I Yosun Chang

How I went from obsessing over “best AI video model” to optimizing for pipeline intelligence after NVIDIA Nemotron 3 release... (This is Big) ⚠️ Nemotron 3 silently got released this month... And in simple words, it signals the next phase: → models will get 10x more efficient → and the cost of “developing a model” will trend toward near-zero $ Nemotron is the direction: → big model on paper → small part active at inference → you can budget how hard it thinks → it can run on your infrastructure In translation ? You stop paying for max intelligence all the time. You pay for the intelligence you actually need. That’s why the economics change. You already guessed it, it is a heaven for large companies. Produce AI video production at x10 less the cost in Tokens. With ANY video model tool. Here’s my view of what happens next. Models won’t be scarce. It fact they will become a commodity and mostly open-source. They’ll be internalized. And they’ll all be excellent at generating Media assets. So the competitive edge won’t be: → “who has the best model?” It will be using Agentic to know: → who runs intelligence with the least GPU consumption → who gets the fewest iterations to a usable output → who can think smarter with less electricity → who can run on smaller cards, reliably → who can keep continuity stable at scale New KPIs will replace token obsession: → reasoning efficiency per task → decision-to-action latency → iteration count to approval → continuity stability across assets → compute cost per deliverable And yes, that also changes the API story. Today, teams track token spend because the intelligence is rented. Tomorrow, the KPI is not “API consumption.” It’s “infrastructure efficiency.” The future of AI production is not “better models.” Forget about them. Agentic is the future of our society. And regarding the production industry, agentic will change everything from its core. Because media production performance becomes an intelligent agentic production line. Not an API bill. Controlled intelligence that is cheaper, faster, and economically predictable.

4

Biological Chaos Entropy: What Deepfakes Can Never Imitate The newly introduced Chaos-D-ID SDK represents a fundamental departure from conventional liveness-detection technologies that rely primarily on visual facial pattern analysis. Built upon HSKG Labs’ core patented technology, Bio D-ID, this system leverages biological chaos entropy—a class of signals that are inherently unpredictable and non-deterministic. Through standard camera sensors, Chaos-D-ID continuously extracts microscopic blood-flow fluctuations (remote photoplethysmography, rPPG) and heart rate variability (HRV) in real time. These signals are not static biometric markers; they are dynamic, stochastic physiological phenomena governed by complex nonlinear biological systems. Crucially, such signals cannot be replicated by AI-generated deepfakes, synthetic avatars, or even high-fidelity 3D masks. No matter how visually convincing a synthetic face may appear, it lacks the irreducible physical entropy produced by a living human body. Chaos-D-ID therefore establishes a new verification paradigm—one that does not ask “Does this face look real?” but instead proves “Is this entity biologically alive, right now?”

2

Luma AI just dropped Ray3 with an impressive demo promising 4K HDR video. I decided to put it to the test. No cherry-picking, just raw results from 2 hours of use. Site: https://lumalabs.ai/ray 4k download link to watch without compression: https://lnkd.in/eBfte2gX ✨ What stood out: -Interface is smooth you can queue shots and run multiple variants at once. -Cheap to run half of my $30 credit gave me ~100–150 generations, draft mode is a great feature! -Camera motion looks strong. -Fast turnaround. ⚠️ Where it struggles: -720p looks solid, HD is fine but 4K falls apart. The upscale looks more like low-poly CGI than real 4K. -Prompts aren’t followed well. Even simple prompts like “static camera” took multiple retries. -Every usable shot took 6-10 variants. -Close-ups look good, but medium/wide shots suffer from the usual AI quirks. 🔥 What really annoyed me The tech demo on their website looks like absolute fire. Trying to recreate those shots was a bit of a failure. Sure, I managed to get the camera motion and scene in there but only after a shotgun approach of endless iterations. And even then, the image quality never came close to their demo material. It suffers from the same issue as Veo3: amazing marketing, but the actual results require endless “slot machining.” 💡 Takeaway Ray3 has potential the speed and pricing are great but right now, Seedream still delivers stronger results overall for a cheaper price. Curious if anyone else here has tried Ray3 yet. What’s your take? 4k link: https://lnkd.in/eBfte2gX https://lumalabs.ai/ray https://x.com/lumalabsai

76

37 Comments

Adopt XR Labs just integrated MetaXR SDK with UE5 to enable Real Time Handtracking on the MQ3. Up next, Depth Occlusion and Scene Detection API! Only if UE5 Source Builds build properly 🤦‍♂️ #unrealengine #meta #metaquest #mixedreality #virtualreality

7

Latent Forcing: Reordering the Diffusion Trajectory for Pixel-Space Image Generation Link: https://lnkd.in/esveDAjy Authors (dm - add / change) Alan Baade, Eric Ryan Chan, Kyle Sargent, Changan Chen, Justin Johnson, Ehsan Adeli , Fei-Fei Li Abstract #Latent #diffusion models excel at generating high-quality images but lose the benefits of end-to-end modeling. They discard information during image encoding, require a separately trained decoder, and model an auxiliary distribution to the raw data. In this paper, we propose Latent Forcing, a simple modification to existing architectures that achieves the efficiency of latent diffusion while operating on raw natural images. Our approach orders the denoising trajectory by jointly processing latents and pixels with separately tuned noise schedules. This allows the latents to act as a scratchpad for intermediate computation before high-frequency pixel features are generated. We find that the order of conditioning signals is critical, and we analyze this to explain differences between REPA #distillation in the #tokenizer and the diffusion model, conditional versus unconditional generation, and how tokenizer reconstruction quality relates to diffusability. Applied to ImageNet, Latent Forcing achieves a new state-of-the-art for #diffusion #transformer-based pixel generation at our compute scale.

22

2 Comments

EON Reality’s Virtual Campus Custom Builder lets institutions stand up fully branded XR learning environments in hours—not months—by auto-converting existing PDFs/PPTs/Docs into XR-ready courses. What it automates: • Brand extraction & enforcement: logos, colors, typography applied consistently • Multi-format ingestion: .pdf / .pptx / .docx → unified, semantic course blocks • AI mapping: dedupes files, infers modules, outputs program.yaml & course pages • XR-ready outputs: microsites, Train-AI docs, tool cards, JSON for Brainy/LMS • Compliance & quality: WCAG checks, validation reports, zero-silent-failure logging Why it matters: pilot in ≤1 week, scale campus-wide in 90 days, cut build costs by 80–95%, keep faculty in the loop with approve-to-deploy, and update curricula in minutes—not semesters. Request a demo and sample automation report to see your content transformed end-to-end. #XR #VirtualCampus #EdTech #AIinEducation #InstructionalDesign #HigherEd #WorkforceDevelopment

1