Sherman Sall

ComfyUI has a shipped an incredible upgrade 🌊 The ACE-Step 1.5 is going to turn AI music into a real AI production asset And this will go far beyond creating cool demos. It plans full song structure with a language model (lyrics, sections, arrangement), then a diffusion transformer generates the audio. This results in complete songs within 10 seconds, all running locally on 4GB VRAM, in over 50+ languages, with track coherency. Winner teams need to treat this upgrade as a part of their infrastructure 🎯 If you run an agency, brand, or production house, the real question is: “How do we turn this into a repeatable, compliant, internal capability - not a one-off stunt?” Here is your guide for implementing this upgrade into an efficient pipeline: 1. Design the song workflow: Define standard song templates per use case → 15s social stings → 30s brand tags → 90s hero tracks for films For each template, standardise: → Structure (intro - build - hook - outro) → Allowed genres and moods per brand → Language rules and lyric constraints Encode those rules inside your ComfyUI workflow, not inside one-off prompts. You are running a controlled, documented workflow that any trained operator can trigger and reproduce. 2. Wrap ACE-Step 1.5 inside an internal, ownable pipeline: Open-source inside a framework creates advantage. Treat ACE-Step 1.5 as one component inside an internal audio pipeline: → Run it inside your own environment, close to your brand or client data → Version-control models, LoRA styles, and workflows just like code → Attach metadata to every render (prompt, model version, use case, license notes) → Define routing: where the track goes next (editor, mixer, supervisor) and how feedback loops are handled The tool is ACE-Step 1.5 and the moat is the pipeline you build around it. 3. Train operators like you train editors: Create a small “AI Audio Operator” track inside your team: Define the role: who is responsible for prompt design, LoRA management, and quality control and build a short playbook: → What “good enough” sounds like for each client → How to iterate when the output is 80% right and 20% off → How to document final decisions so the next project starts at version 1.5. → Pair operators with supervising creatives at the beginning, then slowly decouple as standards are met consistently This way, you are giving them an internal engine that multiplies their output while keeping control, compliance, and quality. The pattern here is simple: Tool → Workflow → Blueprint. ACE-Step 1.5 shows what happens when open models reach commercial-grade quality on consumer hardware. The advantage will go to the teams who integrate it into a documented framework ⚡ P.S. Do you want to integrate an operationally precise pipeline into your AI production workflow? Explore how the Xtendency™ Blueprint can help your business ⬇️ → https://lnkd.in/eTxfbAVp

5

I was left a little perplexed by the response to the OpenAI/Jony Ive video discussing how Agentic AI actually empowers us to be in better control of our technology. By and large, the vast majority of comments were pessimistic—and at worst, combative. I mean… here are two people with enormous platforms and resources, thinking about how to paint a healthier relationship with technology for consumers. One that should put people in positions of power rather than turn them into servants to their devices. But then I remembered the reality: a lot of the folks building these devices are the same ones who built the last ones—the ones that left a majority of people feeling disempowered and disconnected from the realities outside their screens. No wonder they’re angry. Can you really blame them? History tells us that when there are generational leaps in computing power, those moments come with a shift in interaction modality. It also tells us that while large companies may have the distribution to introduce new modalities to the masses, the most innovative design tends to come from the edges—not from those previously incentivized to create feedback loops that kept us glued to our screens. People don’t trust big tech to provide the solution because big tech is the one that left us with the problem. The next interface revolution should feel like an extension of yourself: intuitive, seamless, and—most importantly—quiet until you want it. It’s not a buddy or a fake AI girlfriend like in the movie Her. It’s a butler who’s always asleep on the job until you wake it up to solve a problem. Then it handles it instantly, and goes back to sleep. People are looking for more reasons to be and feel alive. The technology to help us do that is already here. I’m glad that some of the previous generation’s designers are finally trying to solve the problem. But I’m more excited about the people whose brains hold the keys because they lived with the problem every day— not the ones who made money off of creating it.

