The videos are all created with AI. It’s a pipeline of Flux (images), Kling (video), and mmaudio (audio). The videos aren’t always historically accurate to the last detail. They might incorporate elements of folklore or have details from popular beliefs about the way things looked rather than the latest academic research on how they looked.
I’m thinking a lot about how to make the game more interactive. One thing that makes Geoguessr so fun for me is that you can move infinitely and always find more details to help you pinpoint the location. I want Time Portal to have a similar quality. I have a few ideas to try soon that will hopefully make the game more interactive and infinite.
Also I didn't listen to many of the sounds, but I got English voices for something happening in France (the Fauvisme guess).
But still I had some fun and it's nice to see a good use for AI
However, I also have a "complaint": the "balloon" page (trying to avoid spoiling it) has three videos in a big city and one in the countryside, yet the correct location is the countryside location, not the city (although the city is even mentioned by name in one of the prompts!). So a little more care here is probably warranted...
Aaaand the "pope" page has (according to the prompts) three scenes set at the Vatican, but the location seems to be somewhere else in Rome :)
Are they ever?
> They might incorporate elements of folklore or have details from popular beliefs about the way things looked rather than the latest academic research on how they looked
Like the one of the age of castles man loading an American civil war cannon by holding another cannon up to it: https://www.eggnog.ai/timeportal/37e02fea-bbb2-4b88-ae8c-0a3...
I must have missed that folklore.
I got a trippy one that was supposed to be about Da Vinci painting the Last Supper, but the people were moving, so I thought it was supposed to be the actual supper: https://www.eggnog.ai/timeportal/56571c14-8f13-48ba-b60f-d82...
This game is much closer to Trivial Pursuit than it is Assassin's Creed. The importance of accuracy depends on the type of game and when the point of the game is specifically testing the player's knowledge, a lack of accuracy is game-breaking.
I do want to lend support to the idea that he series tries to balance accuracy with its gameplay/narrative foci, though. IIUC, they get a lot of details that Hollywood movies and even documentaries get wrong (by virtue of digital production). Origins and Odyssey included modes that allowed player to move through the world as if it were a living museum, reading passages about people and places as they were encountered.
Overall, it's nice when artists dealing with historical subjects understand that they're presenting imperfect models of the past, and state so, so that viewers can consider what might be accurate and what might not be.
Overall it's a fun idea though. I was able to consistently get pretty close in time and location so even with the anachronisms there was enough there.
This is an example of something that looks like an educational game but has the opposite effect. I had fun doing my round, but I won't do it again, because I think I will learn more wrong things from it than right things.
> This star-lit scene in a [...] garden showcases [...] using a seven-foot reflecting telescope, with speculum metal mirrors, in 18th-century [...]. The [...] row house architecture, oil lantern lighting, and his period attire all pinpoint the late 1700s. The meticulously ground mirror indicates the telescope’s custom design that enabled this revolutionary astronomical finding.
The result is a picture (with some panning animation) of a guy who for some reason has a pistol in his belt, looking through a complicated-looking telescope that is however definitely not a seven-foot mirror telescope in broad daylight with an oil painting inexplicably hanging in mid air behind him. No starlight, row houses or oil lanterns to be seen.
That said I too was a bit annoyed by how it rendered the byzantine empire. I thought Carthage was meant. So i was over by a thousand years.
A few super nitpicky comments:
- I dropped my pin for "Seward's Folly" on Alaska. The videos were clear enough that I knew that's what it was, which made me excited. But then it said it happened in Washington, DC.
- It might be sample bias, but I've only gotten events after year 0 (and technically, it went from 1 BCE/BC to 1 CE/AD.
I'd love to play with this my seven year old, but some of the images are too violent. A "PG mode" would be awesome.
The video of the signing in the White House shows Rutherford B. Hayes, not Andrew Johnson. Andrew Johnson was president in 1867, not Rutherford B. Hayes.
Your location estimate was off because you matched the 3-out-of-4 of videos showing Alaska / Russian Army / Tligits.
My time estimate was off because I matched the only video from the White House ...which was showing Rutherford B. Hayes in office.
Having said that, it's a first occurrence when I see AI-generated videos that provide something of a value.
OP: You might also want to change the title in your HTML header for the game - it just says Eggnog which is kinda funny but not sure if that was intentional.
Project "Eggnog" FTW!
Also some of the temporal clues were very good, some were 'wtf'.
I also laughed at some of the hallucination I witnessed. Like a group of people staring in a telescope pointing straight at a white wall.
Fun though, just needs to be honed in a little.
It would help to have markers on the timeline for the different ages, at least for the first round! e.g. Bronze age.
You already sort of do, being a Gregorian timeline and marking 0 AD as Christ's birth. That's a dead giveaway when you see crosses. So I think it would be fair and useful to give a range of eras as markers on the timeline.
The map could also be continental, and the locations more precise than the country.
The map could be more exciting, and change based on the timeline selection! It's currently showing the "current" map and not the map of the era; which in some respect is relevant.
