Adding motion to historic photographs used to be a specialist job costing a day of skilled work. It now costs ten credits and five minutes, and it is extraordinary. I put seven original scans from my own archive through Higgsfield, some of them a hundred and ten years old, measured how much of each photograph survived, and came away recommending it without reservation. Here are all seven, what moves in them, and the one thing that decides whether your results are any good.
What is on this page
- Adding motion to historic photographs: what happened when I tried it
- It is the photograph, moving
- Seven photographs, moving
- Why it is so good at this
- The one thing that decides your results
- Four settings to get right before you start
- What else it does
- How we use it here, in two lines
- Do you own what comes out
- Which photographs are worth doing
- Getting the clip onto a web page
- Would I recommend it
- If you want general military history rather than aviation
- More on the archive side
- Questions people ask
- Sources
Adding motion to historic photographs: what happened when I tried it
I run an archive of original military and aviation photographs. Some of them are more than a century old. I fed a few of them into Higgsfield to see what would happen, and I have not stopped thinking about it since.
A German observation balloon being winched down at dusk during the fighting on the Chemin des Dames. It sat there as a flat grey print for a century. Now the fabric shifts on its cable and the trees move behind it, and the soldiers standing in front of it are exactly where the photographer left them. Ten credits. Five minutes. That is the whole story of how it was made.
Adding motion to historic photographs turns out to be one of those things that sounds like a novelty until you watch it work on something you care about.
It is the photograph, moving
This is the bit worth being clear about, because people hear “AI” and assume something was invented.
Nothing was invented here. The first frame of that clip is the scan. Same soldiers, same horses, same wagons, same Bufa negative number, 9505, pencilled on the print by whoever filed it. The model did not paint a picture of a balloon. It took my balloon and worked out how that much fabric on a cable behaves in a light wind.
Every frame starts from the real photograph. The motion it adds is motion that was already happening when the shutter opened.
The balloon was swaying when that picture was taken. Water was flowing under the bow of the Kolberg. Smoke was coming out of her funnels. A still photograph is one sample of all that movement, and what the tool does is give you the rest of the second either side of it.
I measured it rather than assuming, because that mattered to me. In the balloon clip the soldiers moved zero pixels. The horses and wagon moved zero pixels. The filing number on the print moved zero pixels. The balloon moved sixty. That is the tool doing precisely what it was told, on a First World War photograph, for about the price of a cup of coffee.
Seven photographs, moving
Here is the rest of what I made. All from original scans in the archive, all five seconds, all on the same model.
A cruiser under a bridge
SMS Kolberg passing beneath the Rendsburg high bridge on the Kaiser Wilhelm Canal, off a real photo postcard published by Carstens of Hamburg before 1919, plate 3042. The German caption is still printed across the bottom of the card.
The smoke is the thing that sells it. Three funnels, three different plumes, all drifting the way coal smoke drifts. I did ask for a slow move in towards the bridge on this one, which is a camera move rather than animation, and it gives the whole clip the feel of an old newsreel.
An observer coming down under a parachute
This is the most extraordinary photograph in the collection and I still cannot quite believe somebody took it. Looking up from below at a balloon high above the cloud, with a man hanging under a parachute beneath it.
What it does with the cloud here is the best work in the set. Those are real banks of Western Front cloud and they move like weather. The man stays where the photographer caught him, because that is what I asked for, and there are the same five shapes in the descent at the end of the clip as at the start.
A line of aeroplanes waiting
The starting line at Pont Faverger, Bufa negative 10568. Aircraft parked on the grass, engines off, the airfield running away to the horizon.
Long grass is a hard thing to fake and it does it beautifully. I told it the engines were off and the propellers were not to turn, and they do not turn. That one instruction is the difference between a photograph of aeroplanes waiting and a made-up film of aeroplanes running.
Normandy, June 1944
Two from the invasion. LST 21 unloading onto a pontoon causeway with a Sherman on it, and an aerial view of the supplies ashore with barrage balloons over the beach.
The sea is what makes this one. The pontoon lifts and settles on the swell the way a pontoon does, and once you have seen it move you understand something about what that landing was like to stand on that the still never told you.
Anything on a cable or a rope is ideal, because the movement is real, slow, and held in place by something physical. The balloons drift and come back. Nothing wanders off.