18

6 Comments

Most AI video models can move. Few can remember who they’re moving. BindWeave changes that. This new 70GB model from ByteDance and USTC tackles a beast most video generators still fumble: keeping characters consistent across frames, actions, and interactions. One reference image is enough. You get full-body consistency, facial identity, viewpoint shifts, and even multi-subject interactions, all stitched together with a coherent logic. The trick? A multimodal LLM first breaks down your prompt into “who’s doing what, where, and how.” That structure feeds into a diffusion transformer that generates the final video. It’s not just pretty motion anymore - it's controllable, story-aware generation. Some highlights: 🤌 Consistent character identity across time and scenes 🤌 Works with multiple subjects + human-object interaction 🤌 Strong grounding of roles, poses, emotions 🤌 High performance on subject consistency benchmarks Workflow isn't public yet, and the full model weighs in at 70GB, so not exactly plug-and-play. But you can bet someone’s already working on a quantized version. And then there's VACE. VACE is more of a multi-tool. You get ref2vid, vid2vid, inpainting, region edits, scene extension, and it’s already out in the wild. It’s fast, flexible, but messier with identity. You can do more things, but you give up some precision. So the comparison’s simple: 💪 VACE - usable now, broad capabilities 💪 BindWeave - future-focused, tighter control over who appears and what they do 🔗 Git: https://lnkd.in/duc7K5wk 🔗 HF: https://lnkd.in/dvyFAVXX We're following both closely at Lighthouse. For those of you building cinematic GenAI workflows - storytelling with structure, characters, camera logic - these models are worth tracking. Cohorts open - grab your seat! Come build with us.

41

9 Comments

This is how Large Language Models (like ChatGPT) work: GPT stands for “Generative Pre-Trained Transformer”. Let’s break down what this means. Generative: this means the model can write (or “generate”) new text. As opposed to simply analyzing or processing inputs, it can output totally new, unique text. Pre-Trained: when you hear about AI being trained on massive data sets, this is what they are talking about. In the pre-training phase, they give the model tons of data, which in the case of LLMs, it uses to learn the patterns of language as well as general knowledge. Transformer: this is the key that has made all modern LLMs and the recent explosion in AI possible. In 2017, a team at Google introduced transformers in a landmark paper called “Attention is all you Need.” Transformers allow the model to analyze every word in a sentence simultaneously in order to understand which ones are most important. (Previously language models tried to analyze words in sequence and wound up having memory issues). — Ok, so how does this all work? At the beginning they start with a shit-ton of data. They can’t just dump it all on the Large Language Model, it needs to be cleaned up and formatted properly. This takes a lot of time. Next, all the words need to be turned into numbers. Computers don’t actually understand words, so the words are broken up into “tokens” and assigned a unique number identifier, which the computer can use to reference that word. They then use these numbers to create an insane multi-dimensional map of the entire language. And when I say multi-dimensional, I’m not talking about 3D space like our puny human brains can comprehend, I’m talking about THOUSANDS of dimensions. They place these word-numbers (“vectors”) into a multi-thousand dimensional space map. This is essentially a meaning map that allows the computer to comprehend language. All existing words are placed somewhere on the map based on their meanings and their relations to each other. This is what it uses to predict words. The next phase of training is basically playing a prediction game where the model is given examples of sentences and asked to guess what the next word will be. If it guesses wrong, its “weights” are tweaked and then it tries again. This happens over and over again billions of times, until it gets really good at predicting words. Once it reaches a good place, its weights are fixed. There’s an additional layer of “fine-tuning” for personality, telling it what its role is, what style it should communicate in, and setting some parameters around how it should “behave”, but this is the basis of the mechanism it uses to generate text in realtime when you send it a prompt. — I’m curious if this makes sense or is still too technical? My goal is to make technical stuff like this more accessible, so would love any feedback if you’re feeling like: “what the hell are you talking about?”