Finally, the scoring could be more explanatory, you got 5,000 / 10,000 for the following reason / calculation method. Maybe a graph of points per time correction and location. It could also be more comprehensive scoring, with a slight multiplier for streaks, a badge for being good at temporal location vs. geolocation etc.
Scoring could animate up, to gamify the experience, create a sort of level end screen that builds up excitement. The map could animate and so could the timeline in this end phase.
I like the idea, there is a lot you could do to push this further.
It also dresses up the timeline so that the game gets its own identity.
The Alaska sale is in DC? One of the videos has the exchange of the flags on the post.
I was only off by 600 years.
I still enjoyed it, however.
I had one for the US purchase of Alaska which I got from the images of Americans building log cabins in an icy landscape and another image showing an American signing a document. I assumed it would be either Washington or Alaska (Anchorage I guess), but wasn't sure which because it depends on if you weight the signing of the agreement over the building of US settlements. It could have been either given the images were of different locations.
Similarly I had picture of British dude creating telescopes and realised it was very likely Herschel. But I also knew Herschels early work was done in Bath, while his most famous telescope was built later in Slough. Again, it wasn't entirely clear which location it would have been referring to.
Maybe I'm just being stupid though. I think you could have argued that right answers in both cases were more likely to be Washington and Bath.
That said, I really really liked it and think you have something here. Personally I'd play this over Geoguessr any day and I'll show my GF it tomorrow because I think she'll also like the history aspect of it.
Also, might worth lowering the distance penalty if someone guesses the right country, but the wrong point? Events in large countries are more risky just because of their size. Eg, if an event happened in France but you click Germany you'll often get less of a distance penalty than correctly guessing an event happened in the US but clicking the wrong part of the US.
I was 40 years off and less than 5km from the place and "only" got 7829. Assuming 99% accuracy for location, and based on some back-of-the-napkin math, that means you have to be within 70.7 years of the actual event to register any points for guessing the correct time. I think If you're within 100 years of the event you should get points. I think ideally it should be a curve, but if you can't program a curve then perhaps create brackets, like -10 points for each year within 10 years, -25 points for each year between 10 and 25 years, and -50 points for each year past that. Using this method you can be +/- 115 years of an event before getting 0 points for the time portion, and the closer you are, the closer to 5000 points you achieve.
Also, the one event was credited as being in Rome, although the picture shows, and the description says, St. Peter's Basilica, which is in Vatican City.
Fair point on the scoring, seems like a lot of people feel that way. Right now, half your points come from distance in km, and half come from distance in years. So you probably got 5000 points for location, and lost points on the year. But it's probably a bit too harsh right now.
And thanks for pointing out the Vatican City nit.
For dates, I think it needs to take into account how far back something occurred, with recent history being +/-5 years, early modern being +/- 50 years, and ancient history being +/-500 years. That's closer to the resolution we think of events in (decades, centuries, millennia). You could probably even update the UI to have these graduations.
Maybe it works for people who like having everything filtered through a modern cinematic cgi filter. In this case, sure, it is a neat tool for seeing how a hollywood studio might have imagined events of the past to look like. At least you admit upfront here that they are "fantastical imaginations" of historical events, but maybe you should clarify that on the website too.
I've always found it better to hear from actual historians, or better yet, dive into the source material when learning about events of the past. This takes some actual work and requires doing good research. It would be nice if AI could help those folks do their jobs more easily instead of being used to generate more fake looking slop.
If people dont want to spend time deep-diving history but are still interested I think this is an absolutely awesome way to start/learn. I would just suggest adding more ways to learn about the event itself with links to source material/wikipedia etc.
The whole point of this kind of thing should be to reward people who can recognize "that architectural style wasn't invented until the 13th century" but that's precisely the sort of thing image models cannot do reliably.
For example, consider this imagery from today's challenge: https://firebasestorage.googleapis.com/v0/b/fastab-f08e9.app...
These are some incredible monoliths: if they were real, I feel like I would have heard about them? And if they did... that's so cool. But because it's AI generated, I have a very low confidence level that this ever existed at all. Which is sad.
Which is funny, because the monoliths in the AI video look more eroded than the real ones today.
This looked like a nice idea at first glance. At second glance, it's really bad because you have to assume that everything you see in these videos can be wrong or misleading.
This is entirely possible, as the incredible accuracy[1] of non-generative picture location models (a very similar problem) shows.
[1] https://paperswithcode.com/sota/image-based-localization-on-...
Excellent idea and can’t wait for the next version.
I really enjoy time guessr! Playing it on the TV with family is fun too.
Constantinople wasn't captured by the Ottomans until almost 1,000 years after the events depicted in the quiz
I guessed it was supposed to be Agincourt because of the prominence of archers.
I built a prototype version similar to this called PastPort last year, but I like your idea better.
This uses Flux then image to video? Good quality generations, it would be wonderful to see the accuracy of the images improve. I saw you want to make it interactive like moving mode in geoguessr; that would be fantastic. I can imagine a few ways to do both.
I don't believe this is open source - is there a way to contribute to this? One man operation or are you a team?
