Drop an image into ChatGPT and watch it think. A photograph of a circuit board returns component labels. A snapshot of a restaurant menu in Mandarin returns an English translation. A blurry whiteboard photo becomes clean, typed meeting notes. GPT-4V processes visual information with the same depth it brings to text.
From handwriting recognition and chart interpretation to document scanning and accessibility descriptions, ChatGPT Vision turns any camera into an analytical instrument.
ChatGPT Vision runs on GPT-4V (GPT-4 with Vision) and GPT-4o multimodal models. It accepts JPEG, PNG, GIF, WebP, and BMP images up to 20 MB each, with up to 10 images per message. Core capabilities include optical character recognition (OCR) for printed and handwritten text, chart and graph interpretation, photo description and scene analysis, document scanning, mathematical equation solving from photos, UI screenshot analysis, and detailed alt text generation for accessibility. Vision is available on all plans, with free users receiving limited message counts. All uploaded images are encrypted in transit and at rest, complying with SOC 2 Type II, GDPR, CCPA, and ISO 27001 standards.
GPT-4V does not simply label objects. It reads, interprets, and reasons about visual content.
Standard OCR reads characters. ChatGPT Vision reads meaning. Upload a photograph of a handwritten grocery list, and ChatGPT does not just transcribe "eggs, milk, bread" — it recognizes the list format, suggests recipes based on the ingredients, and estimates a shopping budget. Upload a medical prescription, and it identifies the medication name, dosage, and frequency despite the notoriously poor handwriting.
Printed text recognition works across languages. Photograph a Japanese product label, a German legal document, or an Arabic street sign, and ChatGPT transcribes and translates in a single step. Recognition accuracy exceeds 95% for well-lit, clearly printed text and drops to roughly 80% for handwritten cursive in poor lighting conditions.
Large-scale digitization projects have embraced similar OCR-based technologies to make millions of historical documents searchable and accessible to researchers worldwide.
Upload a bar chart, line graph, pie chart, scatter plot, histogram, or candlestick chart, and ChatGPT reads it like an analyst. It identifies axis labels, data ranges, trend directions, outliers, inflection points, and relative proportions. Ask "what does this chart show?" and receive a paragraph-length summary suitable for dropping directly into a report.
Multi-series charts work too. ChatGPT distinguishes between series using color, pattern, and legend references. A stacked bar chart showing quarterly revenue by product line returns individual product breakdowns and year-over-year growth rates. A scatter plot with regression lines returns the approximate correlation coefficient and statistical significance assessment.
This capability bridges a critical gap for professionals who receive charts in presentations, PDFs, or screenshots without access to the underlying data. Instead of manually reading values off axes, ChatGPT extracts the numbers and performs calculations on them — all from a photograph.
ChatGPT describes photographs with remarkable granularity. Upload a landscape photo and receive descriptions of terrain features, vegetation types, weather conditions, lighting direction, and estimated time of day. Upload an interior design photo and receive furniture identification, color palette analysis, style classification (mid-century modern, industrial, Scandinavian), and spatial layout observations.
Object counting, spatial relationship mapping, and material identification all function reliably. ChatGPT can estimate distances based on known object sizes, identify brand logos, read text on signs and packaging, and describe the emotional mood conveyed by composition and color choices.
Product photography analysis helps e-commerce sellers optimize listings. Upload a product photo and ask ChatGPT to evaluate lighting quality, background consistency, angle selection, and how well the image communicates the product's key features. The AI provides specific, actionable feedback rather than generic photography advice. Combine this with ChatGPT's DALL-E plugin to generate improved product imagery.
Photograph a receipt, invoice, business card, form, or contract page, and ChatGPT extracts structured data. A receipt photograph returns itemized line items, subtotals, tax amounts, and totals in a clean format. A business card yields name, title, company, phone, email, and address parsed into separate fields.
Multi-page document processing works when you upload sequential page images. ChatGPT maintains context across pages, understanding that page 3 continues the argument from page 2. Legal professionals use this to process discovery documents. Accountants photograph expense receipts for extraction into spreadsheets. Field researchers digitize handwritten field notes into structured databases.
For organizations handling sensitive documents, all uploaded images process within ChatGPT's SOC 2 Type II certified infrastructure. Team and Enterprise plans include data training opt-out by default, ensuring uploaded documents are not used to improve models. The National Institute of Standards and Technology (NIST) maintains benchmarks for document analysis accuracy that inform the evaluation of tools like ChatGPT Vision.
What you can upload and what ChatGPT returns for each image category.
| Image Type | What ChatGPT Extracts | Accuracy Level | Best Use Cases |
|---|---|---|---|
| Photographs | Object identification, scene description, spatial relationships, colors, text | High | Product analysis, real estate, field documentation |
| Screenshots | UI elements, text content, layout structure, error messages | Very High | Bug reports, UI review, tutorial creation |
| Charts & Graphs | Data values, trends, axis labels, outliers, series comparison | High | Report writing, data validation, presentation prep |
| Handwritten Notes | Text transcription, list structure, mathematical notation | Moderate-High | Meeting notes, lecture capture, field notes |
| Scanned Documents | Full text OCR, table structure, form field values | Very High | Digitization, legal review, accounting |
| Mathematical Equations | LaTeX notation, step-by-step solutions, variable identification | High | Homework help, research verification, tutoring |
| Maps & Diagrams | Labels, spatial relationships, flow direction, component identification | High | Navigation, architecture, engineering review |
| Product Labels | Ingredient lists, nutritional facts, multilingual text, barcodes (text only) | High | Health tracking, translation, comparison shopping |
| Medical Images | Structural description (non-diagnostic), comparison with reference images | Moderate | Education, preliminary review (not clinical diagnosis) |
| Art & Design | Style identification, technique analysis, color palette, composition | High | Art education, design feedback, creative inspiration |
Visual AI that makes the world more accessible to people with vision impairments.