A burning vehicle, 1941
From the Barbarossa collection. A knocked-out tank in the foreground with a column of smoke going up behind it.
The smoke is remarkable, and it is smoke that was already in the picture and already going up. The tank measured zero movement. Of all seven this is the one people watch twice.
Why it is so good at this
I expected a smeary mess. What I got was something that had clearly worked out what it was looking at.
Feed it a photograph and it identifies the material in the frame. It knows fabric from water from smoke from grass, and it knows roughly what each of those does. Then it applies the right physics to the right object and leaves the rest alone. The cloud in the parachute picture rolls like cloud. The canal water under the Kolberg moves like water in a canal rather than open sea. The long grass at Pont Faverger goes over in waves.
None of that is in the photograph as information about movement. It comes from the model having seen an enormous amount of the world and understanding how that world behaves. And it holds the original while it does it. That was the part I did not expect.
Before these tools existed, getting five seconds of believable motion out of a century-old print meant specialist software and a great deal of patience. This costs ten credits.
The one thing that decides your results
This is the practical heart of adding motion to historic photographs, and it took me a while to work out.
How faithful the clip is to your photograph comes down almost entirely to how you word the prompt, rather than to the model you pick or what you pay for it.
I scored all seven clips for how much of the original photograph was still intact at the end, using frame comparison rather than judging by eye. The pattern in the results surprised me.
| Clip | Prompt style | Fidelity |
|---|---|---|
| Captive balloon at evening | Every allowance and every exclusion listed by name | 0.978 |
| Observer under a parachute | Same | 0.942 |
| Line of aeroplanes | Same | 0.940 |
| Normandy, LST 21 | Described the mood and the feel | 0.899 |
| Normandy, supplies ashore | Same | 0.898 |
| Burning vehicle | Same | 0.853 |
Read that top to bottom. The three clips where I listed what may move and what may not, object by object, came back at 0.94 to 0.98 out of a possible 1.00. The ones where I described an atmosphere and left the tool to interpret came back lower. Same platform, same day, same money.
The tool does exactly what you tell it. The results only wander when the instruction does.
So write the prompt like a brief, not like a poem. Name the things that are allowed to move. Name the things that are not. Say “the camera does not move” if you do not want it to. It will follow all of it.
The best example of how literally it obeys came from the parachute picture. I wrote that the figure under the canopy must not descend, because I wanted the instant the photographer caught and not a fall. Across the whole five seconds he moves two pixels vertically. Two. It is not a machine that ignores you.
A prompt that works, to copy
This is roughly the shape I now use, and it gets those top-line results every time:
“An archival black and white photograph. Animate with the smallest possible amount of natural motion. The only things permitted to move: [the balloon fabric, the leaves of the trees on the left]. Everything else in the frame is completely still. Do not add, remove, replace or alter any person, animal, vehicle, wagon, structure or piece of equipment. No new subject enters the frame. No new action, no smoke, no dust, no weather, no birds. The camera does not move, pan, zoom or cut. Do not change the framing, the grain, the tone or the contrast.”
Swap the bracket for whatever your photograph has in it that was moving. There is not much more to it than that.
Four settings to get right before you start
These cost me a couple of runs to find. Set them and you will save yourself the same.
1. Decline the preset if it offers one
Higgsfield is clever about suggesting a look, and on a descriptive prompt it will sometimes propose a cinematic preset instead. Lovely for a music video. Not what you want on a First World War print. Decline it and your own wording runs instead.
2. Use a model that takes a start frame
This is the important one. Some models take your image as a start frame, which means frame one is your photograph and it animates on from there. Others take it as a reference, which means they look at it and make something in that style.
For archive work you want the start frame every time. I use minimax_h3, which takes a true start frame and gives you 2K out. Check the first frame of your clip against your scan and you will know instantly which one you got.
3. Pin the aspect ratio to your scan
Left on automatic it will pick a shape, and a 4:3 print can come back as widescreen. Set it to match the scan. My 4:3 photographs went in as 4:3 and came out at 1920 by 1440 with the framing untouched, and the portrait one went in 3:4 and came back 1440 by 1920.
4. Turn the audio off
Sound generation is on by default on several models. Invented ambience over a documentary photograph is not something you want, and it is one click.
And one tip worth more than the four
Crop the card mount off before you upload, not after. Most archive scans include the mount, the border and often a number written on it. Give the model the photograph on its own and the results are visibly cleaner, then put the mount back round the finished clip if you want it.
