Duolingo describe the image: what the scorer compares your answer to
Aman Batth · 9 min read · · Updated
Every guide to the Duolingo photo tasks gives you a four-part template and a pile of sample answers. None of them explain what the machine does with your words — which is why a short or generic response does not merely score lower, it collapses three marks at once.
The Duolingo English Test shows you a photograph twice, in two different tasks. Write About the Photo gives you one minute to type a description. Speak About the Photo gives you a short look and then asks you to describe it aloud until the timer stops.
Search for advice on either and you will find the same thing: a four-part template — subject, setting, emotions, extra details — followed by several dozen worked examples. That advice is not wrong. It is just answering a question nobody is stuck on. Candidates do not fail this task because they cannot think of a sentence shape. They fail it because of what the scorer does next, and no template guide tells you that part.
This is written from the marking side — we build the engine that scores these tasks on Hilingo. For the current task list and timings themselves, the source is the Duolingo English Test, and those are set by the test owner and change; a blog post is not where to confirm them.
Your answer is not compared to the photo
Here is the mechanism that reframes everything else.
An automated scorer cannot look at a picture and admire your observation. To judge whether you described this image rather than any image, it has to compare what you produced against some stored representation of what is in the frame. Duolingo does not publish its scoring internals, so nobody outside the company can tell you the algorithm — but we build the engine that scores speaking and writing on Hilingo, and any automated scorer of this task has to work on overlap of that kind, because there is no other way to check that an answer belongs to a particular photograph.
That single fact decides what earns marks:
- "A woman in a yellow jacket is checking a timetable at a bus stop" overlaps heavily. Woman, yellow, jacket, timetable, bus stop are all in the description.
- "This picture shows an interesting everyday scene and the atmosphere feels quite pleasant" overlaps with nothing. It is grammatically clean, it sounds like English, and as content it is close to empty.
Every piece of advice further down this page falls out of that. Name things. Concrete nouns, visible colours, what people are physically doing. A clever inference about how someone feels is worth less than the object they are holding, because the object is in the description and the inference is not.
The expensive part: you do not lose one mark, you lose three
This is the bit that never appears in template guides, and it is the reason short answers are catastrophic rather than merely weak.
Duolingo reports Speak About the Photo against content, pronunciation and fluency. You would assume these are independent — that a thin answer scores badly on content and keeps whatever pronunciation and fluency it earned. It does not work that way.
Our engine, modelled on the real test's behaviour, applies an off-topic collapse. When content comes back near zero and the response is either very short or shares almost no content words with the image description, pronunciation and fluency are not scored on their merits. They are forced to the floor with it.
So the failure mode is not "I described it a bit vaguely and scored a little lower". It is that a fifteen-word generic answer takes your pronunciation score down with it, even if every word you said was pronounced perfectly.
Two practical floors follow. These are our engine's thresholds, not a published Duolingo rule — we are describing where our own scorer stops crediting an answer, because that is the part we can actually show you:
- Get well past fifteen words. A floor, not a target — it is the length below which an answer starts to read as a non-answer.
- Name at least two things that are actually in the picture. Two shared content words is the difference between "described this image badly" and "described no image at all".
Silence is the worst possible answer for exactly this reason. There is nothing to assess, so nothing can be credited.
Why a memorised template can trigger the same collapse
The scorer also watches for answers that are the right length but carry no new information — a response whose content words are almost entirely lifted from the prompt rather than produced by you.
A fully memorised photo template is vulnerable to this in a way its users do not expect. If your opening thirty words are "In this image I can see a scene which appears to be taking place in a particular setting, and overall the atmosphere is quite positive", you have produced fluent English that describes nothing. To a scorer measuring overlap with a specific image, that is indistinguishable from an off-topic answer.
Use a template as scaffolding for word order if it stops you freezing. Do not let it eat your time. The template is the container; the specific nouns are the content, and only the content is scored.
Both photo tasks land on the same subscore
The DET does not report Reading, Listening, Writing and Speaking. It reports four integrated subscores, and every task feeds two of them. Here is where these two land:
| Task | Section | Feeds |
|---|---|---|
| Write About the Photo | Writing | Literacy, Production |
| Speak About the Photo | Speaking | Conversation, Production |
Production is speaking plus writing — pure output, nothing to recognise or select. It is the subscore both photo tasks push on, and the one candidates most often find sitting below the others on their report.