Upload any image and ask ChatGPT to "describe this image for a screen reader." The response includes object identification, spatial relationships, colors, text content, and emotional tone — formatted specifically for assistive technology consumption. Web developers use this to generate accurate alt text for thousands of product images. Content managers retrofit existing image libraries with descriptive metadata.
A visually impaired user photographs a restaurant menu, a bus schedule, a medicine bottle, or a street intersection. ChatGPT reads all visible text, describes the scene layout, and answers follow-up questions. "What is the second item on the lunch menu?" or "Which bus arrives next?" Combined with ChatGPT Voice Mode, the entire interaction happens through speech — photograph with the phone camera, hear the description through earbuds.
Textbook diagrams, mathematical figures, scientific charts, and geographic maps become accessible through verbal description. A biology student who cannot see a cell diagram receives a detailed spatial description: "The mitochondria are shown as oval organelles near the cell nucleus, colored in orange, with internal folded membranes called cristae visible as parallel lines." This level of detail transforms access to visual educational content. Visit our GPT Models page to understand which models power these vision capabilities.
Combining image analysis with other ChatGPT capabilities produces workflows that no single tool handles alone.
Upload two or more images and ask ChatGPT to compare them. Before and after renovation photos yield a detailed change list. Two product packaging designs get evaluated on visual hierarchy, color impact, and text legibility. Competitive product screenshots reveal feature differences and UI patterns. ChatGPT tracks the comparison across up to 10 images per message, maintaining consistent criteria throughout the analysis.
Photograph a hand-drawn wireframe or screenshot a design mockup. ChatGPT generates corresponding HTML, CSS, and JavaScript code that reproduces the visual layout. Buttons, forms, navigation bars, grids, and typography choices translate from pixels to functional code. A designer's napkin sketch becomes a working prototype in minutes. This workflow pairs naturally with the ChatGPT API for automated design-to-development pipelines.
Researchers upload microscopy images, gel electrophoresis results, spectroscopy charts, and astronomical photographs. ChatGPT identifies structures, measures relative sizes, reads wavelength values, and compares patterns against known references. While not a replacement for peer-reviewed analysis, it accelerates preliminary assessment and generates descriptive text for figure captions. NASA (NASA AI) has explored analogous multimodal AI applications for processing satellite imagery and planetary observation data.
Upload your first image on the free plan. Upgrade to Plus for higher limits and priority vision processing.
Get Started FreeAnswers to common questions about image upload, analysis capabilities, and privacy.
ChatGPT Vision handles photographs, screenshots, charts, graphs, diagrams, handwritten notes, scanned documents, maps, UI mockups, product photos, medical images (non-diagnostic only), and artwork. Supported file formats include JPEG, PNG, GIF, WebP, and BMP. Each image can be up to 20 MB, and you can upload up to 10 images per conversation turn. Higher resolution images produce more detailed analysis, though ChatGPT processes lower-resolution photos effectively as well.
ChatGPT uses GPT-4V and GPT-4o multimodal vision to interpret handwritten text from photographs. Photograph a notebook, whiteboard, sticky note, or handwritten form, and ChatGPT transcribes the content into clean, editable digital text. Accuracy is highest for clearly printed handwriting in well-lit conditions (above 95% character accuracy). Cursive and stylized handwriting accuracy drops to approximately 80-85%. For best results, photograph at a straight angle with even lighting and minimal shadows. See our Plugins page for using Advanced Data Analysis to process the extracted text further.
Yes. ChatGPT identifies chart types (bar, line, pie, scatter, histogram, candlestick, waterfall, area, and radar), reads axis labels and data values, calculates approximate percentages and growth rates, identifies outliers and inflection points, describes trends in plain language, and generates written summaries suitable for reports. Multi-series charts with legends are supported. Accuracy is highest for clean, well-labeled charts and decreases with cluttered or low-resolution images.
Yes. Free-tier users can upload images for vision analysis through GPT-4o with a limited number of messages per day. ChatGPT Plus ($20/month) provides higher message limits with priority processing. Team ($25/user/month) and Enterprise plans include the highest usage allocations and admin controls for image data retention. All plans support the same image formats and analysis capabilities. Compare access levels on our plans page.
ChatGPT Vision generates detailed alt text descriptions optimized for screen readers, transcribes text from photographs for visually impaired users, describes scene layouts and spatial relationships, identifies colors and objects, reads foreign-language text and translates it, and processes handwritten content that scanning tools miss. When paired with ChatGPT Voice Mode, users receive spoken descriptions of uploaded images, enabling fully hands-free and eyes-free visual information access. The Custom GPTs Builder lets developers create accessibility-focused assistants for specific use cases.
Vision works alongside every other ChatGPT capability for comprehensive AI assistance.
GPT-4V and GPT-4o power vision analysis. Compare model capabilities and performance benchmarks.
Combine image uploads with spoken conversation for hands-free visual analysis workflows.
Use Advanced Data Analysis to process data extracted from charts, tables, and document scans.
Build specialized vision assistants for product photography, medical education, or document processing.
Integrate ChatGPT Vision into your applications via the Vision API endpoints with image URL or base64 input.
See vision message limits and processing priority across Free, Plus, Team, and Enterprise tiers.
ChatGPT