something like “Latest AI Breakthroughs: Open‑Source Video Generation, Real‑Time 3D Motion, Mobile Upscaling and Audio‑to‑MIDI Tools”

We need to rewrite transcript into SEO blog post, 800-1200 words, using only information from transcript. No hallucination. Must not copy sentences verbatim; fully rewrite. Must include sections: SEO_TITLE line, intro, at least 3 H2 sections, bullet points where appropriate, conclusion, SEO_KEYWORDS list.

We must stay within length 800-1200 words. Let’s aim ~900 words.

We need to ensure we only use info from transcript: topics: open-source AI catching up to frontier, AI video generation locally on mobile phone, using phone to run Quen 3.6 and Jam Packets? Actually transcript says: “we can also now use one of the best consumer open-source models, Quen 3.6 and Jam Packets run locally on just a phone.” Not sure exact but we can mention.

Google releases an open source model that allows precise control of heads and faces.

Open-source AI that can make any character dance to any music.

AI turning full song into notes per instrument (audio to MIDI).

OpenAI reveals a new model designed to attack other AIs.

Ultra-fast image upscaler and background removal.

Ridiculous humanoid robot demos, including humanoid cage fight.

Nvidia releases AI called RD (realistic 3D human movements real time). Input text prompt, movement, optional controls like paths, waypoints, key poses, joint positions. Generates continuous 3D human animation real time, can stream forever. Two-stage planning: overall position then detailed body motion. Code released, model size 765 MB. Can control humanoid robot Uni Tree G1 in simulation.

Mobile video generation: Mobile One (Alibaba’s model) runs on phone, 5-sec video 480p 16fps in ~20 sec. Achieved by chunked generation, pruning attention heads, reducing steps to 3, memory-efficient decoder. Code released, total size 9.9 GB.

Nvidia PID upscaler version 1.5: better detail and color fidelity, works with image models Quinn image, Flux, Z image. Fast: 512×512 latent to 2K in few seconds on consumer GPU. Code available.

Audio to MIDI by Mirell: upload full song with vocals and multiple instruments, generates separate MIDI tracks per instrument. Free online via ML studio. Detected voice, drums, synth strings, synth pad; bass not detected but can add manually. 80% accurate. Open source, three variants small medium large, large 1.4B parameters.

Also mention open-source AI catching up to frontier, maybe mention open-source models like Quen 3.6, Jam Packets.

Make sure not to add info not present.

We’ll produce sections: maybe “Breakthroughs in Open‑Source AI Capabilities”, “Real‑Time 3D Human Motion Generation with Nvidia’s RD”, “Mobile‑Friendly Video Generation and Upscaling Innovations”, “AI‑Driven Music and Robotics Demonstrations”. Need at least 3 H2; we can have 4.

Need bullet points where appropriate: e.g., features of RD, steps for mobile video generation, benefits of upscaler, capabilities of audio-to-MIDI.

Make sure not to copy sentences; rewrite.

Word count: Let’s draft ~950 words.

Will need to count approximate words. We’ll aim ~950.

Let’s write.

First line: Now intro paragraph ~80-100 words.

Then H2 sections.

Let’s