kpcg

update: thanks to collective efforts in the discord (https://discord.gg/tHB3nzn2) there's now a piece of code in go (https://github.com/depy/RevenantRE) that can produce this: Cave/cavbones1.i2d Misc/dragonent.i2d Forest/forbirch001.i2d KeepInt/KInRug.I2D Now the plan is to start placing the images onto maps using map data

@pvtjokerr i started https://discord.gg/tHB3nzn2 for keeping the RE efforts on this, if you're interested in more realtime-ish collaboration, following https://github.com/WiegerWolf/revenant-map-builder/issues/1

okay, ross responded to this idea, saying he's not interested in recycling content this way (https://www.youtube.com/watch?v=sbNZ4LhxVHg&t=11184s). but, in case anyone among viewers wants to have a giant single movie file with all FM1 episodes stitched together, here's some scripts i used. i assume you have a folder with videos downloaded using `yt-dlp`, that looks like this: ``` user@pc /home/user/Videos/YouTube/accursedfarms/FreemansMind/1 > ls 'Freeman'\''s Mind - Episode 0 [gzY_Sjg7qgg].webm'* 'Freeman'\''s Mind - Episode 10.5 [VqqJ6euyHoY].webm'* 'Freeman'\''s Mind - Episode 10 [GTt6SoonzpY].webm'* 'Freeman'\''s Mind - Episode 11 [HiBh99OANs4].webm'* 'Freeman'\''s Mind - Episode 12 [8alyPGvqhoE].webm'* 'Freeman'\''s Mind - Episode 13 [BO686RdEthk].webm'* 'Freeman'\''s Mind - Episode 14 [kVmvVKRY290].webm'* 'Freeman'\''s Mind - Episode 15 [choM0W2tc78].webm'* 'Freeman'\''s Mind - Episode 1 [5SQhfkpX9bc].webm'* ... ``` and so on. in such folder you can run this python script (https://pastebin.com/eZncNaW6) to get info from sponsorblock api and cut each episode, removing things like into, credits etc. the script will create a sponsorblocked subdir that would contain all FM episodes, trimmed. to stitch them together, go into sponsorblocked dir and use this bash script https://pastebin.com/z6sgKn0m after all this you'll end up with a single output.webm file about 9 hours long. congratulations!

heres more in this vein https://www.worldlabs.ai/blog

okay, after spending about a week on this problem here's where we are. I think I'm close to figuring out the compression/chunked data format of the image files. here's the output I got for Imagery/Misc/dragonent for example: As you can see, it's not quite correct yet, but you can still kinda see a dragon there, it's just the chunks are incorrectly aligned or appear scewed. Here's another example Imagery/Forest/forbirch001: Again, you can kinda see a birch tree here. but the pixels are not aligned properly. What this tells me is that the decompression of each chunk works more or less as it should (I based it on ASM code in https://github.com/benjcooley/Revenant/blob/83f806bb65b762b2cc3e1ea509010f05342caa12/chunkcache.cpp#L249). The part that is missing is chunk alignment/sizing. On another note, I'm also having some troubles with palletes for 8bit images. For example, here's Imagery/Misc/blueflower: As you can see, it's not blue, but a yellow in fact. Here's another example Imagery/Misc/bread: Again, not quite the color of bread. The thing is, I'm short on time ATM, so unfortunately, I have to stop here for now. Maybe I will pick this up later when I have more free time In case, anyone would like to continue or learn from my efforts, I've uploaded my JS code to Github: https://github.com/WiegerWolf/revenant-map-builder and also attaching it to this post. It's a single long js file for Node.js. main.js.txt (mirror https://pastebin.com/UY7cbc2w) Oh and I also found this https://github.com/IgorZyktin/Revenant which contains some useful things too in python

Wow, this is huge find! Thanks a lot @pvtjokerr! With this it shouldn't take long to crack this puzzle! I don't know how I missed it, but thanks again for pointing to it! I'll get to work in it.

