You have a stack of invoices to archive, a training video to summarize, an audio recording of a meeting to transcribe. Until now, that meant three different tools, three subscriptions, three integrations to maintain. With Gemini multimodal, a single model handles all of it. For a small business, that means fewer costs and less complexity.
At Meteora Web, we think in numbers: every extra tool is a hidden cost. A model that handles text, images, audio, and video removes integrations and reduces errors. Let's see how it actually works, with examples you can try today.
How does image analysis work with Gemini?
We're not talking about recognizing a cat in a photo. Gemini analyzes visual content contextually: it reads diagrams, extracts text from screenshots, understands charts and tables. For a business, this means automating processes that previously required a human eye.
The classic example is extracting data from invoices or documents. Send an image, Gemini returns structured fields. But it also works with dashboard screenshots, product photos for e-commerce, even technical drawings.
Extracting data from an image with the API
With the Google AI SDK, a few lines of code are enough to send an image and receive a structured analysis. Here's a Python example:
Sponsored Protocol
from google import genai
from google.genai import types
client = genai.Client(api_key="YOUR_API_KEY")
# Upload a local image
image_file = client.files.upload(file="invoice.png")
response = client.models.generate_content(
model="gemini-2.5-flash",
contents=[
"Extract from this invoice: number, date, total amount, VAT ID. Respond in JSON format.",
image_file
]
)
print(response.text)The model returns clean JSON, ready to be inserted into an ERP or a spreadsheet. Zero OCR to configure, zero templates to maintain.
Beware of common mistakes: don't ask Gemini to "describe" the image if you need a specific data point. Be explicit about the output format. The prompt determines the quality of the result.
How does Gemini handle audio and transcription?
Transcription is not the real value. The value is understanding the content: extracting key points from a meeting, identifying action items, classifying the sentiment of a support call. Gemini does this directly from audio, without going through a separate transcription engine.
Sponsored Protocol
For a business managing customer support, this changes everything. Send a recording, receive a summary with the issues raised and the solutions proposed. No more listening to hours of calls to understand what's broken.
Transcribing and summarizing an audio file
The API accepts audio files in various formats. Here's how to turn a recording into an operational summary:
from google import genai
client = genai.Client(api_key="YOUR_API_KEY")
# Upload the audio file
audio_file = client.files.upload(file="meeting.mp3")
response = client.models.generate_content(
model="gemini-2.5-flash",
contents=[
"Transcribe this meeting and then list: decisions made, assigned actions, deadlines. Use bullet points.",
audio_file
]
)
print(response.text)Common mistake: sending hours of audio without clear instructions. Gemini handles the length, but the result is better if you specify what you care about: "list only customer objections" or "summarize in 5 points".
How do you analyze a video with Gemini without watching it?
Video is the most powerful and least exploited use case. Gemini can watch a video and answer questions about its content: what happens, what is said, which objects appear. No need to extract frames one by one — the model processes the video directly.
Sponsored Protocol
Imagine having hours of footage from a store or a production facility. Gemini can identify anomalies, count customers, verify that procedures are followed. For the retail sector, which we know well from managing a clothing ERP, this means customer behavior analysis without expensive smart cameras.
Video analysis with the Gemini API
The process is similar to other media, but with a difference: you can ask for temporal analysis.
from google import genai
client = genai.Client(api_key="YOUR_API_KEY")
# Upload the video
video_file = client.files.upload(file="store_camera.mp4")
response = client.models.generate_content(
model="gemini-2.5-flash",
contents=[
"Analyze this retail store video. Count customers entering, identify checkout queue moments, and flag if any shelf remains empty for more than a minute. Provide a timeline.",
video_file
]
)
print(response.text)Technical note: for long videos, upload the file with client.files.upload and pass the reference. Don't convert the video to base64 in the prompt: the model has a token limit and the file must be referenced correctly.
Sponsored Protocol
Which formats and sizes does Gemini support for media?
The right question is not "which formats", but "how to prepare files to avoid wasting tokens". Every API request has a cost. Sending a 10MB image when 500KB would suffice is money wasted. We've seen it with e-commerce clients: optimized files reduced API costs by 60%.
For images, common formats like PNG, JPEG, and WEBP work well. For audio, MP3 and WAV. For video, MP4. But size matters more than format. Resize images to 1024px on the long side. For audio, trim silences. For video, reduce resolution if detail isn't needed.
Optimizing files to reduce API costs
Here's an operational checklist before sending any media to Gemini:
- Resize images to a maximum of 1024px on the long side
- Convert to JPEG or WEBP with 80% compression
- For audio, use MP3 at 128kbps — sufficient quality for voice
- For videos, reduce to 720p if full definition isn't needed
- Split very long files into 10-15 minute segments for more precise responses
This isn't just saving money: smaller files are processed faster and responses are more accurate because the model doesn't have to handle visual or audio noise.
Sponsored Protocol
What to do now to start with Gemini multimodal
You don't need a complex project to understand the value. Start with a single use case and measure the time saved. Today's concrete actions:
- Try image analysis on a real document: an invoice, an ERP screenshot, a warehouse photo
- Test transcription on a recording of a meeting or a support call
- Upload a short video (max 2 minutes) and ask for a summary with timestamps
- Compare the time spent with your current manual method
- Estimate cost per volume: multiply the cost per request by the number of documents you process in a month
If the use case holds up to economic comparison, then it's worth building an automated flow. If it doesn't, you've saved time and money. At Meteora Web, we build custom integrations for SMEs that want to automate without depending on lifetime subscriptions. Technology must bend to your numbers, not the other way around. For a deeper look at the entire Gemini ecosystem, start with our main guide on Google Gemini models, APIs, and integrations.
