John LePore

🚨 This is what 450+ shots in Luma AI's Dream Machine looks like 🚨 I maxed out the second highest paid tier on Dream Machine to really test the boundaries of what the model can do. I was really impressed with its spatial coherency, dynamic camera shots, vehicles traveling correctly among a trajectory - and the fast Drone shots... Man the drone shots...(have a look and you'll know what I mean!) 🚨 So enjoy this compilation of ethereal AI dreams brought to life with Dream Machine, LensGO AI, krea.ai & Midjourney. Watch the uncompressed 4K original here: https://lnkd.in/d-CUNW_V Dreamed, Not Designed. A Feeling, Not A Story. ✅ All Images done with Midjourney. ✅ All Images Creatively Upscaled with Leonardo AI Universal Upscaler https://t.ly/0n8kW ✅ Videos Generated Luma Lab's Dream Machine https://lnkd.in/d-bCWTkB ✅ Style Transfers done with Stable Diffusion (Skull melting shots) ✅ Space transition shot done with krea.ai video. ✅ All Videos Upscaled with Topaz Labs video AI: https://lnkd.in/gVZWh8mh ✅ Video Editing in After Effects with Mr Horse Product Plugin: https://lnkd.in/duQpVMSZ ✅ And AE Juice Plugin: https://lnkd.in/dhxM484x ✅ Additional VFX made with Blind Dusk. Music Track: "Watch The World Burn" by Falling In Reverse. www.billyboman.com

236

115 Comments

Ok Luma AI 👏 👏 I didn’t think anything would come close to Runways capabilities but I was proven wrong - happily. 🧪 I’ve been testing all the new video AI tools that’s been popping up bts and not really been thrilled because a majority of them seem to have a plasticky look/ sharpened filter/ lose stability pretty easily (so does Luma and Runway but they hold out longer) 🧠 What also needs to be kept in mind is that none of these tools are mind readers- you need to be super specific with what you’re looking for / source image needs to be spot on. Else you’re going to let the AI’s intrusive thoughts take over (which it always does in the videos- it’s just edited out - will be posting the intrusive thoughts results in the comments) #genai #ai #productdesign #artificialintelligence #creativedirection

14

A new #Design #future is here and moving fast. Real time material scanner instantly places pattern and texture on any 3d model. I can only imagine how helpful this will be in game design, fashion and the auto industry. How could you see this #tech being applied to your industry?

1

2 Comments

💎 12 helpful Luma AI Dream Machine tips to boost your generation workflows. Hot of the press. Tech Specs: + Rendering duration: 4 minutes + Number of Generations: 1 video per prompt (was 2 at the start) + Video duration: 5 sec + Resolution: 1360x752 + Frame Rate: 24 fps VIPs + Dream Machine does not know any VIPs + You have to provide an initial image to get the desired results Image 2 Video + Image to video can yield impressive results with coherent backgrounds and detailed characters. Extending Shorts + Use the final frame trick to extend shots + save the last frame of your clip + use it as the starting image for the next clip + don't forget to apply a different prompt + this will help to create longer scenes but decoherence might still occur Prompting Actions + Specific actions can sometimes result in weird outputs + e.g., morphing hands or arms + You will need two prompts + #1 for your image generation + #2 for your video generation + Save your video generation prompt somewhere else + While the video is being generated, you cannot access it any more Mood + Prompts with atmospheric descriptions will impact the mood and tone of the video + Try out foggy sky, dark clouds, moody light, warm scene Comparing + Comparing multiple generations from the same prompt + This helps to understand the range and limitations + Also compare generations with and without 'Enhance Prompt' + This helps to determine if the enhancer really helps you. Sadly we cannot set a seed for now Camera Control + Dolly Shot + Tracking Shot (or Trucking Shot) + Crane Shot + Steadicam Shot + Handheld Shot + Pan Shot + Tilt Shot + Zoom Tip use GPT4 to enhance this list and provide descriptions. Overall I encourage you to use images for a starting reference as this provides you with a lot more control. Prompt: cinematic slow camera dolly shot towards will smith and taylor swift, sitting at a table in a fancy restaurant. They are talking to each other and seem happy. cinematic light I'm currently creating my first video ad but generation times are still sooo slow. Stuck on a 2 minute timer for about 1 hour ;) ↓ ☝️ Follow for more AI-News Jacques Alomo 👋 Produce content on our Gen AI platform creamAI: http://www.creamai.de #ai #luma #dreammachine #aivideo #videogeneration #sora #kling

