Commons talk:Featured picture candidates

For procedural questions about featured pictures (FP) or featured picture candidates (FPC), please read the complete rules and image guidelines before posting here. Most of your questions have answers on those pages. For unfamiliar photography-related words and expressions used on FPC, also see Photography terms.

overwrite files

Hi. Is anyone else having problems overwriting files with a new version? The following file shows me the new version in the preview in the media viewer. File:Daucus-carota-Sitia-Crete-Greece.jpg. As soon as I click on the image to view it in full size, the old, previous version is displayed. The file history shows two versions: the old version and the current version. The file size and image dimensions are displayed correctly in the current version, but as soon as I click on it, the old version is displayed. Even when I download the image via the download link, the old version is downloaded and not the current one. I cleared my browser cache and also used a different browser. Any ideas? Petro Stelte (talk) 13:11, 13 May 2026 (UTC)

I had the same problem recently, but it’s working again now. Achim Lammerts • Syntaxys (talk) 13:27, 13 May 2026 (UTC)

Had the same issue. It was fixed the next day, but for any case where time to update matters (fixing misinformation in an image displayed on Wikipedia, for example), it's more than just a minor frustration. Might be worth opening a phabricator ticket at this point. — Rhododendrites ^talk |  13:38, 13 May 2026 (UTC)

Thanks for your replies. I guess I'll have to wait then. It's just too bad that I nominated this picture and no one can review the new version. Petro Stelte (talk) 13:48, 13 May 2026 (UTC)

The new version is now visible. Petro Stelte (talk) 02:59, 14 May 2026 (UTC)

Same problem today for me, I guess overwriting files now takes a few hours/days to be updated on the server side cache -- Giles Laurent (talk) 20:00, 17 May 2026 (UTC)

I noticed the same thing with my latest nomination from yesterday where I overwrote the file just an hour before nominating it with a new version where I removed some heavy backlight-CAs. Now it's gone, the latest version will be displayed. --Granada (talk) 13:44, 18 May 2026 (UTC)

It looks like a case of Wikimedia’s intermediate CDN cache serving an old copy of the file to certain network routes. Try do the test using Tor, Tor is exiting through a different IP route and is probably hitting a different CDN cache node, while my regular network is receiving an old copy from some Wikimedia cache Wilfredor (talk) 19:08, 22 May 2026 (UTC)

FPC helper scripts: feedback wanted

Hi all,

I made two helper scripts for FPC. (more info)

Script	What it does
fpc-voter.js	Adds quick buttons for Support, Oppose, Neutral, Question, Info, Comment, Request and Withdraw. It also adds your signature, prevents duplicate votes, hides buttons when they do not apply, and warns about unclear opposes.
fpc-archiver.js	Adds small labels with vote counts, time left, close status, possible image issues, invalid votes, camera type and GPS when available.
Hover tools	Shows a small FPC vote summary for users, a zoom loupe for images, and basic EXIF data near images.
Closing tool	Shows a "Close..." button only when a nomination is ready. It counts votes, ignores late votes, checks eligibility, marks invalid votes, writes the result template, and updates the heading.
Image checks	Shows possible issues such as low resolution, missing metadata, high ISO, blur, compression, color cast, low contrast, missing EXIF, AI software, or HDR processing. These are warnings, not final decisions.

How to try it

Add this to your Special:MyPage/common.js:

mw.loader.load( '/w/index.php?title=User:Wilfredor/fpc-voter.js&action=raw&ctype=text/javascript' );
mw.loader.load( '/w/index.php?title=User:Wilfredor/fpc-archiver.js&action=raw&ctype=text/javascript' );

I would like feedback on
What is confusing
What does not work
What feels intrusive
Which warnings are wrong
Whether this should become an opt-in gadget
Whether voting and closing should be separate gadgets

Edits made by the scripts are marked in the edit summary, so they can be checked later.

