Bertrand Charpentier

Founder, President & Chief Scientist @PrunaAI

Prev. @Twitter research, Ph.D. in ML @TU_Muenchen

[Bertrand Charpentier @Bertrand_Charp]

Today, we are happy to release P Video 2 Pro, Based on H3 from @MiniMax_AI .

Highest Quality, Highest Speed.

More benchmarks (and more models?!) coming soon!

[Pruna AI @PrunaAI]

Today, we launch P-Video-2-Pro, based on MiniMax H3: The best quality. The best speed.

Text or image in. Video and audio out.

• 𝗜𝗻𝗽𝘂𝘁: Text or first-frame image, with optional last-frame conditioning
• 𝗦𝗽𝗲𝗲𝗱: ~2.0s for a 5s clip at 480p · ~4.3s at 768p in Speed mode
• 𝗣𝗿𝗶𝗰𝗶𝗻𝗴: From $0.02/s Speed · $0.04/s Quality
• 𝗠𝗼𝗱𝗲𝘀: Speed or Quality
• 𝗖𝗼𝗻𝘁𝗿𝗼𝗹: First and last frame, seed, aspect ratio, and prompt upsampling with Off, Turbo, or Max

Available via our inference partners @replicate, @Cloudflare, @eachlabs, @scenario_gg, @TellersAI, @wavespeed_ai, @wiroai, @runware, and @togethercompute

Validated by our benchmark partners @datapointai, @RapidataAI, and @DesignArena. Benchmarks to be released soon 👀

👉 Free Playground: https://t.co/5CmcoZKi6O
📚 Model docs: https://t.co/epxE9Dobog
🧩 API: https://t.co/DVBQ8K71iX

[𝐙𝐞𝐧𝐠 💜 @zeng_wt]

Tried the new talking portrait model and ended up with something that felt more like a short memory.

Same person, same soft voice shown in different styles and ages.
The expressions and lip sync is 🔥

Made with the new P-Video-2-Pro-Avatar on @PrunaAI.

[Bertrand Charpentier @Bertrand_Charp]

You can now play with all the fastest models on one interactive playground!

[Pruna AI @PrunaAI]

𝐓𝐡𝐞 𝐏𝐥𝐚𝐲𝐠𝐫𝐨𝐮𝐧𝐝 𝐢𝐬 𝐛𝐚𝐜𝐤 𝐚𝐧𝐝 𝐛𝐞𝐭𝐭𝐞𝐫 𝐭𝐡𝐚𝐧 𝐞𝐯𝐞𝐫!

If you haven’t noticed yet, the Playground is available again, now powered directly by the API. This means:

1) Forget about sharing it with others and being limited
2) No more 5-generation limit -> Use your credits to create as much as you want.

Sign up now in the Dashboard, add your payment information, and get $20 in free credits (+100 video generations), with no other commitment required!

Start creating in the Playground today: https://t.co/NQcQumC4N3

[ComfyUI @ComfyUI]

Pruna P-Video-2-Pro Avatar will be available in ComfyUI!

[liam. @liambraus]

cuando descubres que puedes convertir cualquier foto de tus amigos en un vídeo hablando

EL GRUPO DE WHATSAPP NO VA A SOBREVIVIR A ESTO

[Bertrand Charpentier @Bertrand_Charp]

@Danzeronero @PrunaAI @MiniMax_AI Thanks @Danzeronero !

[Bertrand Charpentier @Bertrand_Charp]

@1littlecoder @PrunaAI @MiniMax_AI Thanks @1littlecoder !

[LordArtCry @lordartcry]

A single photo can become a video of a talking avatar, complete with natural expressions and lip-syncing.

P-Video-2-Pro-Avatar also supports audio alone or speech from text, up to 3 minutes. It's interesting to create content without needing to re-record it.

The 50% launch discount is valid until November 5.