140

22 Comments

Thinking of archiving my immersive enviros as Gaussian Splat vignettes, and trying different output solutions. So far the Spline editor has it all in terms of intuitive ease and numerous options for both building and exporting/embedding. https://lnkd.in/grXGWhQS

4

Holoportation - The Next Big One in Communications. Holoportation is a new type of 3D capture technology that allows high quality 3D models of people to be reconstructed, compressed and transmitted anywhere in the world in real-time. When combined with mixed reality displays such as HoloLens, this technology allows users to see and interact with remote participants in 3D as if they are actually present in their physical space. Communicating and interacting with remote users becomes as natural as face-to-face communications.

1

Making sure your design work performs well on the platforms you intend to is something designers don’t account for anymore because modern workflows allow for up to 8k outputs. With new technologies that require higher frame rates on two high resolution screens and in realtime, following devices requires is again something modern designers have to account for. Here’s a video where we discuss this.

1

WARNING! Scary stuff 😱 I used Luma AI's Dream Machine to create my own 1970's horror movie scene. Made with a single Midjourney image plus 2 extends. Initial prompt: 𝘚𝘭𝘰𝘸 𝘵𝘳𝘢𝘤𝘬𝘪𝘯𝘨 𝘪𝘯 𝘴𝘩𝘰𝘵 𝘰𝘯 𝘢 𝘨𝘪𝘳𝘭 𝘸𝘢𝘭𝘬𝘪𝘯𝘨 𝘪𝘯𝘵𝘰 𝘢 𝘤𝘳𝘦𝘦𝘱𝘺, 𝘴𝘸𝘪𝘳𝘭𝘪𝘯𝘨, 𝘨𝘩𝘰𝘴𝘵𝘭𝘺 𝘮𝘪𝘴𝘵 𝘵𝘩𝘢𝘵 𝘤𝘢𝘳𝘳𝘪𝘦𝘴 𝘩𝘦𝘳 𝘢𝘸𝘢𝘺 Dream Machine ignored my prompt to track in and didn't show the girl being carried off, but it had her walking really nicely so I went with it. Extend prompt 1: 𝘌𝘯𝘨𝘶𝘭𝘧𝘦𝘥 𝘪𝘯 𝘨𝘩𝘰𝘴𝘵𝘭𝘺 𝘮𝘪𝘴𝘵 𝘢𝘯𝘥 𝘴𝘸𝘢𝘭𝘭𝘰𝘸𝘦𝘥 𝘶𝘱 𝘣𝘺 𝘢 𝘱𝘰𝘭𝘵𝘦𝘳𝘨𝘦𝘪𝘴𝘵. 𝘛𝘦𝘳𝘳𝘪𝘧𝘺𝘪𝘯𝘨! Dream Machine did the fast tracking in part here, despite there being no camera move in the prompt. I may have had 'enhance prompt' switched on. I tend to switch between it being on and off and often generate one version with it on and another with it off. You tend to get more unexpected camera moves with it on. Extent prompt 2: 𝘈 𝘨𝘪𝘳𝘭 𝘲𝘶𝘪𝘤𝘬𝘭𝘺 𝘵𝘶𝘳𝘯𝘴 𝘩𝘦𝘳 𝘩𝘦𝘢𝘥 𝘵𝘰 𝘧𝘢𝘤𝘦 𝘶𝘴 𝘢𝘯𝘥 𝘳𝘦𝘷𝘦𝘢𝘭𝘴 𝘴𝘩𝘦 𝘪𝘴 𝘢 𝘩𝘪𝘥𝘦𝘰𝘶𝘴 𝘮𝘰𝘯𝘴𝘵𝘦𝘳. 𝘎𝘳𝘰𝘵𝘦𝘴𝘲𝘶𝘦, 1970𝘴 𝘩𝘰𝘳𝘳𝘰𝘳. 𝘛𝘩𝘦 𝘮𝘰𝘯𝘴𝘵𝘦𝘳 𝘴𝘤𝘳𝘦𝘢𝘮𝘴 𝘥𝘪𝘳𝘦𝘤𝘵𝘭𝘺 𝘵𝘰 𝘤𝘢𝘮𝘦𝘳𝘢 𝘪𝘯 𝘦𝘹𝘵𝘳𝘦𝘮𝘦 𝘤𝘭𝘰𝘴𝘦-𝘶𝘱 This prompt was followed to the letter. I tried around 2 or 3 re-rolls for each prompt and chose the best. The clip you see is exactly as it came from Dream Machine with no editing or colour grading. The music came from Artlist and effects from Motion Array. Only the footsteps were AI generated using ElevenLabs. It's scary how good AI video is getting! ---------------------------------------------- If you'd like to learn how to enhance your own creativity using the very latest AI tools then check out my live, online course. I've just opened the August cohort and places are limited - so grab your spot while you can! 🏃‍♀️🏃🏾‍♂️⏰ https://lnkd.in/eDj8Vm4N #lumadreammachine #midjourney #aitraining

