A night-sky photo accused of being AI: what photographers should keep

Flat illustration of an abstract night sky with curved satellite trails on a navy card, beside a clock and a teal check-mark badge. No people, text or logos.

Short answer: In early November 2026 The Guardian reported that New Zealand astrophotographer Brendan Larsen was fighting accusations that a viral night-sky image over Mount Taranaki had been made with AI. Larsen told the paper he shot 578 eight-second frames over about two hours in September, then spent about 45 hours in Photoshop building a composite of some 1,400 satellites and 14,000 trails, and that he filmed the process because he expected the accusations. If your own photograph is called fake, what helps is not an argument in the comments: it is the camera originals, the sequence, and a dated copy of the finished file. A TimeBind cannot prove a picture is a photograph. It can date the exact files you kept.

Last checked 12 November 2026. General information for photographers, not legal advice and not a finding about any image. The account of Brendan Larsen’s photograph follows The Guardian’s report of 4 November 2026. A dated file does not prove that a photograph is “not AI,” and TimeBinder does not detect AI-generated images.

Table of Contents

What was reported

Eva Corlett’s report in The Guardian, published on 4 November 2026 and updated on 11 November, describes a composite photograph of the night sky over Mount Taranaki, on New Zealand’s North Island, so dense with satellite trails that viewers accused Larsen of using AI. Larsen, 51, told the paper he set his camera to take an unlimited number of eight-second exposures with a two-second gap, left it running for two hours, and “captured 578 photos.” That, he said, “was the easy part.” He then spent about 45 hours layering the frames in Adobe Photoshop, including time spent erasing stars so the satellite trails would show. The finished image, as the paper describes it, shows about 1,400 satellites and 14,000 trails.

Larsen uploaded the image to his social channels in mid-October. The Guardian says it was viewed tens of millions of times, gained him 15,000 new followers, and drew thousands of comments. Some viewers dismissed it as AI. Larsen said he understood why: generating software is becoming hard to tell from real photographs, “especially if you’re not familiar with the location or with astrophotography.” He also said it was frustrating to see the image downloaded and recreated by AI, with obvious flaws.

Importantly for anyone in the same position, he had already documented the process on video, because he was expecting the accusations. The Guardian published a still of him and a timelapse of the photographs. That is not a legal test, and a process video can be questioned too. It is still a lot more than a finished JPEG on its own.

The story travelled quickly. On 5 November 2026 Olhar Digital posted a YouTube report that had 21,204 views when checked on 12 November. An 18 October X post by @Rainmaker1973 had 138,217 impressions when the same metrics were read on 12 November. A 10 November post by @terrinakamura, inside the last 14 days, had 1,789 impressions. Google autocomplete on 12 November did not suggest the photographer’s name. The demand is the coverage and the traffic, not a search cluster.

Why a real photograph gets called AI

Stacked or composited photographs often look unlike a single glance at the sky. Larsen’s image is a record of many short exposures, not a claim that a person standing on the mountain would have seen 14,000 trails at once. People who do not shoot the night sky can reasonably find that surprising. Meanwhile, generated images of landscapes have become common, and viewers have learned to ask. Those two facts together explain the comments. They do not decide whether any particular file is a photograph.

Two other limits matter. First, a process video, a set of RAW files or a dated export can support your account of how an image was made. None of them is a detector, and none of them binds a social platform or a contest jury. Second, Larsen also told the paper he has seen his image remade by AI. A later generated lookalike does not erase the earlier photographic work, but it can confuse anyone who only has a screenshot. That is a reason to keep the originals, not a reason to argue in every thread.

This page is not about whether satellite numbers in the night sky are a good or a bad thing. Larsen said the comments on his Facebook page were often both admiring and alarmed. The practical question for photographers is narrower: if you publish a surprising picture, what can you later show about the file?

Practical checklist if your photograph is called AI

  1. Keep the camera originals, not only the export. For a night-sky stack that means the RAW or original frames (in Larsen’s telling, 578 of them), not just the finished JPEG or a social upload. See how to prove a RAW photo existed before a date.
  2. Keep the sequence. File names, timestamps, a folder that still has every frame in order, and, if you use it, a sidecars or catalogue file, all help show that the export came from a shoot, not from a single generated still. For a series, see how to timestamp a photo series before you publish it.
  3. Keep a copy of the finished composite as it was published. If you later re-export or re-edit, the file that went public may no longer match. Date the version you actually posted.
  4. If you have a process video or a timelapse of the shoot, keep that too. Larsen filmed his process because he expected the accusation. A video can be questioned, and it is not required, but it is easier to keep now than to reconstruct later. It does not “prove a photo is not AI” on its own.
  5. Capture the accusation and the post, with dates. A screenshot of the published image, the URL, and the comments or emails you are answering can matter if the dispute moves off the thread.
  6. Do not lean on a detector score. Tools that look for a SynthID, a C2PA manifest or a “likely AI” label can be useful clues, and they can also miss a photograph or misread an edit. They do not replace the originals. For what Content Credentials show and do not show, see the C2PA field guide and the free Content Credentials checker.
  7. Date the folder or the key files. A TimeBind of the RAW set, a hash manifest of the folder, or the published composite lets you show that those exact files existed no later than a recorded time and still match. For the broader method, see how to prove a photo is not AI — the honest version of that title is: show the original, the sequence and the date, not a detection result.
  8. Reply calmly and point to the files, or do not reply. You are not obliged to convince every comment. If a contest, a client or a publication asks, offer the originals and the dated export rather than a thread of screenshots.

