Pinned Tweet
A thread about the subtle detail in #KashmirFiles that most of you missed.
1/n
ಗಣ retweeted
JUST IN: Riemann Hypothesis has been solved.
alpo.ge/remark.pdf
ಗಣ retweeted
FastVideo is open-sourcing Dreamverse Harness 🚀
Since we released FastH3, the community has picked it up widely, especially on local machines. But a polished film takes more than a single clip.
FastVideo Dreamverse Harness enables any agent to work with you through the whole video production, from the first script to the final cut. And you can run all of it fully local on your own NVIDIA DGX Sparks or through cloud providers.
📝 Blog: haoailab.com/blogs/dreamvers…
💻 GitHub: github.com/fastvideo-project…
@NVIDIARTXSpark @NVIDIAAI @MiniMax_AI
#OpenSourceHiggsfield
ಗಣ retweeted
Upcoming Talks in Bengaluru
Dear friends and supporters,
Hope to see you at one or more of these upcoming talks in #Bengaluru .
Culture as Living Knowledge: The Continuum of Indian Civilization
NMKRV- Date : Saturday, 21-11-2026, Time: 11 am
Historicity of IKS: Focus on Math & Astronomy
Chanakya University - Date: Monday, 23-11-2026, Time: 11 am
Uncovering India’s Deep Past
IIWC, Basavanagudi - Date: Monday, 23-11-2026 Time: 6 PM
Please forward in your circles. 🙏
Wow! Is this the first time I feel I have ever agreed with RaGa and I feel he agrees with the rest of us Indians!
Pakistani man: Is there room for India and Pakistan to move beyond blame game and work together?
RG: Pakistan supports terror in our country. It is absolutely unacceptable that our innocents are killed and we still talk to PAK. Once it stops there may be a possibility of talks
72
ಗಣ retweeted
Latent diffusion needs no decoder😵
New paper: DWD, a linear head on DiT predicts pixels directly
• Same or better quality on every tested latent model, less flicker in Wan video
• Even pixel models improve: ImageNet FID 1.55 ➡️ 1.43 (pixel SOTA), photorealistic texture in T2I
Wow! Wondering if the quality of the denoising will be as good as PIKA as well and that too in 2-4 steps?. Thats my only worry. But great work! Really excited to try it out. #MiniMaxH3
Introducing pi-H3⚡, the fastest MiniMax H3 yet. 10 s of 768p video with audio generated in 9.2 s on a single B200.
How we did it:
Direct pixel output in 2 or 4 steps, no video VAE decoder.
Blog: hanshengchen.com/blog/pi-h3/
App: huggingface.co/spaces/Lakoni…
@StanfordAILab @LumaLabsAI
61
Cant agree more. I have first hand seen both the sides of the coin.
India is firstly a very competitive society. A competitive society produces great people, but I wish the prize at the top was sufficiently rewarding too. Europe and US are competitive too - but not as much as India is, but the prize at a the top is far more rewarding.
So the main reason most Entrepreneur's and Indian Software Engineers gravitate towards Europe or US is mainly this - its much easier / less competition and greater reward for your efforts.
Path of least resistance !
25
At the end of the day even though #Adipurush sucked as a movie, its song still lives with us ( #RamSiyaRam ). I dont think @WorldOfRamayana will have the same impact, even though its done by "legendary" music director! A R Rehman has lost the magic.
31
The thing that I notice more is the giant woman from India who towers over them all and I think she is his translator! What genes!
17
I have just finished writing Book 1 of a series of 5 books. Its a YA fantasy (think Harry Potter!) and it introduces #Ramayanam to them in their own language. I am looking for a set of 10-20 early readers who I can send the book for feedback and corrections. Any takers with 10-20 yr olds please DM me ? Just looking for honest feedback and their reactions.
2
2
30
Just the Camera work is amazing!!!
I dont think this was AI - but IF it is really AI, salute to the person who did it, I couldnt tell and having created AI movie myself, I respect the quality here!
ಗಣ retweeted
Rishi Sunak: South Asian descent
Kamala Harris: South Asian American
Diljit Dosanjh: Punjabi and South Asian
Muslim Groomer Bilal Kazi: Indian
Sky News is a bunch of Islamist Racist Fucks @SkyNews
ಗಣ retweeted
"𝐔𝐧𝐭𝐢𝐥 𝐭𝐡𝐞 𝐋𝐚𝐬𝐭 '𝐁𝐫𝐚𝐦𝐡𝐢𝐧' 𝐂𝐨𝐦𝐦𝐢𝐭𝐬 𝐒𝐮𝐢𝐜𝐢𝐝𝐞,
𝐑𝐞𝐬𝐞𝐫𝐯𝐚𝐭𝐢𝐨𝐧𝐬 𝐒𝐡𝐚𝐥𝐥 𝐂𝐨𝐧𝐭𝐢𝐧𝐮𝐞!!"
Now please imagine, if someone would have used the same words, wishing the annihilation of any another community, wouldn't the person rightfully be in Prison!
Here people in the auditorium are clapping wishing death of a community!
@OGTeamBharat
I think in this time and day, if someone wrote a verse as beautiful as this, it would be scandulous --
Consider another example from our Sanskrit textbook during school days -
रय भिन्न वल्कल - The upper clothing that parted due to her speed.
which was the euphemistic version of the original according to our Sanskrit teacher who smilingly said, it had to be
स्थन भिन्न वल्कल - The upper clothing that parted from her breasts.
Now imagine writing something as "vulgar" or "blasphemous" as this for Devi Parvati.
A fun little anecdote for Children’s Day!
When little Skanda was telling Brahma about how his mother’s lip got bitten by Shiva.
Sanskrit verse by the poet Chittapa (included in the comment).
34
Do you know why the West - from the British Raj to Navarro today - keeps targetting Brahmins?
Reason- Brahmins led revolts, guarded knowledge, and were the biggest impediment to conversions.
The fear that haunted the 1800s still drives the hate now - here's the history.
ಗಣ retweeted
4 ways to take a 768p MiniMax H3 clip to 2K, head to head on my 5090.
Same four 5 second H3 drafts through each one in ComfyUI:
- Open-H3-Regenerate-2K (@NVIDIAAI, the original)
- Open-H3-Regenerate-2K-IC (NVIDIA's new in-context LoRA via Efficient-Large-Model)
- LTX Restore IC-LoRA (@ltx_io, on LTX-2.5)
- ReCaVSR 4x (USTC, one step, int8 @ComfyUI port by PedroMarinhoDev)
My numbers (mean per 5s clip, whole ComfyUI job):
- Regen-2K: 308 s, face match** 0.25, detail 15.9x
- Regen-2K-IC: 504 s, face match 0.64, detail 2.9x
- LTX Restore: 363 s, face match 0.70, detail 4.5x
- ReCaVSR 4x: 47 s, face match 0.78, detail 10.0x
**Face match is ArcFace against the draft, 1.00 = the same face. Detail is sharpness against a plain resize.
Honest take: ReCaVSR is the one. It is about 6.5x faster than the next fastest, it keeps the person, and it still adds a ton of detail. The original Regen-2K adds the most detail of all but it redraws faces, 0.25 is a different person, and it loses all lip-sync. The IC LoRA fixes exactly that, which is the whole point of it, but on my card it is the slowest lane and adds the least detail, preserving more of the original draft. LTX Restore was built for old film and still held up on clean H3 footage, giving the Regen-2K-IC LoRA a run for its money!
All local. ComfyUI 0.45, one RTX 5090.


























