Agent skill
openai-vision
Analyze images and multi-frame sequences using OpenAI GPT vision models
Install this agent skill to your Project
npx add-skill https://github.com/benchflow-ai/skillsbench/tree/main/tasks/jpg-ocr-stat/environment/skills/openai-vision
SKILL.md
OpenAI Vision Analysis Skill
Purpose
This skill enables image analysis, scene understanding, text extraction, and multi-frame comparison using OpenAI's vision-capable GPT models (e.g., gpt-4o, gpt-4o-mini). It supports single images, multiple images for comparison, and sequential frames for temporal analysis.
When to Use
- Analyzing image content (objects, scenes, colors, spatial relationships)
- Extracting and reading text from images (OCR via vision models)
- Comparing multiple images to detect differences or changes
- Processing video frames to understand temporal progression
- Generating detailed image descriptions or captions
- Answering questions about visual content
Required Libraries
The following Python libraries are required:
from openai import OpenAI
import base64
import json
import os
from pathlib import Path
Input Requirements
- File formats: JPG, JPEG, PNG, WEBP, non-animated GIF
- Image sources: URL, Base64-encoded data, or local file paths
- Size limits: Up to 20MB per image; total request payload under 50MB
- Maximum images: Up to 500 images per request
- Image quality: Clear, legible content; avoid watermarks or heavy distortions
Output Schema
Analysis results should be returned as valid JSON conforming to this schema:
{
"success": true,
"images_analyzed": 1,
"analysis": {
"description": "A detailed scene description...",
"objects": [
{"name": "car", "color": "red", "position": "foreground center"},
{"name": "tree", "count": 3, "position": "background"}
],
"text_content": "Any text visible in the image...",
"colors": ["blue", "green", "white"],
"scene_type": "outdoor/urban"
},
"comparison": {
"differences": ["Object X appeared", "Color changed from A to B"],
"similarities": ["Background unchanged", "Layout consistent"]
},
"metadata": {
"model_used": "gpt-4o",
"detail_level": "high",
"token_usage": {"prompt": 1500, "completion": 200}
},
"warnings": []
}
Field Descriptions
success: Boolean indicating whether analysis completedimages_analyzed: Number of images processed in the requestanalysis.description: Natural language description of the image contentanalysis.objects: Array of detected objects with attributesanalysis.text_content: Any text extracted from the imageanalysis.colors: Dominant colors identifiedanalysis.scene_type: Classification of the scenecomparison: Present when multiple images are analyzed; describes differences and similaritiesmetadata.model_used: The GPT model used for analysismetadata.detail_level: Resolution level used (low,high, orauto)metadata.token_usage: Token consumption for cost trackingwarnings: Array of any issues or limitations encountered
Code Examples
Basic Image Analysis from URL
from openai import OpenAI
client = OpenAI()
def analyze_image_url(image_url, prompt="Describe this image in detail."):
"""Analyze an image from a URL using GPT-4o vision."""
response = client.chat.completions.create(
model="gpt-4o",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": prompt},
{
"type": "image_url",
"image_url": {
"url": image_url,
"detail": "high"
}
}
]
}
],
max_tokens=1000
)
return response.choices[0].message.content
Image Analysis from Local File (Base64)
from openai import OpenAI
import base64
client = OpenAI()
def encode_image_to_base64(image_path):
"""Encode a local image file to base64."""
with open(image_path, "rb") as image_file:
return base64.standard_b64encode(image_file.read()).decode("utf-8")
def get_image_media_type(image_path):
"""Determine the media type based on file extension."""
ext = image_path.lower().split('.')[-1]
media_types = {
'jpg': 'image/jpeg',
'jpeg': 'image/jpeg',
'png': 'image/png',
'gif': 'image/gif',
'webp': 'image/webp'
}
return media_types.get(ext, 'image/jpeg')
def analyze_local_image(image_path, prompt="Describe this image in detail."):
"""Analyze a local image file using GPT-4o vision."""
base64_image = encode_image_to_base64(image_path)
media_type = get_image_media_type(image_path)
response = client.chat.completions.create(
model="gpt-4o",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": prompt},
{
"type": "image_url",
"image_url": {
"url": f"data:{media_type};base64,{base64_image}",
"detail": "high"
}
}
]
}
],
max_tokens=1000
)
return response.choices[0].message.content
Multi-Image Comparison
from openai import OpenAI
import base64
client = OpenAI()
def compare_images(image_paths, comparison_prompt=None):
"""Compare multiple images and identify differences."""
if comparison_prompt is None:
comparison_prompt = (
"Compare these images carefully. "
"List all differences and similarities you observe. "
"Describe any changes in objects, colors, positions, or text."
