Articles from Gavin Johnson

AI can help camera admin...

nobody@ipernity.com (Gavin Johnson) — Sat, 16 May 2026 13:20:33 +0000

Gavin Johnson has posted an article:

One of this morning's little jobs was to upgrade the camera bodies to the latest version of firmware, released this week. While a lot of Canon lenses go their whole production run on their release firmware, there are sometimes updates to fix bugs or improve performance. I'm sure the other lens manufacturers do the same thing. Checking up on all the lenses is a dull chore, but I decided to make sure everything was properly up to date.

There's an IT management theory that says you should give repetitive tasks to the laziest engineer, as they'll figure out a way to automate the job, making it much easier to do in future. I'm lazy, so I decided to enlist my AI assistant to do the hard work for me.

I created a spreadsheet with columns for mount type, lens name, firmware version and release date, leaving the last two columns empty. I uploaded the spreadsheet to Claude AI (the free tier) and gave it the following prompt:

Attached is a Numbers spreadsheet containing rows for all my Canon camera lenses. Can you find for each lens the current firmware version number and the date the firmware was released and add that info to the spreadsheet?

Claude worked surprisingly hard on this. It loaded a module to read and write Numbers spreadsheets, went down the list and politely corrected a couple of typos I'd made in the lens names, and searched several websites for the information (after it struggled to understand the Javascript formatting of the official Canon website). It compiled all the information, and presented me with an updated spreadsheet with the firmware columns filled in.

I did consider getting Claude to write a program to let me do this on demand, but I was impressed with the creative way it sought out multiple information sources. Instead, I can now upload the spreadsheet to Claude in the future and give it a different prompt:

Attached is a Numbers spreadsheet containing rows for all my Canon camera lenses. Can you check for each lens whether there is a newer firmware version than the one listed in the spreadsheet. If there is, update the spreadsheet and give me a link to download the newer firmware file.

Being lazy isn't always a bad thing!

Will using an AI reviewer help make me a better photographer?

nobody@ipernity.com (Gavin Johnson) — Sat, 28 Mar 2026 18:20:46 +0000

Gavin Johnson has posted an article:

I've been reprocessing a few old RAW files with Lightroom recently, and uploaded one here as my 'picture of the day'. As AI chatbots have been helpful in setting up my IT environment, I wondered how it would fare as a critical eye.

Here's the picture I uploaded:

I gave Claude AI the following prompt:

I'd like to try something different. I will upload a photograph, and I want you to critique it. I'm not looking for flattery; I want an objective assessment, and practical suggestions for composition, technical execution and post-processing to make the image objectively and subjectively better.

Here is its assessment. I'd be really interested in feedback from folks here on what you think of the critique. I definitely agree with some, but not all, and I think it was a worthwhile exercise. Don't hold back - all criticism is very welcome!

Here is Claude's assessment:

Good subject — a classic English rapeseed field in full bloom with a lone oak. There's a strong image in here, but it needs work to realise its potential. Here's an honest breakdown:

Composition

What works: The lone tree as a focal point is the right instinct. The tractor tracks add a leading line, and the colour contrast between yellow, green and blue is inherently powerful.

What doesn't:

The horizon is nearly centred, splitting the frame roughly 50/50 between sky and field. This is the single biggest compositional problem. You need to make a decision: is the sky the story, or is the field? Right now it's neither. Dropping the horizon to the lower third would give the sky more drama; raising it would emphasise the sea of yellow.

The tree is slightly left of centre but not convincingly so — it's in an awkward middle ground between centred and a proper rule-of-thirds placement. Moving your shooting position to place the tree clearly on a third intersection would have created more visual tension.

The second tree on the right is a distraction. It's too small to be a meaningful compositional counterpoint, and too prominent to be ignored. A slight reframe to the left at the time of shooting would have cropped it out entirely.

No true foreground interest. The rapeseed fills the foreground but it's uniform — there's nothing to draw the eye in from the bottom of the frame. Getting lower and closer to the crop canopy would have created depth and scale.