3

Automation won’t kill your job. It’ll kill your positioning. Anthropic just launched Agent Workflows, a visual editor for designing multi-agent conversation flows. You can now route tasks between specialized subagents, each with its own logic, tools, and knowledge base... optimizing for cost, latency, and accuracy. It’s elegant, scalable… and dangerous for anyone selling “cheap automation.” The rise of agent orchestration changes the professional landscape: companies will soon deploy hundreds of specialized AI agents, not just to answer questions, but to replace entire process chains. That’s where the problem starts. If your agency or consulting offer is based on “we automate tasks with AI”… you’re already a commodity. Because automation alone doesn’t sell anymore → Structure does → Trust does → Narrative does The future belongs to those who turn automation into a luxury experience. Every leap in AI efficiency creates a new price war. When AI tools make execution cheap, value shifts back to perception and precision. That’s why premium positioning matters more than ever: → Build your offer architecture like a brand, not a service → Protect your margins through client experience, not feature lists → Sell certainty, not automation Automation is the engine. But luxury is the interface. The Agent Era won’t replace you. It’ll expose whether you’ve built a brand or just a service. P.S. If you’re serious about scaling your business with AI Production & Customized Pipeline Systems, start here → https://lnkd.in/eTxfbAVp

6

🤖 How a Snarky Cyborg Is Shaking Up Apple TV+'s Sci-Fi This week's 90sec Vault AI Rapid Insights dives into 'Murderbot' — the sci-fi-action-comedy everyone's talking about... **Analysis powered by ✨VaultGPT (DM me for more info) Click here for the full memo: https://lnkd.in/gzDB-Urj And click here to subscribe: https://lnkd.in/gtFZaXx9

8

1 Comment

Disney paid $1 billion to OpenAI to stop fighting an AI war they can't win. The deal has been hailed as a breakthrough for AI storytelling. I see it differently. This is just the inevitable monetization of AI's impact on IP - and luckily it seems to be happening sooner than in other IP piracy cycles. I've watched this pattern play out for decades. New tech emerges. IP gets used without permission. First it is ignored because it is too small. Then IP owners try to stop it, usually with lawsuits. They mostly fail. Then after it is long clear they lost, they cut a deal. Fighting rarely works. The technology and consumer adoption happens whether you participate or not. Disney isn't looking for a 20x return here. A billion dollars is roughly 0.2% of OpenAI's valuation. This is about securing a seat at the table. My guess? It's a round trip deal. Disney invests, OpenAI provides favorable licensing terms. Smart positioning, not a moonshot. IP owners need to accept a harder unfortunate truth: AI will use your intellectual property whether you want it to or not. I think of it this way: Disney couldn't stop us from drawing Mickey Mouse because we have pencils. (Mickey literally might be the only thing I can draw decently.) But the difference now is scale. AI enables instantaneous mass creation and distribution of derivative content. Legal action won't change that underlying capability - just like piracy it'll just keep popping up whack-a-mole style. I suspect ultimately the future looks a lot like what YouTube and Facebook already built for music and video IP on their platforms. Copyright protection systems that identify IP usage and let the owner decide: take it down, or take the monetization. The music labels learned this lesson. They do far better collecting revenue behind the scenes than fighting every infringement in court. AI will follow the same path. More IP will get licensed to these models until eventually the technology has built-in systems to identify pre-existing IP and route monetization to whoever owns it. This isn't the end of IP protection. It's the beginning of a new era for it. Companies that figure out how to monetize AI usage of their content will come out ahead. Those still fighting the last war will see their IP used anyway, with nothing to show for it. Just look at Viacom and their decade long lawsuit with Youtube. The question isn't whether AI will use your IP. It's whether you'll get paid when it does. (Witness the 10 seconds it took me to make the below image.)

