How to do it, step by step
- Choose the photos first A caption is written to a set, not the other way round. Cut the carousel to eight to twelve photos before you write a word, or you will end up captioning the wrong thing.
- Upload the photo, do not describe it ChatGPT and Gemini both accept the image. Giving the model the actual frame produces a far more specific caption than typing "beach photo with friends".
- Give it constraints, not a request Ask for ten options under eight words, no emoji, no hashtags, no rhetorical questions, and ban the words that mark a caption as generated.
- Paste three captions you have actually posted Tell it to match that voice, including your lowercase and your punctuation habits. This one step does more for the result than switching apps ever will.
- Reject the first batch and ask for drier The first set is the average of everything ever captioned. Ask for something drier and more specific to what is in the photo, and the second set is usually usable.
- Cut the winner in half before you post Take the caption you liked and delete half of it. Almost every generated caption improves when the explanation at the end is removed.
The five caption writers, ranked
Ranked on the only thing that matters, which is whether you would actually post what comes out.
- ChatGPT. Free tier, accepts the photo itself, and will give you ten options in one go and rewrite them all shorter when you ask. Verdict: the best caption writer available, and the one every other tool on this list is quietly competing with. Its weakness is a default voice that is upbeat, tidy and slightly American.
- Canva Magic Write. Built into the editor, so the caption arrives next to the post you are already making. Verdict: convenience rather than quality. Choose it if you are designing the graphic in Canva anyway, not for the writing.
- Picsart. A caption generator inside a full editing app, which suits people who edit and post in the same place. Verdict: fine, and no better than the free chatbot at the actual sentence. Some AI features consume credits or leave a watermark, so check before you rely on it.
- Copy.ai. Built for marketing copy, so it does volume: a month of captions in a batch, in a defined tone. Verdict: overkill for one holiday carousel, genuinely useful if you run an account and plan a grid.
- Gemini. The same shape as ChatGPT and just as capable on a photo. Verdict: worth trying purely because the two default voices differ, and one of them may be closer to yours.
ShutterAI writes captions in your voice from inside the camera, learned from how you actually write. It is free, and it is the narrowest tool here, which is covered honestly below.
The prompt that makes a chatbot caption usable
Most people get bad captions from ChatGPT because they ask for a caption. Ask for constraints instead, and the output changes completely.
Give it four things. A length: "under eight words" prevents the paragraph with a moral at the end. A ban list: no emoji, no hashtags, no rhetorical questions, no words like "vibes", "core memory", "unforgettable" or "living my best life". Examples of you: paste three captions you have actually posted and say "match this voice, including the lowercase and the lack of punctuation". A quantity: ask for ten, because the ninth is usually the one.
Then reject the first set on purpose. The first thing any model produces is the average of everything ever captioned. Saying "all of these sound like an ad, be drier and more specific to what is in the photo" is the single most effective follow-up you can send.
Why AI captions sound like AI
There is a recognisable register, and once you can name it you stop posting it.
- It summarises the photo you have already posted. "A beautiful day at the beach with my favourite people" tells nobody anything they cannot see. Good captions add the thing the photo does not show.
- It reaches for a conclusion. Real captions rarely resolve. They are a fragment, an in-joke, or a line someone said that day.
- It is uniformly warm. No model volunteers "worst coffee in Lisbon, would return", and that is exactly the sort of line people reply to.
- The punctuation is too correct. Most people caption in lowercase with no full stop. A perfectly punctuated sentence reads as written by something else.
The fix is the same in every app: feed it your own old captions as examples, and cut whatever it gives you by half.
What none of these apps can do
Worth knowing before you install four of them.
- None of them know what happened. A model sees a table and four glasses. It does not know whose birthday it was or what got said, and that is where every good caption comes from.
- None of them fix a weak set of photos. If the carousel is nine near-identical shots, no sentence rescues it. Cut it to ten first.
- Hashtag suggestions are mostly noise. Generated hashtag blocks tend to be generic and high-competition, which is the worst combination there is.
- None of them are a camera. Picsart and Canva are editors with a text feature bolted on, so the caption still arrives after the post is built rather than while you are shooting.
Captions in your voice, and the honest caveat
Our own app, so read this as disclosure. ShutterAI is a camera with film looks that render live on photo and video, a glow that keeps freckles and texture, one-tap removal of strangers and a photo dump that curates itself. Captions in your voice is the part relevant here: because the captions are written where the set is chosen, they are shaped by how you write rather than by how the internet writes, which is the exact failure mode of every generator above.
The caveats matter, and there are two. Captions needs Apple Intelligence, so an iPhone 15 Pro or newer on iOS 26, which is a narrower gate than the rest of the app. And it writes to the set in front of it, one post at a time, so it is not the tool for a month of captions planned in a spreadsheet. ChatGPT with the prompt in the section above still wins on range, Copy.ai on volume, and our own written lists if you would rather not generate anything at all.
This page, and our caption lists
This page is about which app writes the caption. If you would rather pick one than generate one, photo dump captions is a written list of more than 150 captions sorted by occasion, with none of the generated register described above. Read that one when you want to copy a line and post in the next two minutes.
Questions people ask
What app writes Instagram captions for your photos?
ChatGPT is the most capable and has a free tier: upload the photo and ask for ten captions under eight words. Canva's Magic Write and Picsart generate captions inside the app you are editing in, and Copy.ai is built for writing a month of them at once.
Is there a free app that writes Instagram captions?
Yes. ChatGPT and Gemini both write captions from an uploaded photo on their free tiers, and Canva's free plan includes a limited amount of Magic Write. Dedicated caption apps almost all cap the free tier at a few generations a day.
How do I stop my AI captions from sounding like AI?
Paste three captions you have actually posted and tell the model to match that voice, ask for under eight words, and ban emoji, hashtags and words like vibes and core memory. Then delete half of whatever it gives you.
Can ChatGPT write captions from a photo?
Yes. It reads an uploaded image and captions what is in it, which produces something far more specific than describing the photo in words. Ask for ten options in one go, because the last few are usually better than the first.
Do caption generators help with hashtags?
Not much. Generated hashtag blocks tend to be generic and highly competitive at the same time, which is the worst combination. A few specific tags you have chosen yourself do more.
Is it better to use a caption app or a caption list?
A list if you want to post in the next two minutes, because you are choosing rather than editing. A generator if the caption has to be about your specific photo, in which case give it the image and your own examples.