What a TimeBind does and does not show

A TimeBind is a dated, checkable record that an exact file existed no later than a recorded time and still matches. The SHA-256 fingerprint is made in your browser and the file is never uploaded. The record is anchored to Bitcoin (a Merkle root in an OP_RETURN). Your first Pooled TimeBind is free; after that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. See how it works.

Here it is useful for the folder of original frames, a hash manifest of that folder, the published composite, and any process video you kept. It shows that those exact files existed no later than the recorded time and still match.

A TimeBind does not detect AI, does not read a SynthID or a C2PA manifest, does not prove that you stood on a mountain, does not prove that a composite is a “single photograph,” and does not create or prove copyright or ownership. It is one piece of a timeline. If someone has run your picture through a generator, that is a different problem; see what a German appeal court said when someone ran a photo through AI.

Limits

We have not inspected Larsen’s files and we are not authenticating his image. The facts above are taken from The Guardian’s report and from dated public posts and a YouTube news clip. Other outlets may have summarised the same story with different figures; where they differ, the Guardian article is the source used here. Contest rules, platform rules and defamation law differ by country. A dated RAW folder will not, on its own, force a platform to restore a post or a jury to reverse a decision.

Tools that can help

Use the bulk SHA-256 manifest for a folder of frames, the Content Credentials checker if you want to see whether a file already carries a C2PA manifest, and how to check when a photo was taken for what EXIF and other metadata can and cannot show.

Frequently Asked Questions

  1. What did Brendan Larsen say he photographed?

    According to The Guardian on 4 November 2026, he photographed the night sky over Mount Taranaki in September, taking 578 eight-second frames over about two hours, then spent about 45 hours in Photoshop on a composite that shows about 1,400 satellites and 14,000 trails. He said he filmed the process because he expected AI accusations.

  2. Does a process video prove a photograph is not AI?

    No. It can support your account of how an image was made, and Larsen made one for that reason. Process footage can also be questioned. Keep it with the camera originals and a dated copy of the published file, and do not treat any single item as a detector.

  3. What should I keep if my photo is accused of being AI?

    Keep the camera originals, the sequence, the exact export you published, any process video, and a capture of the post and the accusation, all with dates. A dated record of those files can later show they existed by a recorded time and still match.

  4. Can a TimeBind prove my photo is not AI?

    No. A TimeBind does not detect AI and does not prove a file is a photograph. It is a dated, checkable record that an exact file existed no later than a recorded time and still matches.

  5. Is this the same as the drone-portfolio story that was also in the news?

    No. A different, scheduled TimeBinder article covers a published drone photograph that was flagged as AI. Larsen’s case is a night-sky composite and a different photographer. The practical habit — keep originals, the sequence and a dated export — is similar.

Sources checked

  • The Guardian, Eva Corlett, “The night-sky image so incredible the NZ man behind it has been accused of using AI,” 4 November 2026 (last modified 11 November 2026), https://www.theguardian.com/world/2026/nov/04/viral-image-nz-night-sky-satellites-accused-of-being-ai-brendan-larsen (read 12 November 2026)
  • YouTube: Olhar Digital, “Foto com milhares de satélites no céu é confundida com imagem de IA,” 5 November 2026, 21,204 views when checked 12 November 2026, https://www.youtube.com/watch?v=eus2zyFTK-w
  • X: @terrinakamura, 10 November 2026 (1,789 impressions, 73 likes, 16 reposts). Supporting: @Rainmaker1973, 18 October 2026 (138,217 impressions, 1,954 likes, 326 reposts, 501 bookmarks); @konstructivizm, 23 and 25 October 2026 (public metrics read 12 November 2026)
  • Google autocomplete, 12 November 2026 (US): no suggestions returned for seeds including “Brendan Larsen night sky” and “astrophotography accused of AI”

Date the camera originals you keep yourself

Your first Pooled TimeBind is free, with no credit card required. After that, TimeBinds start from US$5, and an ASAP TimeBind is US$15. Click here to start, or see plans and pricing.

Related: VSCO says it will not train generative AI on your photos: what creators should keep.