Twitxy Saved

koyal.ai (YC F25)

Verified account This account is verified. Learn more

@KoyalAI

The AI Film Company

Built by researchers from @mit, @meta and @carnegiemellon, Backed by @ycombinator

An AI Research Lab purpose built for filmmaking.
koyal.ai Joined April 2025
18 Following 2.1K Followers 505 Posts

Latest posts 5

Pinned post

koyal.ai (YC F25) Verified account @KoyalAI
Star in your own movie today at xp.koyal.ai

- kookoo
xp.koyal.ai Koyal Experiences Star in your own movie. Create your own adventure.
Mehul Agarwal Verified account @meh_agarwal Aug 31
What if movies were not viewed but experienced?

What if you could star in your own movie?

What if you could change the ending?

xp.koyal.ai

- Choose your genre
- Cast yourself and your friends
- Change the storyline

Create your own adventure. All in 15 minutes.

Introducing Koyal Experiences.

Generate for free now at xp.koyal.ai
View quoted post on X
koyal.ai (YC F25) Verified account @KoyalAI
@OpenAI's GPT Image 2.6 Sunburst ranks among the leaders in image editing.

It places 2nd at the median across 19 editing benchmarks and wins 56.8% of its matchups. Its strongest placements:

- 1st, realistic style transfer (restyling an image while keeping its content)
- 1st, realistic color (recoloring a specific element in the image)
- 2nd in both styles, color palettes (applying a new palette to the whole image), at a 62.2% win rate
- 2nd in both styles, background (replacing what sits behind the subject)
- 2nd, animated lighting (relighting a scene in place)

Expression (changing a facial performance) is its weakest category, at 6th animated and 8th realistic. Semantic consistency (errors in perspective, scale and physics) is its weakest criterion. Worth knowing if facial edits are central to your work.

The model does what the prompt asks more reliably than anything else on the board. It ranks in the top three on edit accuracy in 15 of 19 benchmarks, the most consistent result of any criterion. A common failure in editing is a clean, plausible image that ignores half the instruction. A model that scores high on visual quality can still fail this way. GPT Image 2.6 Sunburst makes the requested edit, and it holds that across both styles, with the same mean rank in animated and realistic.

Measured over 60k+ blind pairwise votes.

filmarena.ai
koyal.ai (YC F25) Verified account @KoyalAI
Excited to share that @KoyalAI won the AI Innovation of the Year at the 10th Digital Studio M&E Awards 2026! 🏆

Honored to be recognized alongside industry leaders like
- Prime Focus / @dneg (Ramayana, Dune, Interstellar),
- Pixotope (virtual production & live sports technology)
- @FanCode (sports streaming)

For decades, filmmaking innovation has been driven by breakthroughs in cameras, VFX and production technology.

We believe the next breakthrough is making filmmaking entirely AI-native.

That's what we're building with Koyal Act 3: an agentic studio for creating and iterating on feature-length films.

Huge credit to Pranjal Chavarkar and our India team!

We're just getting started!

KOOKOO
1 replies 2 reposts 11 likes 551 views View on X
koyal.ai (YC F25) Verified account @KoyalAI
when our founder isn't running his startup.

stick to the end for a cameo by us :)
Mehul Agarwal Verified account @meh_agarwal Nov 2
I played an exaggerated SF tech bro so convincingly that I became one in The New York Times.

A few days ago, the NYT featured me in a story about dating in SF. Many questioned my dating criteria as an AI founder.

Fortunately, I performed a stand-up routine about these exact criteria 3 months ago, including "single until series b" (which yes, is not real).

After months of playing an AI-pilled SF tech bro, I'm not sure if I was method acting or if I just am one.

YOU CAN SEE EVERYTHING here:
View quoted post on X