Some screenshots

• 
  New controls, zoom and EXIF info sample
• 
  Warning more than two nominations
• 
  Nominate image to FP
• 
  Lupe and check list sample

Thanks — Wilfredor (talk) 14:41, 24 May 2026 (UTC)

Hi, @Wilfredor. I really like the idea, and I tested it. However, it is using your signature. heylenny (^talk/_edits) 18:02, 26 May 2026 (UTC)

 Done Thanks for test it, please, let me know if work now, obrigadao Wilfredor (talk) 20:06, 26 May 2026 (UTC)

Hi @Wilfredor! I've also added your scripts and had a peek at their features and so far I quite like it. One small addendum: when an image is technically FP-ready the list is very extensive, but all of those pure technical data is omitted when your script detects issues. Wouldn't it be reasonable to see them anyways? And another remark: FPC should technically be fulfilling a lot of things, but officially they must have a wow factor. How could an AI detect that? ;-) --Granada (talk) 06:01, 27 May 2026 (UTC)

Thanks for trying the Granada script and for your suggestions. I've been researching this and the most realistic options for it:

- NIMA (Google, 2017) — Inception V2 trained on the AVA dataset, returns a 1-10 aesthetic quality score. It correlates ~0.7 with human voting and ~0.5 with FPC-like decisions.

- CLIP aesthetic predictors (LAION-aesthetic, SAC) — regression head on CLIP embeddings. They capture "pretty pictures" but are biased towards Discord/Pinterest aesthetics, not documentary photography.

- Saliency models (SAM, BASNet) — detect where the eye looks. Useful for detecting "decisive" composition.

- Composition scorers — detect rule-of-thirds, leading lines, and diagonal balance.

The problem is that correlation ≠ wow. They can point out prerequisites (composition, light, contrast, defined subject), but the real "wow" factor of a First-Person Photo (FPP) usually comes from:

- The rarity of the subject or moment (not measurable without context)

- Documentary significance (a portrait of the last speaker of a language matters more than the same technically perfect portrait of a passerby)

- Emotional state or decisive moment (Bresson)

- The photographer's courage (they went where no one else had)

A viable, useful tool would be an "aesthetic prior," an informative (not opinionated) pill that says "NIMA 7.8/10 · CLIP-aesthetic 6.2/10 · composition: ROT 0.8" without marking an issue. The human reviewer still decides on the "wow" factor; the script simply states "this starting point is not in the aesthetic bottom quartile." This avoids acting as a gatekeeper. I already tried doing this, but Wikimedia Commons doesn't allow loading external APIs. Ultimately, this could potentially lead to many false positives, and my answer to your question is that the "wow factor" is exactly what should remain with the reviewer's eye; that's the correct division of labor. Any attempt to automate it would create more friction than value. Wilfredor (talk) 11:11, 27 May 2026 (UTC)

Your work is great, thank you very much for it! And it points in the necessary direction: to automate the QIC process. Maybe you can bring this old idea back to life. --Granada (talk) 11:30, 27 May 2026 (UTC)

I already made an official gadget approved by Wikimedia commons community, many years ago it is widely used, I don't know if that's what you're referring to, it's called QICvote and you can find it in your preferences, you can activate it there Wilfredor (talk) 11:43, 27 May 2026 (UTC)

If you mean automating the process a lot, this would leave many people with nothing to do, people have fun doing things manually, it makes them feel useful Wilfredor (talk) 11:44, 27 May 2026 (UTC)

I made the change you asked me to not omit the checks Wilfredor (talk) 11:13, 27 May 2026 (UTC)

Hi @Wilfredor, I noticed that now no FPC at all is technically fp-ready and the reason seems to be the depicts statement in structured data. I did not cross-check all current candidates, but I found at least two (e.g. my nom of Monroe at the ESC 2026 where I added P180 after I saw the remark) with existing P180-statement and the "error" showing up in fpc-voter.js. --Granada (talk) 12:10, 28 May 2026 (UTC)

 Done Hi Granada, thanks for the report, confirmed and fixed. The check was reading claims.P180 but Wikibase MediaInfo entities expose their statements under statements (not claims, which is for Q-items). Hard-refresh and the depicts check should now properly detect existing P180 statements. Sorry for the noise! Wilfredor (talk) 12:26, 28 May 2026 (UTC)

Three small rule updates for FPC

Hi all. After looking through ~1,400 closed nominations from the last 8 months, I'd like to suggest three small updates.