110

34 Comments

Luma AI’s new tool Dream Machine has been all over the internet for the past few days, so I figured I’d share what I was able to create. I just wanted to see what it is capable of so I ran some Midjourney images through and added some text prompts with the image. It works best with image + text prompts. The full length of the videos is about 5 seconds, which is a little longer than Runway. And it’s better than Pika IMO. Looking forward to playing around with this some more! All images created in Midjourney Animated with Dream Machine Music created with Suno AI Thoughts on this? Have you used Dream Machine yet?

7

Sound design requires layering. The !Habits checkmark sound is 10 sounds layered together each contributing a different component to the overall sound—build, punctuation, finish. https://lnkd.in/gyDqrt7t

189

12 Comments

🏮 Short-form Narrative Pipelines for Social Platforms, Part 3 🏮 To help us streamline our workflow, we worked with Chaos, Epic Games, NVIDIA, and Lenovo to push the real time rendering envelope by fully ray tracing live using Unreal Engine. For those unfamiliar with the nuances of lighting and rendering, here's a high level breakdown along with an explanation as to why FULLY ray tracing is a big deal: Games use a rendering technique called rasterization to light environments. Basically, individual pixels in a 3D mesh are shaded based on how lights in the environment hit each pixel. While computationally intensive, game engines are able to handle the workload in real time. That's why when you walk down a street in a video game it feels like you're in a 3D world, even though you're viewing that world on a 2D screen -- the horizon/environment is being rendered in real time as you walk, so it feels like you're in it. Movies and TV shows on the other hand use ray tracing to light environments. Working backwards from the POV of the viewer, ray tracing is a physics-based rendering technique that follows (traces) light paths (photons) as they bounce off objects. This technique is more computationally intensive than rasterized rendering and is done offline on servers after a scene has been filmed. Harnessing the power of modern GPUs, ray tracing produces images that are indistinguishable from reality, and that's why films look more "real" than games. The holy grail for games is to achieve film quality rendered environments, while the holy grail for 3D films is to be able to see a final frame image in real time. Certain game engines, like Unreal Engine, come equipped with a ray traced effect added on top of the rasterized image. This gives the game an incredible look right out of the box, but to get to that film quality bar, you need to go that extra little mile. You need to FULLY ray trace. That means film teams will film once in engine and then re-render everything they just filmed offline to reach that film quality bar. That means more time and more $$. Working with the team at Chaos, they live linked Chaos Vantage directly into Unreal Engine. This gave us two screens when filming. One that showed the scene in Unreal and one that showed the same scene in Vantage. Both were capturing in real time. The scene in Vantage was fully ray traced live at about 30fps (see example below). This allowed us to get really close to a final frame image in real time, which we presented at NVIDIA GTC22. Why is this a big deal? Because the director can CHOOSE what shot to finish rendering offline instead of re-rendering everything. It's a much more agile workflow that reduces the time between concept and market. If you want to learn more about what an interoperable 3D asset future looks like, check out my 'Interoperable Playbook' on my profile page or download it for free here --> https://lnkd.in/gGHDMZ4p Cheers! 🖖

1

Here’s the rendered video of my latest art project R&D test where #motioncapture data drives fluid #simulations and emissive #shaders . Made with INSYDIUM Fused #xparticles vertex maps and #cycles4d