This isn't open source, no. We're a team of two through our company Eggnog AI. That said, happy to talk some time if you're interested!
Boy is that overselling it. And you don't really need to, this is cool even though there's a ton of AI jank.
But overall, I love the concept and will probably continue to play and ignore the scoring.
This would make for a fun party game in the style of Jackbox.
On the Guess page it just has a blank screen with the 4 cinematic clips below...
I for the chinese question, I think it was way off. It showed the great wall of china, which is the wrong place for the answer.
The rise of the mugals was also interesting, I got it right but only because I assumed it was confusing Ottoman architecture with mugal.
Two small comments - if I know the precise year it is kinda hard to seek, especially at the ends of the scale. Maybe if you could hold the button to go faster?
And the text overlay on the map at the end mostly prevents me scrolling it (and seeing where the correct place was)
why don't you drag to get close enough and then use the buttons?
I think this might work better if you fed the LLM some real image and asked it to expand it than by using prompts.
https://en.wikipedia.org/wiki/King_Ezana%27s_Stele
And this is the Obelisk of Axum
https://en.wikipedia.org/wiki/Obelisk_of_Axum
You can see they have nothing to do with what is shown here, which is a weird cross of greek, middle-eastern, and stonehenge.
https://www.eggnog.ai/timeportal/93fb81ef-002c-459d-9726-b14...
---
Edit: dang suggested we should add more about the link to our startup, so here's more if anyone's interested:
Originally, we were making an AI video creator tool with a focus on character consistency in long videos (https://news.ycombinator.com/item?id=39853474). A year ago, character consistency was a big problem and we developed a solution for it in AI videos. Now though, a lot of other AI video creator tools have come out, and 1) character consistency has been solved in other ways 2) the other tools just make better quality videos than ours.
So we decided to pivot from building video creator tools to building apps with AI video as the core format. When I say that we plan to sell apps like Time Portal, it's still pretty open ended. First, we just want to build an app/apps that people really like using for free and figure out the best way to monetize them later. Time Portal is not for sale right now. It's free to play on the web and there is a free app in the App Store.
And I’d love to see the idea expanded further. “AI recreations of historical scenes” is an idea with tons of interesting potential.
One nit is recent events (1980+) will fall into an uncanny valley.
1. The AI imagery is often misaligned with the actual answer, with some anachronistic elements.
2. The scoring seems harsh. I got a couple of answers within 50km and/or 10 years in the time scale but was still severely penalized.
Good luck with this, I will definitely watch your progress and pass it along
I had real trouble with the battle of towton just now - the armour was “off” and someone was wondering around with a really cool white rose icon on their breastplate - and I could not work out if it was trying to be accurate or imaginative (accurate woukd look more like these things https://en.m.wikipedia.org/wiki/Armorial_of_the_House_of_Pla...)
I mean the fact that there were moving videos of Dutch astronomers or Ethiopian rulers is god damn amazing - it looks luscious
But it also looks … cut-scene. It’s brilliant. But it’s also a work of imagination (LLM imagings).
So it’s quite hard to do the game itself - but it’s amazing to drop people into the context and excite a historical interest.
Seems to be tied to the videos playing — when they start, the button disappears. I’m seeing it on iOS as well.
It looks nothing like the Aztec Pyramid of Teotihuacan, which is flat on top and has no structure.
In other words, this is AI slop that makes something that looks plausible, and is utterly misleading.
As someone who has spent decades of my life on history, this makes me weep for humanity.
@Samplank2 - this may hurt to hear, but your assertions that better models and pipeline improvements will solve this are pure cope. What you really need to do here is manually curate and tune the prompts, then cherry-pick with a fine eye for detail. There’s no substitute for actual effort and knowledge, but you seem disinterested in that part.
The thing that makes Geoguessr cool is that it drops you into a real place.
This is like if Geoguessr showed you the output from "Midjourney, show me China".
You are shown historical photos (from the past 100 years, even up to the past year) and need to guess the location on a map and the year of the image. Extremely fun and varied gameplay because of the varied events the photos capture (some mundane while others more recognizable).
It would’ve been cool to collect actual images from history. I’m sure there are 1000s of public domain images that could be used.
They literally can’t. There’s no photographs…
What’s the point? You’re just guessing at the prompt used to generate the photo.
That might be more interesting, actually. AI photo game where you need to guess the prompt.
The path forward as a foundation video model company closed, so they worked hard on end-to-end story creation workflows and mobile.
Turns out that's hard to gain traction and distribution amongst dozens of other similarly shaped startups. So they hack on games and fun viral loops.
Keep at it! This is super clever. You're getting noticed.
Video is going to be huge, and even though power law dictates there will be only a few winners, I think there's space for teams hustling this hard if you can find distribution.
Let the foundation video model companies fight to the death. They've over-raised and are being commoditized by Tencent and Alibaba's open source foundation video models (Hunyuan and Wan). You can use their APIs on the cheap and still provide value. And value will accrue to the application layer.
Focus on what the creators want and need.