66

44 Comments

Nvidia’s AI boom runs on a handful of customers. When two clients account for nearly 40% of your revenue, who’s really in control? One buyer made up 23% of Nvidia’s $46.7B Q2 revenue (~$10.7B). The second largest took 16% (~$7.5B). Together, that’s 39% of the company’s entire quarter. And it’s not new. The trend has been building for over a year and I posted about it before: 📊 Q2 FY2025 → Four customers = 46% of revenue 📊 Q1 FY2026 → Four customers = 54% of revenue 📊 Q2 FY2026 → Just two customers now = 39% of revenue When I wrote about Nvidia earlier this year, I flagged the same problem: hyperscalers like Microsoft, Amazon, and Meta drive the bulk of its growth. If even one of these giants shifts to produce their own chips, Nvidia’s revenue would take a dive, and with them, fair part of the S&P500. ⚠️ It’s not just theory, some customers are already building their own chips. Less powerful, more specialized, but the race is on: 🔹Amazon: Graviton CPUs + Trainium/Inferentia AI accelerators 🔹Google: Tensor Processing Units (TPUs) for cloud AI 🔹Microsoft: Maia 100 AI accelerator for Azure 🔹Meta: MTIA program for training and inference AI’s gold rush is concentrated in a handful of buyers pouring $200B+ into GPUs and concentration cuts both ways. 💬 How long can Nvidia ride this wave before customer dependency becomes its biggest weakness? And at what point do those customers stop being customers, turning into competitors building their own chips?

9

4 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

✅ *Today’s AI News* 🧠🌐 1️⃣ *OpenAI Launches GPT‑5.1 Preview* OpenAI rolls out the GPT‑5.1 public preview with improved reasoning, longer context windows, and stronger safety guards — aiming to reduce hallucinations and boost factual accuracy. 2️⃣ *Apple Unveils AI Health Assistant* Apple announces an AI health feature in iOS that uses on‑device models to monitor wellness trends, suggest habits, and generate personalized daily health insights while prioritizing privacy. 3️⃣ *EU Adopts New AI Transparency Rules* The EU finalizes regulations requiring major AI developers to publish model training summaries and risk assessments for high‑impact systems — a step toward global AI accountability standards. 4️⃣ *AI Helps Predict Earthquakes* A breakthrough research team uses deep learning to identify subtle seismic patterns, improving early detection models that could provide minutes of advance warning in quake‑prone regions. 5️⃣ *AI Content Detection Tools Expand* New AI tools from multiple vendors aim to distinguish AI‑generated vs human content more accurately — targeting misinformation, plagiarism, and academic integrity issues. 💬 *shared as received *** ***Follow and Connect for more updates*** If you are a founder and Investor you can join my Successful whatsapp group Community https://lnkd.in/gdfRcV9T #ai #artificialintelligence #founder #entrepreneur #entrepreneurship #startup #News #openai #Apple #chatgpt #llm #news

5

Big move from Samsung into the XR space — and a strong signal that XR is entering a new phase. A device at nearly half the price of Vision Pro, yet with similar specs, could shift the balance for developers, studios, and enterprise adopters alike. The real story here isn’t just another headset — it’s the growing Android XR ecosystem and how it might reshape collaboration, accessibility, and mixed reality design. #XR #Innovation #Samsung #Galaxy #MixedReality #MR #VR

5

3 Comments

Electronic Arts has agreed to a record $55 billion leveraged buyout by Saudi Arabia’s Public Investment Fund (through Savvy Games), Silver Lake, and Jared Kushner’s Affinity Partners. It marks a turning point for the videogame industry. Why it matters: 🎮 Gaming is at a crossroads. After the pandemic boom, the industry is in a downturn, with spending tightening. Publishers are now focused on how to extend the life — and profitability — of their biggest franchises. 📺 IP crossovers are the playbook. Hollywood and streaming platforms are driving value from game-to-screen adaptations: The Last of Us, Fallout, Arcane, Super Mario Bros, and soon Call of Duty. EA already has plans with Amazon MGM to turn The Sims into a movie. 💰 Saudi Arabia’s strategy. PIF has invested billions in gaming and entertainment, from Nintendo to Take-Two to Toei Animation, alongside moves into sports and cinema. Owning EA gives them Battlefield, Apex Legends, The Sims, Madden, and FIFA — franchises with both gaming and media potential. ⚠️ The risk. Expanding IP comes with massive production costs. Other firms, like Embracer, learned the hard way when acquisitions didn’t translate into successful content. As one analyst put it: “The value of high-end gaming IP is only increasing, but execution will be everything.” This deal highlights gaming’s role as a central pillar of global entertainment. The next great battle for audience attention will play out across consoles, streaming platforms, and movie theater screens. https://lnkd.in/dGhfGFC5