If your Production is the low number, the photo tasks are a direct lever on it. If your Comprehension is the low number, they are not, and drilling fifty-five caption examples will not move it. We cover how to read the four together in Duolingo English Test scores explained.
The caption is the lightest task you will write
One more thing the example-bank guides get backwards.
In our engine, following the official model, Write About the Photo carries a single weight while the Writing Sample and Interactive Writing each carry double. The same holds on the speaking side: the Speaking Sample and Interactive Speaking outweigh the photo description.
The one-minute caption is the cheapest writing task on the test. It is worth getting reliably competent at, which takes about an afternoon. It is not worth a week, and it is certainly not worth memorising dozens of model answers when the two double-weighted tasks are sitting untouched — and those are the ones institutions can actually read and listen to.
Get the photo tasks to "never collapses, always names four concrete things". Then leave them alone.
Adaptive difficulty changes the pacing, not the strategy
The first section of the DET adapts: answer well and the next items get harder, and harder items are worth more. Two consequences for these tasks.
There is no fixed number of items to survive, so you cannot pace by counting questions. Pace by the on-screen timer only.
And keep producing until that timer stops. On a scorer that rewards sustained, relevant output, stopping early is a decision to be marked on less evidence. There is no elegance bonus for finishing neatly at forty seconds.
What to actually do
- Spend your first two seconds counting, not composing. Find four concrete things: a person, an object, a place, an action.
- Open with the most specific one. Not "this is a photo of" — go straight to "a man in a blue apron is arranging bread on a counter".
- Colours, clothing, objects, positions. These are the words most likely to be in the description you are being matched against.
- Foreground first, then background. It is a reliable way to keep generating specific detail when you run dry.
- Keep going. Past fifteen words is survival; keep talking or typing to the end is the score.
- Judgements last, if at all. "They look happy" is a fine closing sentence and a terrible opening one.
Check it against a scorer, not a friend
The one thing you cannot do for yourself here is tell whether your answer cleared the content threshold, because your description always makes sense to you — you were looking at the image when you wrote it.
Hilingo gives you one full scored DET mock free, with no card. We run 15 Duolingo English Test mocks in total. Both photo tasks come back with content scored against the image rather than as a general impression, and Speak About the Photo returns a per-word pronunciation score, your pause count and a transcript — so you can see whether a low mark came from your English or from producing too little to assess. Your result is reported on the DET's own four integrated subscores, and the AI teacher opens with your weakest one rather than waiting to be asked.
- Duolingo English Test scores explained — the 10–160 scale and the four subscores
- DET score chart — the scale next to IELTS, TOEFL and CEFR
- Best Duolingo English Test practice platforms — what to look for in feedback
- PTE Describe Image: a template that works — the same task on a different scorer
- Universities that accept the DET
Frequently asked questions
How many sentences should I write for Duolingo Write About the Photo?
Write as many as you can complete correctly in the minute, and treat length as a floor rather than a target. A very short answer risks being treated as a non-answer, which is far more expensive than a slightly rushed longer one. Three to five full sentences that each name something visible is a sensible working aim.
Do I get marked down for describing the photo wrongly?
Describing something that is not there earns nothing, but it is not punished the way saying almost nothing is. Content is measured as overlap with a description of the image, so a wrong guess simply fails to match. The costly outcome is an answer so short or so generic that it matches nothing at all.
Should I use a template for the Duolingo photo questions?
Use one only to stop yourself freezing on the sentence order. A template made of general phrases that would fit any image contributes nothing to the content score and can make a short answer look like an off-topic one. Fill the structure with specific nouns from the picture in front of you.
Which Duolingo subscore do the photo tasks affect?
Write About the Photo feeds Literacy and Production. Speak About the Photo feeds Conversation and Production. Production is the subscore both of them push on, so if Production is your weakest number, these tasks are a direct lever on it.
Is Speak About the Photo more important than the Speaking Sample?
No. The Speaking Sample and Interactive Speaking are weighted more heavily than the photo description, and the Speaking Sample is shared with institutions, so an admissions officer can listen to it. Get the photo tasks reliably competent, then spend your remaining preparation on the longer tasks.
What happens if I say nothing on Speak About the Photo?
Silence is the worst available outcome, because there is nothing for the scorer to assess and therefore nothing it can credit. A halting, imperfect description of four things you can see will outscore a polished sentence followed by a long pause every time.