Opan — Multimodal Vision Assistant

Built on Qwen3-VL · 2B Parameters · 256K Context · Vision + Text

Opan is a lightweight multimodal assistant tuned for friendly, clear, and accurate responses. Based on Qwen3-VL:2b, it combines strong visual understanding with natural conversational ability — all in a 1.9GB package you can run locally.

Quick Start

ollama run aeline/opan

Requires Ollama 0.12.7 or later

What Opan Can Do

Vision Understanding Send images alongside your text. Opan can describe scenes, identify objects, read text from photos, interpret diagrams, and answer questions about what it sees.

OCR in 32+ Languages Extracts and interprets text from images even under difficult conditions — blurry, tilted, low-light, or densely formatted documents. Preserves structure across multi-page inputs.

Visual-to-Code Describe a UI mockup or flowchart with an image and ask Opan to generate code. Supports HTML/CSS layout generation, JavaScript from diagrams, and logic extraction from flowcharts.

Long Context — 256K Tokens Feed it full PDFs, long documents, extended conversations, or multi-image sequences. Opan maintains coherence across very long inputs.

Spatial & Layout Reasoning Understands object positions, relationships, chart structures, and UI layouts. Useful for analyzing screenshots, design files, or annotated images.

System Requirements

Example Usage

Image + Text

>>> [attach image of a receipt] What is the total amount and date on this receipt?

OCR

>>> [attach image of a handwritten note] Transcribe this text.

Visual Coding

>>> [attach UI mockup image] Convert this design into HTML and CSS.

Document Analysis

>>> [attach PDF screenshot] Summarize the key points from this page.

Model Details

Notes

• Opan uses a custom system prompt tuned for warm, clear, and helpful responses
• It is not a reasoning/thinking model — for step-by-step chain-of-thought tasks, a larger model is recommended
• Performance on complex multi-step visual reasoning may vary at 2B scale

Made by sk16er