Deep dive · module 25 of 100
Multimodal AI
Work with text and image in a single flow.
After this lesson
You know when and how to have images analysed or generated.
In view · Working together towards one result
Text and image come together
Provide the relevant assignment, notes or accompanying text.
Add an authorised image and state what you want to have investigated.
Request a single structured elaboration and check both sources.
Part 1 of 3 · Text source
Your preference for less movement is on. Choose each part yourself.
Analysing images
Upload an image and ask for a description, errors or suggestions. Check the output. Describe what you want to know and which part of the image is important. First ask for visible characteristics and only then for interpretation. Have small or illegible text marked as uncertain; a plausible description is not yet a reliable observation.
Generating images
Describe style, atmosphere, colours and composition clearly. Check for brand and rights. Provide subject, composition, light, format, and purpose. Check details and be clear about synthetic origin where relevant. A generated product photo does not prove a real product property, and an illustration is not a customer reference.
Combining
Use text and image together, for example for presentations or social media. Ensure a caption aligns with what is actually visible in the image. Separately check whether the image, text, and conclusion support each other. For important use, also create an accessible description that explains the core message without the image.
New practice example
From explanation to a usable result.
Text and image can help together, but every input source has its own uncertainties.
Fictional practice situationFictional: an inventory list contains twelve chairs. In a photo, only eight chairs are visible; part of the space is out of frame.
View a possible solution
This is how it can be done
Note: the list mentions twelve, in the photo there are eight visible. That does not prove that four are missing. Check the space or ask for a full count.
Why this works
Different sources may show a different part of the situation. A discrepancy requires investigation, not an automatic conclusion.
Make it slightly more difficult
For every conclusion, indicate whether it is an observation, a source citation, or an interpretation.
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
Different sources may show a different part of the situation. A discrepancy requires investigation, not an automatic conclusion.
A step further
For every conclusion, indicate whether it is an observation, a source citation, or an interpretation.
A prompt to get you started
Describe this image in 3 points and provide 2 suggestions to make the message clearer. [image] Use only the provided fictitious facts. Make missing information visible and indicate what a human must check before use.
Apply it
A AI image analysis reads the small number on a product label as 8. In the photo, the label is blurry; it could also be a 3.
How do you use this analysis?
- Adopt 8 because that is what the model chooses.
- Use the average 5,5.
- Mark the number as uncertain and check the original label or a sharper source.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
What is an important point of attention with generated images?
- They are always free of rights.
- Check brand consistency and rights.
- You do not need to check them.
Extra practice
Describe a situation in which you would want to have an image analysed by AI, and why. Work with a fictitious situation and name one concrete control point before you would use the result.
Having a product photo analysed for clarity and suggestions for better lighting or composition. I compare the outcome with the confirmed input, leave unknown data blank, and adjust the result before use.