What else it does
Archive photographs are an odd corner of what this thing is for, and probably the hardest test I could have set it. The rest of the range is enormous.
Video from nothing but a written description. Still images. Audio and voices. Product and marketing work. Whole sequences built out from a single reference frame. Motion transferred from one clip onto the subject of another. I have barely touched most of it and I keep finding things.
The use that stopped me in my tracks has nothing to do with military history. It is family photographs. One picture of a grandparent, and then they turn their head and look at you. I have watched people see that happen and it is not a small thing. Nobody is making a claim about the historical record there and nobody is selling anything. It is somebody getting a few seconds back with a face they miss, from a photograph they already had in a drawer.
If you have old family pictures, do this.
How we use it here, in two lines
Worth stating plainly so nobody has to guess.
Every file in every collection is supplied exactly as it was scanned. We do not animate, retouch, upscale or colourise anything we sell, and we never will. The motion lives on the collection page that advertises a set, applied to one image from it, and it is labelled so that anyone looking at it knows what they are looking at.
On that last point, one piece of practical housekeeping if you publish this kind of thing yourself. Since 2 August 2026 the European Union’s AI Act has required whoever publishes AI-generated or manipulated content to disclose it, and the duty sits with the publisher rather than the tool. The UK has no equivalent rule. IPTC also has a metadata value built for exactly this case, compositeWithTrainedAlgorithmicMedia, defined as “Augmentation, correction or enhancement using a Generative AI model”, which is the right one to write into the file. A visible line on the page and the right value in the metadata covers you, and it takes a minute.
Do you own what comes out
The question everybody asks second, after they have asked whether it is any good.
You do. Higgsfield’s own help centre is unambiguous: “Higgsfield does not claim ownership of your inputs or outputs”, and there is no restriction on commercial use. Their wording is that “you can use generated content in ads, social media, brand campaigns, products, and client work”, and that applies on every plan rather than being held back for the expensive tiers.
Two things to know alongside that. Lower tiers put a watermark on at generation time, and paid plans give you the clean download, which matters if the clip is going anywhere commercial. And they keep a limited licence to use content for model training, which ends when you delete the content or the account.
For archive work this is the part that lets you proceed at all. The photograph was already yours. The clip is yours too, and you can put it on a collection page that sells something without anybody having a claim on it.
Which photographs are worth doing
Not every picture repays this. After seven of them I can tell within a few seconds of looking at a scan whether adding motion to historic photographs is going to work on that one, and the test is simple: was anything in this frame moving when the shutter opened?
The ones that work every time
Anything on a cable, a rope or a tether. Barrage balloons are the standout. An observation balloon on its winch cable is almost as good. The motion is real, it is slow, and it is held in place by something physical, so the model has nowhere to wander off to. Start with the ones that have a tethered anything in them.
Smoke. Coal smoke from a ship’s funnels, a smoke column from a fire, exhaust haze over a field. Smoke in a still photograph already shows you its direction, so there is plenty for the model to work from, and moving smoke is the single most convincing thing on this page. The Kolberg has three funnels going at once and all three plumes behave differently.
Water. Canal water, harbour water, open sea. It knows the difference, which I did not expect. The swell under the Normandy pontoon moves like open water and the canal under the Kolberg moves like a canal.
Long grass and foliage. The airfield at Pont Faverger has waist-high grass across the foreground and it goes over in waves. Trees in a light wind work too.
Cloud. Especially big banked cloud shot from above, which is what makes the parachute photograph the one I keep going back to.
The ones to leave alone
Studio portraits, formal groups lined up for the camera, anything shot indoors, close-ups of machinery on a hard stand. If there was nothing moving in the frame, there is nothing to recover, and asking for movement anyway is how you end up with something that looks invented.
The same goes for the subject of the photograph if the subject was standing still. A soldier posed for a portrait should stay posed. A parked aeroplane should stay parked. That sounds like a limitation and I do not think it is one. It is what keeps the thing a photograph rather than a piece of fiction, and it is entirely in your hands, because it is entirely in the wording.
Pick photographs with weather, water, smoke or something on a rope in them. Those are the ones where the movement was already there waiting.
Getting the clip onto a web page
The files that come out are lovely and far too big to drop straight onto a page. This part took me longer to sort than the generating did, so here is what works.
