category
fal-ai/florence-2-large/region-to-description: Florence-2 · Region-to-Description
perceptron/isaac-01: Perceptron · Isaac 0.1
perceptron/isaac-01/openai/v1/chat/completions: Perceptron · Isaac 0.1 (OpenAI-compatible)

Common parameters

genmedia schema fal-ai/moondream3-preview/query --json
genmedia schema fal-ai/got-ocr/v2 --json
genmedia schema fal-ai/sam-3/image/embed --json

Frequently exposed:

image_url: source image
prompt / query / question, for VQA or guided segmentation
threshold: confidence cutoff (detection)
output_format: for masks: png alpha, binary, coco-rle, etc.

Discovery