1. Active-recent edits for voter eligibility

Today: 10-day account + 100 lifetime edits.

Proposed: 10-day account + 100 edits in the last 6 months.

Why: A 2014 account with 105 lifetime edits passes today, even if dormant for years and re-activated to vote once. Recent-activity is harder to game with sleeper accounts. Active contributors aren't affected — even 3 edits a week clears it.

2. Raise minimum resolution from 2 MP to 4 MP

Today: 2 MP minimum (set in the mid-2000s).

Proposed: 4 MP minimum for born-digital photos.

Exception: historical content (scans of pre-~1980 photos, daguerreotypes, microfilm, etc.) stays exempt — quality there is limited by the original medium.

Why: Any modern smartphone shoots 12 MP. A 4K monitor is 8 MP. 2 MP no longer means "good enough for FP" — it just preserves a floor that's hard to defend today.

3. Disclose AI-generated content

Today: No explicit rule — every borderline case is argued individually.

Proposed: Nominators must disclose any AI tool that generates new image content not in the original capture.

• Must declare: Adobe Generative Fill/Expand, Photoshop Sky Replacement, Midjourney, Stable Diffusion, DALL-E, Firefly, AI relighting that invents content.
• Need not declare: AI denoise (Lightroom Denoise, DxO, Topaz), sharpening, AI-assisted masks used only in workflow, in-camera HDR/panorama merging.

A simple template, e.g. {{AI-content|tool=Generative Fill|note=sky replaced}}, goes on the file's description page. Failure to disclose = grounds for {{FPX}} or post-promotion withdrawal.

Why: Reviewers can vote on full information; authors avoid retroactive controversy when AI usage is discovered later. The line (generate vs. process) matches what photographers already treat as material.

— Wilfredor (talk) 12:14, 26 May 2026 (UTC)

All looks acceptable for me. Юрий Д.К. 14:27, 28 May 2026 (UTC)

Thanks for the feedback, let's wait for what others say Wilfredor (talk) 16:57, 28 May 2026 (UTC)

•  Oppose rules 1 and 2,  Support rule 3. I think when presenting new rules it is always critical to consider what problem they are trying to solve. In the case of rules 1 and 2, I think they are solutions in search of problems. Re. rule 1, I am not seeing any problem on FPC with sleeper accounts showing up after years of inactivity to sway the results of specific nominations. What I can see is dormant but trusted users coming back to FPC after a long break (including me earlier this month) who would be penalised by this rule in order to solve a problem that doesn't exist. Re. rule 2, I am not seeing lots of photos that barely scrape over the 2 megapixel lower limit get promoted. I do occasionally see a photo <4megapixels get promoted but this is always pointed out in the nomination and a convincing argument for mitigating circumstances is made. The community is generally good at this, and I prefer the flexibility afforded to me as a reviewer by keeping the lower limit as it is - if a subject is particularly extraordinary or hard to capture, I may still support a 2026 digital camera photograph <4megapixels.

Rule 3 feels entirely different and far more sensible to me. This, unlike the others, is a proactive solution to an issue we have actually had, and unlike the others, it does not risk penalising otherwise trusted users or banning otherwise worthy FPs because the megapixel police have come to town. Cmao20 (talk) 19:44, 28 May 2026 (UTC)

• I'd recommend tweaks. #1 - 10 edits in the last 6 months seems sufficient to me. #2 - Not sure this is necessary. Very few 2-4 MP photos get promoted, but every once in a while there's a very small subject that makes it through, and that doesn't seem like a problem. #3 - Better to specify what doesn't require disclosure, and requiring disclosure of all other use of generative AI tools, rather than try to list both sides. — Rhododendrites ^talk |  20:14, 28 May 2026 (UTC)

