ๆญฃๅœจๅŠ ่ฝฝ่ง†้ข‘...

่ง†้ข‘ๅŠ ่ฝฝๅคฑ่ดฅ

๐Ÿ‘€ I built an AI NBA ๐Ÿ€ commentator using OpenAI Vision API + TTS magic in 15min. Clip taken from when Damian Lillard ended PG's career with the game winner in 2019. Same pipeline can be applied for any highlights and buzzerbeaters.

223,614 ๆฌก่ง‚็œ‹ โ€ข 2 ๅนดๅ‰ โ€ขvia X (Twitter)

10 ๆก่ฏ„่ฎบ

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

๐Ÿฅน

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

My heart is full:) If you enjoyed this, follow @thealexker for more on applied AI, startups, and project insights and let's learn together. High-signal content only, no fluff.

๐Ÿ‡บ๐Ÿ‡ธTERRENCE ๐Ÿ‡บ๐Ÿ‡ธ ็š„ๅคดๅƒ
๐Ÿ‡บ๐Ÿ‡ธTERRENCE ๐Ÿ‡บ๐Ÿ‡ธ2 ๅนดๅ‰

Only thing missing is excitement and "from deep"

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

Yes!

Justin Halford ็š„ๅคดๅƒ
Justin Halford2 ๅนดๅ‰

With proper intonation and a little bit cleaner choice of phrasing, this is a killer product! Great job ๐Ÿ”ฅ

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

Thanks Justin! Iโ€™m just surprised the default text to voice is already so natural albeit a bit emotionless

chance ็š„ๅคดๅƒ
chance2 ๅนดๅ‰

Would be even cooler to delay video by a couple frames so the commentator can prepare with enthusiasm

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

Itโ€™s hard to get the timing right but ideally they would talk a little faster

jeffw ็š„ๅคดๅƒ
jeffw2 ๅนดๅ‰

Honestly SO good! Dagger. ๐Ÿ—ก๏ธ No trouble with gpt-4-vision rate limits?

Alex Ker ๐Ÿ”ญ ็š„ๅคดๅƒ
Alex Ker ๐Ÿ”ญ2 ๅนดๅ‰

fixing the rate limit is most of the work haha. at 30fps I sample every 150frames or every 5 seconds to get around it. also the video is short so there isn't a lot of output tokens. the limit is 10k/min atm.

็›ธๅ…ณ่ง†้ข‘