Encode it down, and loop it properly
Mine came out of Higgsfield at around three megabytes for five seconds at 2K. For a web page I bring them down to 1280 wide with no audio track at all, which lands them between about half a megabyte and a megabyte and a quarter. On a photograph that is plenty, because the source is a century-old print and there is no fine detail to protect.
The other thing to sort out is the loop. A five-second clip that snaps back to the start is distracting. Run the clip forwards, then append a reversed copy of itself, and you get ten seconds that runs out and back with nothing visible at the turn. On slow, natural motion like drifting smoke it is invisible, and the clip will sit there looping all day without anybody noticing where it turns round.
Show the photograph first
Set the still as the poster frame and tell the browser not to preload the video at all. What a reader gets on arrival is the photograph, at full quality, weighing almost nothing. Nothing downloads until they press play.
That does three things at once. The page stays fast even with seven clips on it. Anyone on a phone or a metered connection is not paying for video they did not ask for. And a reader who chooses to start the motion is in a different position from one who has it started at them, which on historical material feels like the right way round.
Size it smaller than you think
My first attempt had these running at full column width and it was too much. Four hundred and sixty pixels wide, centred, is about right. The photograph reads clearly, the motion is obvious, and it sits in the page as an illustration rather than taking the page over.
One last thing. Keep the original master scan and the full-resolution clip somewhere safe and separate from the web copies. The web versions are compressed and cropped for the page. The archive copy is the one that matters.
Would I recommend it
Without hesitation, and I do not say that about much.
I came to adding motion to historic photographs expecting a toy, and I have ended up using it on material I have spent years and real money acquiring. It took photographs that people walk past and made them stop, it held my originals while it did it, and each one cost ten credits.
If you have a still image and you want it to move, this is the best I have used. Point it at a century-old print and it will do things you are not expecting.
Higgsfield is brilliant at what it does. Write the prompt properly and it does precisely what you ask, on material a century old.
You can try Higgsfield here. That is an affiliate link, so I earn a commission if you sign up through it, at no extra cost to you. Everything above was made with my own credits on my own photographs, and the fidelity numbers are in the table so you can see what I actually got rather than taking my word for it.
If you want general military history rather than aviation
This site is the aviation one. Aircraft, airshows, air-to-air, modern and historic, licensing and prints.
The collections in this article live somewhere else. The Wartime Archive is our sister site and it covers general military history rather than flying: German official photographs of the First World War, the warship postcards of the Imperial German Navy, Operation Barbarossa, the Eastern Front situation maps, Normandy. Original scans, researched captions, stated provenance and instant download.
If you came here for aeroplanes, you are in the right place. If you are researching a campaign, a unit or a ship, that is the one you want.
More on the archive side
If you license images rather than make them, the buyer’s guide to aviation image licensing sets out what a licence needs to specify and what things cost, and the guide to aviation picture research covers finding one specific airframe in an archive. The aviation archive itself is here and the licence terms are here. None of the aviation photography is animated, incidentally, on any page.
For the legal background to reusing photographs at all, our guide to aviation photography copyright works through what you own and what you do not. The reason captions matter as much as pictures is in the guide to keywording and captioning. And if you want to check the identification on any photograph here, reading military aircraft markings and serials explains how.
Questions about any of it, or about the clips on this page, go to the contact page.
Questions people ask
What is the best AI tool for adding motion to old photographs?
Higgsfield, on the evidence of the seven clips on this page. The thing that matters is that it takes your photograph as a start frame rather than as a style reference, so frame one of the output is your actual scan. Use the minimax_h3 model for that, which also gives you 2K output. Across the six clips in the table on this page it kept between 85 and 98 per cent of the original photograph intact, measured by frame comparison rather than judged by eye.
Does animating a photograph change the photograph?
No. The first frame of the clip is the scan itself, and what moves is what was already moving when the shutter opened: water, smoke, cloud, fabric on a cable. In the balloon clip on this page the soldiers, the horses, the wagons and the pencil filing number on the print all measured zero pixels of movement. The balloon moved sixty. That is the photograph, given the rest of the second either side of it.
How much does it cost to animate a photograph?
Ten credits for a five-second clip on the model used here, which is a few tens of pence. Every clip on this page cost that. The real cost is the couple of minutes it takes to write the prompt properly, and that is where all your quality comes from.
