Jeremy Cowles

Amazing! New update on Luma AI "Keyframes have arrived within Dream Machine! Start and end keyframes give you narrative control, while text instructions guide the journey. Change lighting and seasons, control perspectives, morph subjects, and create impossible transitions!"

55

4 Comments

#Sony drops their #virtualproduction toolkit on the #unrealengine marketplace. Why is this a big deal? Previously, you only had access if you purchased the expensive kit of #venice / #venice2 cameras, or similar for the #arri. It wasn’t possible for other teams or departments to execute #previs in #preproduction, where this work needs to take place. In other words, your #vfx supervisor and vendors, didn’t have a means to understand how something was going to look, to do remote location scouting, setting up shots, and otherwise end up killing expensive production time on-set. Now, anyone can download the tools and handle the preparation that’s critical to this type of work. This is a major step towards #democratization and pushing #vp as THE mainstream studio methodology. Keep in mind that Sony has also invested over $1.5 Billion into Epic Games for technologies released to #streamingvideogames, #uefn, and related #metaverse tech. It’s fair to say we’ll see some rather interesting, new, creative tech developments in the quarters and years to come. If what I saw at #NAB / #NAB2024 is any indication it will be incredible. Way to go #Sony and #EpicGames! Download from the link below!

40

2 Comments

Eager to dive into the world of Large Language Models (LLMs)? Join Sebastian Raschka, PhD at SciPy 2024 for an in-depth tutorial on pretraining and fine-tuning LLMs from the ground up. Event Details: https://mng.bz/DpVR 🎉 🐍 Perfect for Python programmers eager to learn and code in PyTorch. Don’t miss this opportunity to learn from a leading expert! 📅 Mark your calendar: July 8th 🗺️ Discover more in his book: "Build a Large Language Model (From Scratch)" 📗🔗: https://mng.bz/lrp2 #SciPy2024 #LLMs #PyTorch #MachineLearning #AI #ManningBooks

372

16 Comments

⚡️Multi-Agent RAG Online Workshop⚡️ If there’s anything better than agentic RAG, it’s multi-agent RAG! In this event, we’ll explore the big idea behind “multi-agent” applications. These types of workflows combine multiple independent agents, which can be structured to work together to solve complex problems. The patterns of using multi-agent frameworks can take many forms including planning, reflection, and tool use. LangGraph, our library for building stateful, multi-actor applications with LLMs, can help you orchestrate your system. Sign up here ➡️ https://lu.ma/agentrag

236

3 Comments

We’re back with Bob Chesebrough, and he’s back to discuss dinosaurs! More specifically, he’s demonstrating how he created a clipped image with the background clutter removed using monocular depth estimation. This method of creating depth estimates from a single image using a Hugging Face model and Intel Developer Cloud. Read his blog to follow along. https://intel.ly/4aNcxiu #Developer #GenerativeAI #ArtificialIntelligence

141

Tutorial: Learn to build a project using Cesium for #Unity for the Magic Leap 2 headset. https://hubs.li/Q02yF3J60

58

Achieving real-time rendering of volumetrics is always a matter of ray-marching those volumes, and raymarching is an expensive technique. When you develop on Unreal Engine, you always have a choice to use something built-in or develop from scratch your own solution. In the first case, you can be sure you have the best integration and in-engine features, in the second case you can develop a custom render pass with the best performance and features you need. Being aware that UE already has Sparse Volumetric Textures for rendering VDBs we tested it and found that it’s not suitable for us for two main reasons: first of all, it is performance, and the second reason is that it is quite complex to use for non-power users. So we developed our own volumetric rendering component. Though it is still in progress, and we haven’t made any optimizations, our raymarcher is about 3 times faster than SVT and being implemented inside a translucent pass supports all needed features. As for features, we started with the most essential ones like self-shadowing, and received shadows illumination (performed in a separate compute shader). As with any other raymarcher, you can control step count and step size to control the quality/performance tradeoffs. The result?  Excellent visual fidelity and great performance — even combined decompression and raymarching for relatively small VDBs (up to 5GB) allows us to achieve a 2x performance uplift compared to SVT. Also when we use VDBs bigger than 5GB SVT often glitches and stutters making it barely usable when we run them on RTX 3080 Ti 12GB. Now we can run a 40Gb effect in UE5 in just 16ms (combined decompression and raymarching) which was impossible before! #Zibra #ZibraVDB #OpenVDB #Compression #VFX #unrealengine #tools #vdb #rendering #VirtualProduction #VirtualProductionSet

300

16 Comments

Splats are go in UE5.4, check that lighting...

1

How cool is websim.ai??? 😅🕶️🎉 By using Anthropics Claude 3 models (choose between Opus, Sonnet and Haiku) websim.ai generates a website for you by just providing a URL and optional parameters. I tried https://lnkd.in/g54nQrTN and it created a website with a searchable database of movies shot on celluloid 🤯 I mean how cool is that! And I like the fact that a lot of the generated websites look a lot like they are straight from the first days of the world wide web 😅 #websimai #ai #websites #llm #anthropic #claude3

8

2 Comments

Now, the most interesting part, is that Luma made a plug-in available for Unreal Engine 5.3, here: https://lnkd.in/gnkCib-2, that is able to pull Luma app output files into Unreal scene. It's still more like development version, but it works and results are just cool. All you need is Luma AI app on your phone, shoot some objects, process them, pull into UE, fly around. Realtime render quality is not the best, so it might be a good idea to render everything out using sequencer. Here's just a brief example of what could be done:

2

Has anyone used Meta Quest 3 or other AR/VR/mixed reality headsets? I'm looking for something that could be useful as both a capture device (sensors) and for AR/VR/MR/Vision related projects. If you have recommendations, please leave a comment. #AR #VR #ComputerVision #mixedreality

2

2 Comments

🧬🧬 Event-driven SuperResolution 🧬🧬 👉USTC unveils EvTexture, the first VSR method that utilizes event signals for texture enhancement. It leverages high-frequency details of events to better recover texture regions in VSR. Code available💙 𝐇𝐢𝐠𝐡𝐥𝐢𝐠𝐡𝐭𝐬: ✅First event-driven texture restoration in VSR ✅Recovering hi-freq textural info progressively ✅Texture enhancement branch coupled w/ ITE ✅Novel SOTA on 4 challenging VSR benchmarks #artificialintelligence #machinelearning #ml #AI #deeplearning #computervision #AIwithPapers #metaverse 👉Discussion https://lnkd.in/dMgakzWm 👉Paper arxiv.org/pdf/2406.13457 👉Code https://lnkd.in/d-dpztfY

350

14 Comments

Ready to deploy AI faster and safer? Check out how #NVIDIA NeMo Guardrails and #LangChain Templates can help you build safer #LLMs. Read more > https://nvda.ws/4aTZYkM

4