)
content = [{"type": "text", "text": comparison_prompt}]
for i, image_path in enumerate(image_paths):
base64_image = encode_image_to_base64(image_path)
media_type = get_image_media_type(image_path)
# Add label for each image
content.append({
"type": "text",
"text": f"Image {i + 1}:"
})
content.append({
"type": "image_url",
"image_url": {
"url": f"data:{media_type};base64,{base64_image}",
"detail": "high"
}
})
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": content}],
max_tokens=2000
)
return response.choices[0].message.content
Multi-Frame Video Analysis
from openai import OpenAI
import base64
from pathlib import Path
client = OpenAI()
def analyze_video_frames(frame_paths, analysis_prompt=None):
"""Analyze a sequence of video frames for temporal understanding."""
if analysis_prompt is None:
analysis_prompt = (
"These are sequential frames from a video. "
"Describe what is happening over time. "
"Identify any motion, changes, or events that occur across the frames."
)
content = [{"type": "text", "text": analysis_prompt}]
for i, frame_path in enumerate(frame_paths):
base64_image = encode_image_to_base64(frame_path)
media_type = get_image_media_type(frame_path)
content.append({
"type": "text",
"text": f"Frame {i + 1}:"
})
content.append({
"type": "image_url",
"image_url": {
"url": f"data:{media_type};base64,{base64_image}",
"detail": "auto" # Use auto for frames to balance cost
}
})
response = client.chat.completions.create(
model="gpt-4o",
messages=[{"role": "user", "content": content}],
max_tokens=2000
)
return response.choices[0].message.content
Full Analysis with JSON Output
from openai import OpenAI
import base64
import json
import os
client = OpenAI()
def analyze_image_to_json(image_path, extract_text=True):
"""Perform comprehensive image analysis and return structured JSON."""
filename = os.path.basename(image_path)
prompt = """Analyze this image and return a JSON object with the following structure:
{
"description": "detailed scene description",
"objects": [{"name": "object name", "attributes": "color, size, position"}],
"text_content": "any visible text or null if none",
"colors": ["dominant", "colors"],
"scene_type": "indoor/outdoor/abstract/etc",
"people_count": 0,
"notable_features": ["list of notable visual elements"]
}
Return ONLY valid JSON, no other text."""
try:
base64_image = encode_image_to_base64(image_path)
media_type = get_image_media_type(image_path)
response = client.chat.completions.create(
model="gpt-4o",
messages=[
{
"role": "user",
"content": [
{"type": "text", "text": prompt},
{
"type": "image_url",
"image_url": {
"url": f"data:{media_type};base64,{base64_image}",
"detail": "high"
}
}
]
}
],
max_tokens=1500
)
# Parse the response as JSON
analysis_text = response.choices[0].message.content
# Remove markdown code blocks if present
if analysis_text.startswith("```"):
analysis_text = analysis_text.split("```")[1]
if analysis_text.startswith("json"):
analysis_text = analysis_text[4:]
analysis = json.loads(analysis_text.strip())
result = {
"success": True,
"filename": filename,
"analysis": analysis,
"metadata": {
"model_used": "gpt-4o",
"detail_level": "high",
"token_usage": {
"prompt": response.usage.prompt_tokens,
"completion": response.usage.completion_tokens
}
},
"warnings": []
}
except json.JSONDecodeError as e:
result = {
"success": False,
"filename": filename,
"analysis": {"raw_response": response.choices[0].message.content},
"metadata": {"model_used": "gpt-4o"},
"warnings": [f"Failed to parse JSON: {str(e)}"]
}
except Exception as e:
result = {
"success": False,
"filename": filename,
"analysis": {},
"metadata": {},
"warnings": [f"Analysis failed: {str(e)}"]
}
return result
# Usage
result = analyze_image_to_json("photo.jpg")
print(json.dumps(result, indent=2))
Batch Processing Directory
from openai import OpenAI
import base64
import json
from pathlib import Path
client = OpenAI()
def process_image_directory(directory_path, output_file, prompt=None):
"""Process all images in a directory and save results."""