@PrunaAI

[kai. @kaido_xr]

ceo: how much did that 60 second avatar video cost

me: $1.20

ceo: per second??

me: for the whole video

list price. $0.60 during the 2 week launch window

[tobarra @txbrraa]

el ceo de anthropic viendo que ahora puedes subir tu propio audio y que el avatar lo siga exacto sin desincronizarse

[Sara Han @sdiazlor]

I can't believe what they say on this talk show 👀

By P-Video-2-Pro-Avatar

[razornero @Danzeronero]

One image. Two ways to bring it to life.

P-Video-2-Pro-Avatar is here!
Powered by MiniMax H3, delivering realistic expressions, natural motion, and accurate lip-sync from a single portrait.

Your image. Your voice. Your story.
This is where AI avatars get REAL.
@PrunaAI

[Minette Kaunismäki @MinetteKaum]

P-Video-2-Pro-Avatar is live 🚀

[Bertrand Charpentier @Bertrand_Charp]

Animating a Dutch Golden Age painting 🖼️

[Bertrand Charpentier @Bertrand_Charp]

You can now animate all pieces of art in seconds 🎭

We are super happy to release P-Video-2-Pro-Avatar, creating the highest-quality avatars at the highest speed.

Try it on the Pruna API here: https://t.co/YaF4gKuKQw

[Pruna AI @PrunaAI]

Introducing P-Video-2-Pro-Avatar, based on @MiniMax_AI H3 !

Turn a single portrait into a realistic talking video with natural expressions and accurate lip sync.

And for launch, it's 50% off for the first two weeks, until November 5 at 18:00 CET!

• 𝗣𝗿𝗶𝗰𝗶𝗻𝗴: $0.02/s at 480p · $0.04/s at 768p
• 𝗦𝗽𝗲𝗲𝗱: ~0.97 s at 480p · ~2.2 s at 768p per second of output video
• 𝗜𝗻𝗽𝘂𝘁: One portrait with a text script or uploaded audio
• 𝗩𝗼𝗶𝗰𝗲𝘀: 30 voices across 10 languages
• 𝗗𝘂𝗿𝗮𝘁𝗶𝗼𝗻: 1s–3min at 24 fps, in 480p or 768p
• 𝗖𝗼𝗻𝘁𝗿𝗼𝗹: Voice delivery, scene motion, seed, and prompt upsampling

One image, two ways to bring it to life. Use your own audio or generate speech directly from text.

Available via our inference partners @replicate, @eachlabs, @Cloudflare, @scenario_gg, @wavespeed_ai, @WiroAI, Sociaro, @TellersAI, and @togethercompute.

📚 Model docs: https://t.co/4e81CKYbpQ
🧩 Try the model through the API or Playground: https://t.co/DVBQ8K71iX

[Bertrand Charpentier @Bertrand_Charp]

Animating a Greek statue 🗿

[Bertrand Charpentier @Bertrand_Charp]

@datapointai Congratulations!

[Bertrand Charpentier @Bertrand_Charp]

📄 Paper: “Beyond 4 bits: Data-Centric Post-training Quantization and Recovery”
👩‍🔬 Authors: @RheaSukthanker, @riccardograzzi, @jameshensman, @CameronPashmina
🔗 Link: https://t.co/FfJMYGX6CP

[Bertrand Charpentier @Bertrand_Charp]

Below 4 bits, the “implementation details” become the method. Representation, calibration data and recovery data decide whether a 2-bit model works at all.

1️⃣ Representation dominates: At ~2 bits, when scalar PTQ falls to chance, vector PTQ can maintain accuracy.

2️⃣ Calibration and recovery data must be chosen jointly: Using recovery data different from calibration performs better.

3️⃣ Recovery is cheap but model-dependent: while vector quantization recovery works well on some models with 1.4–30x less compute than any QAT, no recovery worked best on some models.

[𝐙𝐞𝐧𝐠 💜 @zeng_wt]

@PrunaAI @AskRayan Pruna is the best!💜

[Bertrand Charpentier @Bertrand_Charp]

@realrebelai @PrunaAI 🫡