• I would like to add a recent example from the FP process that may be relevant to the discussion about minimum resolution. A photo taken by me in 2008 with a simple compact camera was promoted to FP status last year, despite its low technical resolution. In the nomination itself, I had already raised concerns about the limited resolution. During the discussion, however, several reviewers emphasized composition, atmosphere, and timing over pure pixel count. The image was ultimately promoted with a clear consensus. This case suggests that the community already handles low-resolution edge cases contextually, weighing artistic and documentary value against technical limitations rather than applying a strict numeric threshold. It may indicate that flexibility at the lower end of the scale is currently working as intended. Best, -- Radomianin (talk) 21:18, 28 May 2026 (UTC)

  Thanks for the comments so far.

  Regarding rule 2, this is a topic that has already been discussed at length, but I thought it was important to bring it up again. I think I could concede on this point and leave it aside for now, especially because of the arguments about context and not necessarily limiting this aspect.

  Regarding rule 1, I think it could be adjusted to 10 edits in the last 6 months. That seems quite reasonable, perhaps better than 100, because the idea is to show recent activity, not to exclude occasional editors.

  Rule 3 seems to be the clearest and least controversial one. Perhaps it makes sense to move forward with this one first. Wilfredor (talk) 02:22, 29 May 2026 (UTC)

• First let me thank you, Wilfredor, for this clear proposal with three well-defined ideas. When people start to discuss about FPC rules, they sometimes try to do everything at once, declare that FPC is fallen/corrupted/worthless …, and demand a complete overhaul or restoration of FPC. That’s honourable, but not very constructive, and such far-reaching proposals normally trigger extensive and somewhat chaotic discussions which usually end without any clear results. Your proposal is much better: it’s exemplary because it is clear, focussed, and constructive. Therefore many thanks!

Regarding the individual rule updates I mostly agree with Cmao20 and Radomianin.

1. I would be fine with proposal #1, but as Cmao20 has explained it has little actual relevance, so we can change the rule or leave it as it is.
2. I would also be fine with proposal #2, i.e. raising the official minimum resolution from 2 MP to 4 MP; but in order to cope with special cases like the one mentioned by Radomianin we would have to add a long list of possible exceptions. This could confuse nominators instead of helping them. So it may be simpler to keep things as they are, because, as Radomianin has pointed out, “the community already handles low-resolution edge cases contextually”.
3. Like others, I agree with proposal #3. I would just add that people can also still use the traditional {{Retouched}} template to disclose AI-generated content – if one uses that template to describe several kinds of edits, one of them using AI-generated content, it’s simpler and clearer to mention all edits in one description in one template instead of using an additional template for some minor AI-generated content. The important point is that generated content is declared, just like other kinds of far-reaching edits.

Thanks to all participants and all the best, – Aristeas (talk) 15:38, 29 May 2026 (UTC)

Regarding your comment at the beginning, and other comments of appreciation you’ve made lately, I want you to know that I truly appreciate them. I value them a lot, especially after a year away from FPC. It’s good to receive so much positive feedback. Wilfredor (talk) 16:42, 29 May 2026 (UTC)

FPCbot error

This nomination was miscounted by FPCbot, which was missed by the reviewer. I left a message on the page, but it seems nobody has seen it. What is correct procedure so as not to cause problems with the bot? Courtesy ping for the reviewer, Wilfredor. — Rhododendrites ^talk |  13:03, 28 May 2026 (UTC)

Dear Rhododendrites, i'm very sorry for missing your oppose, it was at the *:: indent level so neither FPCBot nor my script picked it up at close time. Corrected the tally on the nom to 17S/1O (result still, featured: 17 ≥ 2×1, so promotion stands). Going forward, my script now warns on tally discrepancies and writes its own count (which picks up sub-thread votes) instead of trusting the bot should, prevent this kind of slip. For unambiguous votes, top-level *  Oppose is still safest since FPCBot uses that strictly. Thanks for the courtesy ping my friend. Wilfredor (talk) 13:33, 28 May 2026 (UTC)