if prompt is None:
prompt = "Describe this image briefly, including any visible text."
image_extensions = {'.jpg', '.jpeg', '.png', '.webp', '.gif'}
results = []
for file_path in sorted(Path(directory_path).iterdir()):
if file_path.suffix.lower() in image_extensions:
print(f"Processing: {file_path.name}")
try:
analysis = analyze_local_image(str(file_path), prompt)
results.append({
"filename": file_path.name,
"success": True,
"analysis": analysis
})
except Exception as e:
results.append({
"filename": file_path.name,
"success": False,
"error": str(e)
})
# Save results
with open(output_file, 'w') as f:
json.dump(results, f, indent=2)
return results
Detail Level Configuration
The detail parameter controls image resolution and token usage:
# Low detail: 512x512 fixed, ~85 tokens per image
# Best for: Quick summaries, dominant colors, general scene type
{"detail": "low"}
# High detail: Full resolution processing
# Best for: Reading text, detecting small objects, detailed analysis
{"detail": "high"}
# Auto: Model decides based on image size
# Best for: General use when cost vs quality tradeoff is acceptable
{"detail": "auto"}
Choosing Detail Level
def get_recommended_detail(task_type):
"""Recommend detail level based on task type."""
high_detail_tasks = {
'ocr', 'text_extraction', 'document_analysis',
'small_object_detection', 'detailed_comparison',
'fine_grained_analysis'
}
low_detail_tasks = {
'scene_classification', 'dominant_colors',
'general_description', 'thumbnail_preview'
}
if task_type.lower() in high_detail_tasks:
return "high"
elif task_type.lower() in low_detail_tasks:
return "low"
else:
return "auto"
Text Extraction (Vision-based OCR)
For extracting text from images using vision models:
def extract_text_from_image(image_path, preserve_layout=False):
"""Extract text from an image using GPT-4o vision."""
if preserve_layout:
prompt = (
"Extract ALL text visible in this image. "
"Preserve the original layout and formatting as much as possible. "
"Include headers, paragraphs, captions, and any other text. "
"Return only the extracted text, nothing else."
)
else:
prompt = (
"Extract all text visible in this image. "
"Return the text in reading order (top to bottom, left to right). "
"Return only the extracted text, nothing else."
)
return analyze_local_image(image_path, prompt)
def extract_structured_text(image_path):
"""Extract text with structure information as JSON."""
prompt = """Extract all text from this image and return as JSON:
{
"headers": ["list of headers/titles"],
"paragraphs": ["list of paragraph texts"],
"labels": ["list of labels or captions"],
"other_text": ["any other text elements"],
"reading_order": ["all text in reading order"]
}
Return ONLY valid JSON."""
response = analyze_local_image(image_path, prompt)
try:
# Clean and parse JSON
if response.startswith("```"):
response = response.split("```")[1]
if response.startswith("json"):
response = response[4:]
return json.loads(response.strip())
except json.JSONDecodeError:
return {"raw_text": response, "parse_error": True}
Error Handling
Common Issues and Solutions
Issue: API rate limits exceeded
import time
from openai import RateLimitError
def analyze_with_retry(image_path, prompt, max_retries=3):
"""Analyze image with exponential backoff retry."""
for attempt in range(max_retries):
try:
return analyze_local_image(image_path, prompt)
except RateLimitError:
if attempt < max_retries - 1:
wait_time = 2 ** attempt # Exponential backoff
print(f"Rate limited, waiting {wait_time}s...")
time.sleep(wait_time)
else:
raise
Issue: Image too large
from PIL import Image
import io
def resize_image_if_needed(image_path, max_size_mb=15):
"""Resize image if it exceeds size limit."""
file_size_mb = os.path.getsize(image_path) / (1024 * 1024)
if file_size_mb <= max_size_mb:
return encode_image_to_base64(image_path)
# Resize the image
img = Image.open(image_path)
# Calculate new dimensions (reduce by 50% iteratively)
while file_size_mb > max_size_mb:
new_width = img.width // 2
new_height = img.height // 2
img = img.resize((new_width, new_height), Image.Resampling.LANCZOS)
# Check new size
buffer = io.BytesIO()
img.save(buffer, format='JPEG', quality=85)
file_size_mb = len(buffer.getvalue()) / (1024 * 1024)
# Encode resized image
buffer = io.BytesIO()
img.save(buffer, format='JPEG', quality=85)
return base64.standard_b64encode(buffer.getvalue()).decode("utf-8")
Issue: Invalid image format
def validate_image(image_path):
"""Validate image before processing."""
