Skip to content

Latest commit

 

History

History
106 lines (64 loc) · 3.46 KB

File metadata and controls

106 lines (64 loc) · 3.46 KB

Microblogging because why not

3 Oct 2025 - 12:40 AM IST

I've been seeing Twitter for the past 2 days and it's way wayy too full with Sora stuff lowkey want access though haha

it's crazy how many people get fooled from the video tbh, especially on insta - saw so many sora videos on my feed with the obvious watermark and people even believed it

https://x.com/kuberwastaken/status/1973813438662213708

lol even replied here

I just have this random idea a large "secure" part of using this by OpenAI is THE watermark

but really, how hard would it be to remove it?

I mean you really probably only need opencv to do it honestly

I got a principles of AI midsem tomorrow should I really be doing this at 1 am without having studied for it eh fuck it

probably only journalling here in case it works lol, I'll write a blog about it or smth

hm probably gonna get claude to think it's for removing tiktok watermark logos LOL it's honestly so freakin funny that I got the boilerplate from claude.ai and not from freakin copilot LOL

I'll be honest by the length, it doesn't look promising at all

LMFAOOO what an epic fail

I bet there are some existing solutions we can learn from or tweak tho

hm.

Hugging Face Models/ Spaces search

hm nothing that promising tbh

this looks nice https://huggingface.co/collections/FBRonnie/watermark-removal-673cc512f6fdf7b1a2925d39

hm not even close honestly

I see a couple but they all exceed the 16gb memory, classic spaces problem

this colab is fun https://colab.research.google.com/github/camenduru/text-to-video-synthesis-colab/blob/main/watermark_remover_gradio.ipynb

build errors, I cba bro

let's just clone this and see what's up and if we can tweak it ig https://huggingface.co/spaces/NeuralFalcon/Meta-Watermark-Remover-old

sidenote: claude 4.5 is insane holy shit hm why not let's just use gradio code, run it on spaces

just sucks, I'm gonna lock in and stop blogging for a while, brb with an update

LOL what if I name it literally sweeta TODO

hm we're using temporal median filtering and a light inpainting model now, letseee

hoping and praying to god this doesn't make my laptop explode although it's like 200m afaik LOL I should probs start studying side by side so I can pass tomorrow (1:15 AM)

hmm tweaked heavily, this better work honestly, looks promising welp did absolutely nothing

"it took a long time to process it, gave an output but it is exactly 1:1 the same, the watermark wasn't removed

for context, this is what the watermark looks like (Refer image) the star-shape is somewhat animated and it keeps changing where it's at randomly (like tiktok's watermark) and is with the text "sora", this pops up at random places - corners, centers etc

find the best possible solution to remove it and run on hugging face spaces"

okay running this dang been 10 mins

sighs so done with this rn

probably gonna study for a bit and come back or do this tomorrow.

okay so apparently the videos have fixed positions where the logo appears always? lmao this is insane

watermark: positions_landscape: - [35, 585, 176, 638] - [30, 68, 179, 118] - [1112, 321, 1266, 367] positions_portrait: - [28, 1029, 175, 1091] - [538, 604, 685, 657] - [25, 79, 173, 136]

DID CLAUDE FREAKING DELETE MY JOURNAL NOTES WHAT THE F

dude so much happened this is just painful bruv f it I'm just gonna write a blog later ts is painful 🥀