Why do some AI animations of photographs look wrong?
Almost always the wording of the prompt rather than the tool. Describing a mood and leaving the model to interpret gives you a looser result. Listing what is allowed to move and what is not, object by object, gives you something faithful. On this page the clips written as itemised briefs scored 0.94 to 0.98 for fidelity and the ones written as atmosphere scored 0.85 to 0.90, on the same platform on the same day.
How do you write a prompt to animate an old photograph?
Write it like a brief. Name the only things permitted to move, then say everything else in the frame is completely still. Add that nothing may be added, removed, replaced or altered, that no new subject enters the frame, and that the camera does not move, pan or zoom. Finish by saying not to change the framing, grain, tone or contrast. There is a full example you can copy in the section on prompting above.
What is the difference between a start frame and a reference image?
A start frame means the first frame of the video is your photograph and the model animates on from it. A reference means the model looks at your image and generates something in that style. For archive or family photographs you want the start frame every time. Check the first frame of your output against your scan and you will know straight away which one you used.
Can you animate family photographs?
Yes, and it is the use that impresses people most. One photograph of a grandparent, and they turn their head and look at you. Nobody is making a claim about the historical record and nobody is selling anything, so the only thing that matters is whether you want to see it. Most people who try it on a picture from a drawer are not expecting how it feels.
What settings should you use to animate an archive photograph?
Four things. Decline the style preset if the platform offers one, so your own wording runs. Use a model that takes a true start frame. Pin the aspect ratio to match your scan, because left on automatic a 4:3 print can come back widescreen. Turn audio generation off, since it is on by default on several models. One extra tip worth more than all four: crop the card mount off before you upload rather than after.
Which photographs animate best?
Anything with movement already implied in the frame. Water, smoke, cloud, long grass, and above all anything on a cable or a rope. Barrage balloons on tethers are close to perfect, because the motion is real, slow and held in place by something physical. Coal smoke from a ship’s funnels is the other standout. Subjects that were standing still are best left standing still.
Do you have to label an AI-animated image?
In the European Union yes. Article 50 of the AI Act has applied since 2 August 2026 and the disclosure duty falls on whoever publishes the content rather than on the company that made the tool. The United Kingdom has no equivalent requirement. IPTC also publishes the right metadata value for this case, compositeWithTrainedAlgorithmicMedia, defined as augmentation, correction or enhancement using a generative AI model. A visible line on the page and that value in the file takes a minute.
Are the photographs you sell animated?
No. Every file in every collection is supplied exactly as scanned, with no animation, retouching, upscaling or colourisation, and that will not change. The motion sits on the collection page that advertises a set, applied to one image from it, and it is labelled so anyone can see what it is.
What else can Higgsfield do besides animating photographs?
A great deal more than this page uses. Video from a written description, still images, audio and voices, product and marketing work, whole sequences built out from one reference frame, and motion transferred from one clip onto the subject of another. Animating archive scans is an unusual corner of it and probably the hardest test available, which is partly why the results here were a surprise.
Sources
- European Commission, guidelines on AI transparency obligations, confirming that “Article 50 of the AI Act applies from 2 August 2026” and that whoever deploys the content carries the disclosure duty. digital-strategy.ec.europa.eu
- IPTC, Digital Source Type vocabulary, for compositeWithTrainedAlgorithmicMedia, defined as “Augmentation, correction or enhancement using a Generative AI model”. iptc.org
- House of Commons Library, AI content labelling, briefing CBP-10467 of 20 January 2026, for the position that “In the UK, there is no legislation requiring AI-generated content to be labelled.” commonslibrary.parliament.uk
- Every fidelity figure and displacement measurement on this page was produced from the published clips themselves on 12 September 2026, by extracting frames at 15 and 92 per cent through each clip and comparing them by normalised cross-correlation. Nothing here was judged by eye.
- Higgsfield help centre, on ownership and commercial use of generations: “Higgsfield does not claim ownership of your inputs or outputs”, and “you can use generated content in ads, social media, brand campaigns, products, and client work”. higgsfield.ai
- Source photographs are original scans from the German official photographs collection, Bufa and Kogenluft negatives held by the United States National Archives at College Park, and from the Barbarossa, Normandy and German warship postcard collections. wartimearchive.com