valid_extensions = {'.jpg', '.jpeg', '.png', '.gif', '.webp'}
path = Path(image_path)
if not path.exists():
return False, "File does not exist"
if path.suffix.lower() not in valid_extensions:
return False, f"Unsupported format: {path.suffix}"
try:
with Image.open(image_path) as img:
img.verify()
return True, "Valid image"
except Exception as e:
return False, f"Invalid image: {str(e)}"
Quality Self-Check
Before returning results, verify:
- API response was received successfully
- Output is valid JSON (if structured output requested)
- All requested analysis fields are present
- Token usage is within expected bounds
- No error messages in the response
- For multi-image: all images were processed
- Confidence/warnings are included when analysis is uncertain
Limitations
- Medical imagery: Not suitable for diagnostic analysis of CT scans, X-rays, or MRIs
- Small text: Text smaller than ~12pt may be misread; use
detail: high - Rotated/skewed text: Accuracy decreases with text rotation; pre-process if needed
- Non-Latin scripts: Lower accuracy for complex scripts (CJK, Arabic, etc.)
- Object counting: Approximate counts only; may miss or double-count similar objects
- Spatial precision: Cannot provide pixel-accurate bounding boxes or measurements
- Panoramic/fisheye: Distorted images reduce analysis accuracy
- Graphs and charts: May misinterpret line styles, legends, or data points
- Metadata: Cannot access EXIF data, camera info, or GPS coordinates from images
- Cost: High-detail analysis of many images can be expensive; monitor token usage
Token Cost Estimation
Approximate token costs for image inputs:
| Detail Level | Tokens per Image | Best For |
|---|---|---|
| low | ~85 tokens (fixed) | Quick classification, color detection |
| high | 85 + 170 per 512x512 tile | OCR, detailed analysis, small objects |
| auto | Variable | General use |
def estimate_image_tokens(image_path, detail="high"):
"""Estimate token usage for an image."""
if detail == "low":
return 85
with Image.open(image_path) as img:
width, height = img.size
# High detail: image is scaled to fit in 2048x2048, then tiled at 512x512
scale = min(2048 / max(width, height), 1.0)
scaled_width = int(width * scale)
scaled_height = int(height * scale)
# Ensure minimum 768 on shortest side
if min(scaled_width, scaled_height) < 768:
scale = 768 / min(scaled_width, scaled_height)
scaled_width = int(scaled_width * scale)
scaled_height = int(scaled_height * scale)
# Calculate tiles
tiles_x = (scaled_width + 511) // 512
tiles_y = (scaled_height + 511) // 512
total_tiles = tiles_x * tiles_y
return 85 + (170 * total_tiles)
Version History
- 1.0.0 (2026-01-21): Initial release with GPT-4o vision support
Recommended Agent Skills
Expand your agent's capabilities with these related and highly-rated skills.
csv-processing
Use this skill when reading sensor data from CSV files, writing simulation results to CSV, processing time-series data with pandas, or handling missing values in datasets.
pid-controller
Use this skill when implementing PID control loops for adaptive cruise control, vehicle speed regulation, throttle/brake management, or any feedback control system requiring proportional-integral-derivative control.
yaml-config
Use this skill when reading or writing YAML configuration files, loading vehicle parameters, or handling config file parsing with proper error handling.
simulation-metrics
Use this skill when calculating control system performance metrics such as rise time, overshoot percentage, steady-state error, or settling time for evaluating simulation results.
vehicle-dynamics
Use this skill when simulating vehicle motion, calculating safe following distances, time-to-collision, speed/position updates, or implementing vehicle state machines for cruise control modes.
web-interface-guidelines
Vercel's comprehensive UI guidelines for building accessible, performant web interfaces. Use this skill when reviewing or building UI components for compliance with best practices around accessibility, performance, animations, and visual stability.
Didn't find tool you were looking for?