8

Two, REALLY big advancements in the bridging of AR/VR to possible ways of "Virtual Touch 3D Scanning". Sensoryx and Gravity Sketch are already showcasing ways to use either the hand controllers with a VR headset or a "Pen". Imagine tracing a physical car with a pen and that going directly into 3D!!! Below are a few links show this very concept is action and this is just the beginning!!! - Tracing a Motorcycle - https://shorturl.at/kmnQ2 - Details around the Pen's capabilities - https://lnkd.in/ewY3xgKu - Using Gravity Sketch and then import into Alias Automotive!! - https://shorturl.at/wIJQ8

16

How I made a 20-second cinematic trailer with Luma AI, Audio Design Desk, and ElevenLabs 🐿 "What year is this? What are neural frames? And what's this squirrel doing in my room?" 🕵‍♂️ Clues below: 1. Wrote an image prompt for MidJourney "A squirrel holding a video camera. The camera has a digital control panel and a MIDI controller synthesizer keyboard embedded on its side. 3D isometric pop art art, futuristic, cyberpunk, purple and tangerine hues --ar 2:1" 2. Luma AI: Picked a Midjourney image and imported to Dream Machine Luma's video output is ~3-4 seconds length per generation. To get a long, continuous shot, I needed to work in short incremental sprints. I extended each clip 3+ times, picking the best option and repeating the process until I had a branch that captured the story I wanted to tell. Luma needed a few minutes to render each video clip. It takes around 10-30 minutes to extend a batch of 3-5 clips on a good day. 3. Audio Design Desk for music, sfx and sound design layers ADD is a digital audio workstation specializing in sound for video. I used it to find and place music, sound effects, risers, and impacts for this video. The DAW comes with 70,000+ royalty-free sounds, plus integrations to all the major libraries like Artlist, Epidemic Sound, Soundstripe, and Envato. So it took me all of 20 seconds to find a jazz noire cue that I liked. Then I started pulling in sound effects - mechanical camera clicks and shuttering drawers, plucked piano keys from the melody library. I needed some riser and impact sounds for the outro. ADD has thousands of sound design cues, already pre-mapped to special QWERTY keys on my laptop. When I hit "r" it triggered a riser. I pressed the "i" key and triggered impact sounds. Each clip automatically showed up on the audio timeline for the video. Next, in an effort to find better sounds, I hit the replace key. It swapped in new risers and new impact hits in real time, during video playback. Each audio file had an embedded sync point, so new files that I swapped in stayed in sync with the video. I quickly found the sounds I liked and exported the final mix to my desktop. 4. Wrote a short script and pasted it into ElevenLabs Auditioned a dozen male and female voices until I found one with the tone and cadence I liked. Then I downloaded the AI narrative voiceover to my laptop. 5. Imported video and audio files to iMovie Lined up the squirrel animation, ADD audio and voiceover. Then slapped on a couple of iMovie titles and transitions for effect. 💡 I've had fun testing out these tools. The prompting, selection and media collage process is an art form. There are also as many software stacks and workflows to be discovered as you could possibly wish for. The squirrel mystery concept was inspired by John Zorn's Spillane, Lynch's Black Lodge from Twin Peaks, and the Cool 3D World animations. More demos coming this week...

38

17 Comments

How about using Luma AI Dream Machine-Keyframes to create animations for mockups? I made some experiments with iPhone mockups, and it totally works! You can animate the screen, the background, or both! All you need to do is choose start and end keyframes, add a prompt, and ta-da! For this one, my prompt was: screen transforms smoothly.

132

8 Comments

Here is a quick case study on the power of Mixed Reality in one short video!

9

3 Comments

🚀🍩 My new YouTube tutorial about mastering 3D Lattice Design using Blender Geometry Nodes is online! In this first start of my Geometry Nodes series, I dive into the art of using custom lattice cells, controlling every aspect of your lattice, and exporting watertight voxel meshes for your #3Dprinting projects. Watch till the end to see how smoothly you can adapt your base geometry while automatically applying the 3D lattice. I was surprised by how swiftly #Blender was able to render the lattice in real-time. Even the exporting to mesh was surprisingly fast and felt more stable than Grasshopper! For a free and open source software I think this is quite competitive. Check out the extended tutorial on my YouTube channel and follow (link in comments). Stay tuned for the next part! #computationalDesign #AM #DfAM #3Dlattice

154

17 Comments

Customizing every detail and texture within an asset doesn't have to be a pain. Sortium's #generativeai allows you to seamlessly use prompts to customize multiple textures in seconds. Ready to try it for yourself?

24