When most people think about AI costs, they think about text — typing a question, getting an answer. But modern AI models accept images, audio, and even video as input. And the token costs for these different modalities aren't just different — they're wildly different.
Understanding how each input type gets tokenized helps you make smarter decisions about when to paste a screenshot versus describing something in words, and why some seemingly simple requests can be surprisingly expensive.
How different input types get tokenized
Text is the baseline. One token equals roughly 4 characters or ¾ of a word. A 500-word prompt is about 375 tokens. Text is the most token-efficient way to communicate with AI by a wide margin.
Images are where costs start to climb. AI models don't "see" images the way we do — they convert images into a grid of tokens. A typical screenshot or photo gets tokenized into roughly 765–1,590 tokens depending on resolution and the model's processing method. A high-resolution image can consume over 3,000 tokens. That single image costs as much as 1,000–4,000 words of text.
Audio models like those used for voice input tokenize speech differently. One minute of audio typically translates to around 1,500–2,000 tokens, though this varies by language and speech density. A quick voice note might seem easier than typing, but it often generates more tokens than a well-crafted text prompt would.
Video is the most expensive input type by far. Video is essentially a rapid sequence of images plus an audio track. Even a short 30-second clip can consume tens of thousands of tokens. Most developers avoid video input unless the visual information genuinely can't be conveyed any other way.
The real-world cost comparison
| Input Type | Typical Token Cost | Text Equivalent |
|---|---|---|
| 500 words of text | ~375 tokens | 375 tokens |
| 1 screenshot (1080p) | ~1,100 tokens | ~1,500 words |
| 1 high-res photo | ~1,600 tokens | ~2,100 words |
| 1 minute of audio | ~1,800 tokens | ~2,400 words |
| 30 seconds of video | ~15,000 tokens | ~20,000 words |
Key insight: A single pasted screenshot costs as many tokens as writing out a detailed paragraph describing the problem. In many cases, a well-written text description is both cheaper and gives the AI model better information to work with.
When images are worth the token cost
Despite the higher token cost, images are sometimes the right choice. UI bugs and layout issues are nearly impossible to describe in text with enough precision — a screenshot is worth the extra tokens. Error messages with stack traces that span multiple lines are faster to screenshot than to copy-paste and format. And diagrams or architecture drawings convey spatial relationships that text simply can't match.
The key is intentionality. Don't paste a full-screen screenshot when the relevant information is in one small corner. Crop aggressively — smaller images use fewer tokens. And when the information is text-based (like a code snippet), always copy the text rather than screenshotting it. The model processes text much more accurately than OCR'd text from an image.
Practical advice for developers
Default to text. It's the cheapest, most accurate input type. Describe what you see rather than showing it whenever possible.
Crop images aggressively. If you must use a screenshot, crop it to just the relevant area. A 200x200px crop uses far fewer tokens than a 1920x1080 full-screen capture.
Avoid video input for AI coding tasks. If you're tempted to record a screen capture of a bug, consider whether a screenshot plus a text description would serve just as well. It almost always does, at a fraction of the token cost.
Track what you're spending. Without visibility into your token usage by input type, you can't make informed decisions. Tools like Kontinuity show you exactly how many tokens each message consumes, so you can see the real cost difference between a text prompt and an image prompt.
See what your prompts actually cost
Kontinuity tracks token usage across Claude and ChatGPT in real time. Free during beta.
Try Free for 14 Days →