Technical Execution

Exposure is broadly acceptable but the yellow is slightly hot — you've likely lost some highlight detail in the brightest parts of the rapeseed. A stop of negative exposure compensation would have preserved texture in the flowers.

Focus appears centred on the tree, which is fine, but the foreground rapeseed is soft in a way that feels accidental rather than intentional. If you were going for a deep depth-of-field landscape shot, a smaller aperture (f/11–f/16) and focus point at roughly the hyperfocal distance would have kept everything acceptably sharp.

The sky is slightly hazy/washed out, suggesting this was shot in bright midday light. Golden hour — particularly late afternoon — would have warmed the yellow dramatically and given the tree's canopy directional light, separating it far more from the background.

There appear to be a couple of sensor dust spots in the upper sky area (top left). Worth cleaning in post.

Post-Processing

The image looks relatively unprocessed, which gives you a lot of room. Specific suggestions:

Dehaze the sky — add clarity/dehaze to the blue channel to bring back the blue and make the clouds pop.

Selective colour work on the yellow — shift the yellow hue very slightly warmer (towards orange) to make the rapeseed feel richer and less acidic. Reduce yellow luminance slightly to recover texture.

Lift the greens in the tree — they're currently quite dark and flat. A targeted green luminance boost and slight hue shift towards yellow-green will make the canopy feel more alive.

Add a slight vignette to draw the eye away from the edges and towards the tree.

Crop aggressively: raise the horizon to roughly the lower third and crop the right side to remove or significantly reduce the second tree. This single edit would improve the image substantially.

Local adjustment on the sky — a graduated filter darkening the upper sky by half a stop will add drama without looking processed.

Summary

The bones are good — you've identified a genuinely strong subject and the colour palette is working in your favour. The main issues are a safe, indecisive horizon placement and midday flat light. The most impactful fix you can make right now is a crop: raise the horizon and lose the right-hand tree. For next time, revisit the location in late afternoon light and get your camera lower.

In defence of saving (and revisiting) those old photos that didn’t make the cut.

nobody@ipernity.com (Gavin Johnson) — Mon, 16 Mar 2026 15:39:58 +0000

Gavin Johnson has posted an article:

I’ll admit up front, I’m a hoarder. I hang on to all sorts of junk just in case it might turn out to be useful one day. In the attic, I’ve got a box of IT components and cables for technologies that have been obsolete for decades. My wife despairs of me. The same is true of my digital image collection. I was an early adopter when digital cameras came out, and I’ve got 25 years of images - mostly RAW - sitting in SSDs, hard drives, backup drives, and backups of backups. Huge numbers of them didn’t get properly processed at the time I took them. Quite a lot, while in focus and reasonably framed, just didn’t look as good as they did in my imagination when I pressed the shutter. Many have exposure issues, visual clutter, or just felt like they were going to be too much hard work to sort out.

The problem has only magnified as camera technology has moved on. When photographing wildlife, I’ll crank the camera up to maximum frame rate, turn on pre-capture, and let fly. I can come back from two hours at the local wetlands reserve with 4000 images, long sequences of which are almost, but not quite, identical.

I try to cope with today’s volume problem with workflow and a bit of AI assistance. After a first pass to eliminate all the bad (out of focus, out of frame…) images, I’m starting to trust tools like Excire Photo to round up near duplicates, and recommend candidates for deletion. I’ve thought about letting the tools loose on the back catalogue, too, but as part of my migration from Flickr to self-hosted Immich, I’ve been going back through my old ‘not good enough’ shots and discovered that I can bring some of them back to life.