Didn't realize FPCbot couldn't see indented opposes. Good to know, thanks. The promotion will probably still stand, but it doesn't qualify for 5-day. — Rhododendrites ^talk |  13:37, 28 May 2026 (UTC)

true, you are right, the winner needs 5 days with no opposition, the nomination did not qualify for early closing; I would have had to wait the full 9 days. The promotion is still on, but from a process point of view the early closure was incorrect. Apologies, in the future, the script I am developing will detect that it now accepts votes with a two-point indentation, so your vote would be included. Both the discrepancy warning and day five winner suggestion now use that full count instead of FPCBot's strict top-level count. We can leave the nomination as it is by reverting all post-close edits (gallery list, reviews, chrono file, talk page notifications, etc.) so that a result that would be identical on the 9th would create more noise than it solves. It should be enough to mark the closure as procedurally early in the conversation thread. thanks my friend Wilfredor (talk) 14:19, 28 May 2026 (UTC)

I’m not sure I understand what we are talking about. FPCBot counts indented votes, too, just like unindented ones. It just ignores votes which are commented out with <!--...-->, have been crossed out (with <s>...</s> and similar tags) or are part of a <nowiki>...</nowiki> section. Please see the definition of the function filter_content() for details. At least this applies to the version which is available from GitHub and AFAIK is still used here. Are we talking about some modified version of FPCBot? How was this edit done – with some modified copy of FPCBot? Or have you found an unknown bug? Then please describe it and we can fix it, no problem. – Aristeas (talk) 14:59, 29 May 2026 (UTC)

OK, I have tested it. I have copied the last version of that nomination before the vote counting to my test nomination (result). Then I used my copy of FPCBot to close it – this is what FPCBot did. AFAICS the result is correct: it found 17 support, 1 oppose, 0 neutral votes. No offence, but please allow me to clarify this: in this case the miscounting was not a FPCBot error. FPCBot has many weaknesses (also because I did not want to rewrite code from scratch, but to improve it step by step), but it has no problems with indented votes. – Aristeas (talk) 15:08, 29 May 2026 (UTC)

Oh, I see. This was a purely human error. [1]. I thought FPCbot did its tally, and Wilfredor didn't double-check to make sure it was right, but it looks like both tasks were done at the same time. That seems to defeat the point of the review step. I'll note that this invalid promotion is still marked as promoted/closed. — Rhododendrites ^talk |  15:19, 29 May 2026 (UTC)

Agree. A good thing about FPCBot is that leaving over vote counting to the bot and reviewing the results afterwards adds an additional step which can help to discover unclear votes or vote counting mistakes before a nomination is finalized and parked. So we better do not mix up these steps, but keep them separate. – Of course please forgive me if my comments above were a bit harsh – I did not want to criticize you or Wilfredor. I was just confused because I should know such a FPCBot bug by now if it exists, and was eager to defend the poor bot because in this case it was not it’s fault ;–). Best, – Aristeas (talk) 15:48, 29 May 2026 (UTC)

Thank you both for checking this, and sorry for the confusion. You are right, this not was an BPCbot bug, I misunderstood the situation and incorrectly assumed the indented oppose had not been counted by the bot. The bot is dead, long live the new script (just kidding). Feel free to criticize me; that way I can improve the script, or you can improve the bot, or both. I understand your desire to defend the bot; often we treat a bot like a child, feeding it commits regularly and giving it edits to make. The script is still in testing, so it's normal for these issues to occur. I added a validation check to see if the bot has already started doing the same thing, so the chance of a conflict is 0.7%. On the other hand, I agree with the ability to review and count. The person closing the session is the one who should confirm this. The script facilitates this operation by displaying the count result before closing. I should adjust my workflow so that future closures keep the FPCBot count and the human review as separate steps? Wilfredor (talk) 16:39, 29 May 2026 (UTC)