Thanks for the responses. To clarify the core idea of my post: The main suggestion is for Ross to create long-form compilations of his own Freeman's Mind series. The key benefits would be: Potentially increased revenue from these longer videos Minimal additional work required, as it's just repackaging existing content Ross would be capitalizing on the apparent popularity of these compilations, rather than other channels To further clarify the potential profit aspect: The real benefit comes from how YouTube monetizes longer videos. When viewers watch these long-form compilations (often several hours long) on autoplay, such as while sleeping, it allows YouTube to insert multiple ads throughout the video. We're talking about 6-7 ads across several hours, which can generate significant ad revenue. This is why creating these longer compilations could be particularly lucrative for Ross. It's not just about view count, but about maximizing ad placement opportunities in a format that viewers often consume passively over extended periods. So by creating these long-form versions of Freeman's Mind, Ross could tap into this revenue stream that other channels are currently benefiting from using his content. It's a way to optimize his existing work for the YouTube algorithm and viewing habits, potentially resulting in a nice boost to his income with minimal extra effort.

watching a recent video of spiffing brit i got an idea how @Ross Scott can easily boost his revenue, without too much work. recycling long series like freeman's mind and recutting them into long (several hours) form (watch brit's video for why this works and brings income). the proof is right here (YT search results for "freeman's mind"): a 2:47:10 hour video compilation has almost as much views as the finale of freeman's mind 1. so why some random youtuber profits from this and not ross? I even suggest asking LLM to write some sort of python script that just uses sponsorblock api to cut out all intro, credits from an existing freeman's mind YT playlist and just uses ffmpeg to stich the whole saga together and reupload it as is. so no manual labour required, as ross is as busy as he can be already.

based on `map sector file format` doc, i was able to conjure up this Modules/Map/*.dat file format spec: https://pastebin.com/XeMvt9aw sadly, i2d files are still a mystery to me. but thanks to this work: https://www.moddb.com/mods/the-forsaken/images/example-of-dat-images#imagebox image files with .dat extension (mostly in root game folder) are quite easy to extract, but they are not relevant for the task of stitching together a map (they mostly have game UI elements). a draft version of the file spec for images.dat files is here (its WIP, but with a bit of work it can be usable): https://pastebin.com/r52j4Wbt so, at this moment 2/3 pieces of the map extraction pipeline are possible (first, extract all *.rvr files as zip files that you can find in your game dir and sub dirs): read Modules/Map/*.dat files and extract TILEs from it (with coordinates, rotation, etc) - DONE (use https://pastebin.com/XeMvt9aw KaiTai scheme to generate a parser for your language of choice or just use https://ide.kaitai.io/ to check it out first) Using `unique_id` of each TILE in a Modules/Map/*.dat file, get the actual tile texture path - DONE (open `class.def` file in game root folder and search for 0xblablabla unique_id that you got from the tile on the map in step 1) Read the Imagery from the texture in path Imagery/Folder/file.i2d - NOT_DONE I've tried to reverse engeneer the format myself, but was not succesful in this yet. Any help of either reverse engeneering from scratch or better yet, somehow downloading this file https://web.archive.org/web/20000817185014/http://tux.telefragged.com/file.pl?filename=fileformat.zip&dir=telefragged/revenantqs/ If you know where to find it, it would be tremendously helpful in achieving step 3 here. it's a: Anyway, I started a very drafty file spec for i2d files here (but it's mostly trash at the moment): https://pastebin.com/UqCXNLry

found more lucrative things here: https://web.archive.org/web/20000817185014/http://revenantqs.telefragged.com/downloads.shtml Unfortunately the i2d file format spec is unavailable here, i'll see if i can find it elsewhere. I feel like solving i2d is last piece of the puzzle here

Couple of useful tools for this: https://www.moddb.com/games/revenant#downloadsbrowse the Map format reference is of particular interest: https://docs.google.com/document/d/1WAAecVsqAqZGLR4oOd3tpel7QC0YjrZOVuqwYt-sLnw/edit Also ingame builtin editor is of interest (press enter in game, type "lookunderthehood" and press F12) I don't have time to dive into all this now, but leaving my findings here for someone to pick up.

lo and behold. Luma did exactly what Ross was asking for, easy import of captured 3D data into Unreal

And here's attached zip file of a low-poly (because of the forum 8Mb upload file size limitation, full hi-poly geometry is about 70Mb in this example) 3d OBJ file that came out of it. With proper game settings (like first-person view and intent on actually capturing all angles of the level) the results would be better. This is what it looks like in 3D view (note, this was obtained from a YouTube video, not from GPU video memory locally): 83ac1cf7fd3e0ff01696cde41dd81cc06e5c8c3970489109a65bde6c94c16d22.zip

This is what Luma AI was able to create from about 20 seconds from this moment (from 0:02:41 to 0:03:15) out of a random videogame footage.

Yeah, maybe not all parts of dream software are done, but "capture gameplay video -> 3d static scene" seems to be possible with things like F2-NeRF today. teaser.mp4