5

Steve Jobs at his finest especially with "You've got to start with the customer experience and work backwards to the technology."

8

Google I/O 2025: AR/AI Smart Glasses, not XR Headsets, were front and center at the Google I/O stage. This reinforces what I've been saying for a year now. The next frontier for Hardware is not AI Pins (StarTrek comm/badges), and it is not full on XR headsets. Open AI and other AI startups have no idea what they're doing (typical). The vast majority of humanity wants NOTHING like the Humane AI Pin (broken and sold for patents, the Teenage Engineering Rabbit (just an android app as a dev kit with a cut down, cheap Android phone) or the OpenAI/io/Jony Ive collaboration that tries to make Sam Altman into Steve Jobs (laughable). None of these will ever sell in numbers. Humane AI Pin? 10k units sold then sold for patents. Rabbit R1? 5k units in active use a year later. OpenAI's magical bullshit device? Aiming for 2027 to 2029. OpenAI is projecting they sell in 2029... JUST LIKE THE IPHONE in 2020. Yeah right. Meanwhile what does sell? AppleWatches. Wearables. Glasses. Apple sells tens of millions of AppleWatches every year. Apple is ensuring that future Apple Watches have a microphone & speakers, but also a camera. Like the Apple Smartglasses will have. Like Google Smartglasses will have. The next frontier that every hardware manufacturer is exploring is AI Smartglasses (no display) and AR Smartglasses (not True AR, monocular or binocular display). The issue is where is the user data privacy from datamining and AI training? Which BigTech do you trust with your biometric and real world data? Apple? Google? Meta? Samsung? Microsoft? Sony? LG? Huawei? Who? None? ---- UploadVR: "Google I/O 2025: Smart Glasses With HUD On-Stage Demo" (May 20, 2025) YouTube: https://lnkd.in/ga9u6jPW ---- TED: The Next Computer? Your Glasses | Shahram Izadi | Android XR (Apr 18, 2025) YouTube: https://lnkd.in/gbd7nxiF ---- UploadVR: "Google Is Working With Gentle Monster & Warby Parker On Gemini Smart Glasses" (David Heaney) (May 20, 2025) UploadVR: https://lnkd.in/gNtAxKA3 ---- The Verge: "We tried on Google’s prototype AI smart glasses. The eternal problem with smart glasses is you’ve got to see it to believe it." (Victoria Song) (May 20, 2025) The Verge: https://lnkd.in/gMu2SbJq --- The Verge: Xreal teases Project Aura smart glasses for Android XR Details are sparse, but this will be the second official device to launch on the Android XR platform." (Victoria Song) (May 20, 2025) The Verge: https://lnkd.in/gEe5trKS --- Bloomberg: "Apple Is Developing Specialized Chips for Glasses, New Macs and AI Servers" (Mark Gurman) (May 8, 2025) Bloomberg: https://lnkd.in/gJjjRUiU --- Bloomberg: "Apple Explores Push Into Smart Glasses With ‘Atlas’ User Study" (Nov 4, 2024) https://lnkd.in/gDZ2HAKy --- Forbes: "Samsung Smart Glasses May Launch This Year" (Mar 25, 2025) Forbes: https://lnkd.in/g3edr5e6 --- UploadVR: Apple Seems To Be Exploring Releasing Smart Glasses In 2027 David Heaney) (Nov 4, 2024) UploadVR: https://lnkd.in/gta6FPmn #xr #ai #ar #smartglasses #wearables

23

16 Comments