A lot of processing techniques that would have been fiddly and ridiculously time consuming twenty, or even ten years ago are now remarkably straightforward with today’s software and desktop computing power. While I don’t like being tied to Adobe’s software suite, their Lightroom / Photoshop subscription bundle at around £10 a month feels worth the money and offers ever-improving capabilities, but where Adobe goes, others follow, and Affinity and open source image processing tools are not too far behind the cutting edge of their higher-cost cousins. I’m still trying to decide in my own mind what I consider reasonable use of AI processing, and what feels like ‘cheating’, but at the moment I feel comfortable using AI denoising and limited distraction removal. (So far, I’m drawing the line at letting generative AI create new image elements with the exception of bits of background behind clutter that’s been removed).

What feels like the big change for me is the ease with which I can pull out information hidden in the raw files, for example correcting images that felt too badly exposed to be worth bothering with. Tidying up wires or other unwanted details used to mean hours of cloning and smudging, and now it’s done in the press of a couple of buttons. I had a load of images I’d rejected because I let the camera sensor get dusty (back before self-cleaning technology!). Now that’s fixed in seconds, along with lining up wonky horizons - seriously, what was wrong with my sense of the horizontal back then?!!

Washed out or muddy images are easy to turn into sharp pictures with good contrast, and a more pleasant dynamic range, and the ultra grainy high-ISO pictures look great after the AI’s done some smoothing.

There’s no magic here, of course, though the denoising sometimes feels that way. What has changed is that photos that weren’t really worth the time and hard work to process and correct are now fixable with minimal effort, and can turn into ‘keepers’ after all. And, of course, it’s fun going back and remembering the travels and excursions where we took those so-so photographs all those years ago.

Go on, fire up Photoshop, Lightroom, Affinity, Darktable, Luminar or your tools of choice and see if there are any could-be masterpieces hidden away in your ‘not quite good enough’ archives!

Moving away from Flickr

nobody@ipernity.com (Gavin Johnson) — Fri, 13 Feb 2026 10:17:21 +0000

Gavin Johnson has posted an article:

Fortunately, I found Ipernity, which is the sort of community that I was looking for in Flickr. Perhaps Flickr used to be like that, but now - at least for me - it’s a pretty sterile environment with minimal engagement, and few technological advances in recent years.

Ipernity answers half my needs, but over the years I’ve been using Flickr as an on-line dumping ground for photographs, with over 4000 images that I don’t want to get rid of, and would like to have accessible on-line. I’ve had a career as an ‘IT guy’ so I’m fairly comfortable with software, if a bit rusty at writing my own, and I have a Mac Mini running as a multimedia server, so I decided to look at a self-hosted image gallery.

After trying out a few alternatives, I decided that Immich (https://immich.app) was going to meet my needs pretty well. After a bit of trial and error I got it running on the Mac, using Docker (https://www.docker.com) to take care of most of the installation and configuration. Once it was stable, I had a go at retrieving my photos from Flickr.

Flickr make it easy to get a copy of your images, but painful to port them elsewhere. For some reason, the site strips out all the useful EXIF metadata from the images and stores it separately. When you request your photos, Flickr processes for a few minutes or hours, and sends you a link to download a bunch of zip files, including a subdirectory of files in JSON format that relate to all the images in directories elsewhere.

With a bit of searching I found another open source utility called exfitool (https://exiftool.org) that manipulates the data, but it looked like it would take me quite a while to figure out how it works, and many hours to dust off my coding skills and write a program to merge everything. I decided to see if AI could help.

AI ‘chatbots’ and Large Language Models get a lot of well deserved bad press, but as a coding assistant they can be a huge timesaver. It took less than five minutes to describe what I wanted to the Claude AI (https://claude.ai/), with it asking a few clarification questions, after which it created nicely commented Python script that did the job flawlessly; even including a ‘dry run’ mode to test it before modifying any files.

My only outstanding problem is that the Flickr export mixes photographs from different albums into the export directories. There’s another JSON file that seems to contain the mapping, but with the time I’ve saved I’ve decided to restructure the albums manually and curate my collection a bit.

So, it turns out that self-hosting a Flickr alternative isn’t too tricky, and AI can really help in generating easy to follow instructions, and program code that would take a rusty human a lot of hours to get right.