I just see a small chunk of the name but If you’re testing with this https://discuss.eroscripts.com/t/thelittlered-insane-1-hour-edging/291873 it’s sadly not representative of real world performance because that is in the training set. 
Other than that I’m not sure If I follow. I think you may to loosen the dwell setting so it’s less eager when creating these plateaus.
This creates an impressive result. Do you plan on expanding it to support multi-axis? Or even on certain axis anyway would be a big benefit.
probably not. don’t have enough data for that nor am I sure it would even work.
How about scripts for vibrators?
also no. same difference. I don’t have enough data for that
Bro this is fire 
I compared GoblinScript v0.2.0 against my manually authored script for the same 10-minute VR scene, and I’m genuinely impressed.
The generated script contained 2,049 actions versus 2,042 manually placed actions. The best global offset was exactly 0 ms, motion correlation reached 0.84, and around 85% of my manual reversals had a generated reversal within 100 ms.
Most of the visible differences occur during transitions, where I intentionally scripted things differently, or in the depth/amplitude rather than the rhythm. During sustained action, the timing frequently overlaps almost perfectly. This looks like a genuinely useful first draft rather than random motion detection.
This is seriously impressive work. Well done !
All hail the goblins !
I just ran it against the whole of Shrek (2001 VHS) on the Internet Archive, and frankly, it didn’t do a very good job at all.
I assume the goblins were too busy enjoying the movie to focus on the task at hand!?
jokes aside I did add 20 hours of youtube movie trailers and movie scenes to the training set for the model to not predict any motion. so I’d say it’s working as intended 
That’s really smart. I’ve actually not tried v0.2.0 for anything other than Shrek yet but I will during the course of this week.
What are you thinking the next iteration will bring - any plans as yet?
well I hope it’s going to be marginally better. I’m just letting claude try a million things and hope that it has a breakthrough. I’m not sure how much more room for improvement there is, probably not that much. I want to fix the phantom stroke issue introduced in v0.2.0.
I do still believe more training data is still helping but quality is such a big factor. sometimes it feels like just one bad sample can make the model universally worse.
let’s hope that meta graces us with some more upgrades to their v-jepa models or other open source competition in that space. this whole thing wouldn’t be possible if they didn’t provide such relatively cheap to run distilled models.
in conclusion I don’t expect anything super transformative to happen
Meant to post this a while ago. These goblins are far from perfect, but I was able to achieve shockingly decent results for essentially a hands-off tool! Just point it at a file and wait.
I have used multiple AI/automated scripting tools with all sorts of manual sliders and settings, and achieved worse results. I would still want to edit anything by hand, but if youre really desperate for a script, RIGHT NOW, this is a solid solution.
I’m curious to see how this develops.
Added another update Releases · ofs69/GoblinScript · GitHub
Hello goblins!
A small Chinese-speaking crew has found GoblinScript. The scripts are easy enough to understand, but the picker and review bench are still speaking only English.
Would it be possible, someday, to keep the goblin words in a little language file beside the exe? Something like languages/zh-CN.json would let us translate the screens without taking a screwdriver to goblinscript.exe.
English can stay as the fallback when a goblin loses a word. We would be happy to help teach the crew Simplified Chinese.
/\ , /\
(^.^)
|___|
/ \
Have fun. hope this helps Release v0.3.2 · ofs69/GoblinScript · GitHub
Edit: I added v0.3.2 fixing a bug on the review page
You spoil us
Is it possible to run this on a Network Drive? I only see my C & D folder.
Will it run on Sub-folders as well?
Ran this on one local file, wow results are great. Awesome work 
Not sure what you mean with sub-folders but I guess you need to map it to a drive letter.
I got mine mapped to Z: and it works
Thanks for your help and your reply,
I was using Network Location instead of mapping a drive, your tip has helped me resolve that one! 
As for the question regarding Sub-folder my media storage structure is as below, meaning each video file is in its own sub-folder inside a larger folder. So would it be pussible to select the larger Folder to handle all files in folders within the bigger folder
no I don’t think that is implemented. you can go in a directory select all and go out again into the next directory select all again. it does remember selections across directories.
Maybe I can add a one layer deep selection so that you can select a directory and it processes every video in it but no recursive processing.
That would be greatly appreciated if that is something that could be implemented.
