Which one generates a better 3D model from video? Luma AI or 3DPresso?

Olli Huttunen
22 Jul 202316:41

TLDRIn this video, 'oliotron' compares Luma AI and 3DPresso, two AI-assisted applications for creating 3D models from video. Utilizing 360-degree cameras for scanning, the video explores the process from recording to editing and the final 3D model output. Luma AI offers a fast, free service with impressive environment capture but less accurate models. 3DPresso, while slower, provides more accurate and usable models, especially for larger objects. The verdict? 3DPresso is superior for detailed models, while Luma excels in capturing environments.

Takeaways

  • ๐Ÿค– Luma AI and 3DPresso are two AI applications that create 3D models from video images using neural Radiance Fields (NeRF) technology.
  • ๐ŸŽฅ The video demonstrates the process of using a 360 camera for 3D scanning, including recording and editing techniques for best results.
  • ๐Ÿ” Luma AI and 3DPresso have different strengths; Luma AI is better for capturing environments and larger subjects, while 3DPresso is more suited for creating accurate, usable 3D models.
  • ๐Ÿš€ Luma AI is currently free to use and offers features like automatic NeRF rendered animations, but it has limitations on rendering camera animations.
  • ๐Ÿ’ฐ 3DPresso uses a credit-based billing system, with new users receiving free credits, and is designed to produce clear polygon-based mesh models.
  • ๐Ÿž๏ธ Luma AI excels at capturing the background and surroundings of the scanned object, but the accuracy of the 3D model can vary.
  • ๐Ÿ“ 3DPresso's models tend to have a cleaner 3D mesh and better textures, making them more suitable for 3D printing or use in games.
  • ๐Ÿ”ง Both applications have limitations and may not produce perfect models every time, especially with challenging objects like those with transparent or narrow features.
  • ๐Ÿ”„ The process involves uploading and processing video files through the respective AI platforms, with Luma AI being faster in processing times.
  • ๐Ÿ“ˆ The technology behind these applications is rapidly developing, and their potential uses in the future are promising but still uncertain.
  • ๐Ÿ‘ The video encourages viewers to experiment with these services themselves, using readily available tools like smartphones for video capture.

Q & A

  • What are the two AI-assisted applications mentioned for creating 3D models from video images?

    -The two AI-assisted applications mentioned are Luma AI, developed by Luma Labs, and 3DPresso, a Korean alternative.

  • What technology do Luma AI and 3DPresso use to create 3D models from video?

    -Both Luma AI and 3DPresso use the neural Radiance field, also known as NeRF technology, to create 3D models from video.

  • What type of camera does the video recommend for 3D scanning and why?

    -The video recommends a 360-degree camera for 3D scanning because it can capture larger areas at once, making it suitable for subjects of your own size or larger.

  • How does the video suggest recording video for 3D scanning with a 360 camera?

    -The video suggests recording in three rounds at different heights, utilizing the 180-degree area that the front lens can see, and ensuring the person recording stays behind the camera to avoid appearing in the final image.

  • What is the recommended process for editing 360 videos for 3D scanning?

    -The recommended process involves uploading the material to a computer, cropping the video into a 16x9 format, adding keyframes to keep the object centered, and exporting the video as an mp4 file in 1920x1080 resolution with a high bitrate.

  • How does Luma AI handle the processing of 3D models compared to 3DPresso?

    -Luma AI is currently faster in processing 3D models than 3DPresso, and it automatically creates a NeRF rendered animation for each model.

  • What is the difference in pricing between Luma AI and 3DPresso?

    -Luma AI is free to use, while 3DPresso uses a credit-based billing structure with new users receiving a hundred beans for free to start using the service.

  • What are the limitations of Luma AI when it comes to rendering camera animations?

    -Luma AI limits the renders to 30 times per month, but it remains free to use.

  • How does the video compare the accuracy of the 3D models produced by Luma AI and 3DPresso?

    -The video finds that 3DPresso produces more accurate models with better textures, while Luma AI's models tend to be less accurate with lumpy surfaces and holes.

  • What is the final verdict on which application is better for creating usable 3D models from video?

    -The final verdict is that 3DPresso is more useful for creating accurate and complete models for 3D printing or game assets, while Luma AI is better for capturing environments and larger entities.

  • What does the video suggest about the future of these AI applications in 3D modeling?

    -The video suggests that as the technology is developing rapidly, it will be interesting to see where the true potential of NeRF models and applications like Luma AI and 3DPresso will take us in the future.

Outlines

00:00

๐Ÿค– AI-Powered 3D Modeling from Video

The script introduces the challenge of creating 3D models from video images using AI-assisted applications. It mentions two such applications, Luma AI by Luma Labs and 3D Presser from Korea, both utilizing neural radiance fields (NeRF) technology. The narrator, Oliutron, plans to compare these services by using a 360-degree camera to record videos of three different objects. The video explains the principles of using a 360-degree camera for 3D scanning, emphasizing the camera's advantages for larger subjects and detailing the process of recording and the importance of the recording angles.

05:01

๐ŸŽฅ Post-Processing for 3D Scanning

This paragraph describes the process of editing 360-degree videos for 3D scanning using Insta360 Studio application. The narrator demonstrates how to crop the video into a 16x9 format, add keyframes to keep the object centered, and export the videos in mp4 format with a high bitrate for quality. The focus is on preparing the video material for processing by AI programs like Luma AI and 3D Presser. The paragraph also discusses the time taken for these programs to process the 3D models and their respective pricing models, highlighting Luma AI's free usage and 3D Presser's credit-based billing system.

10:02

๐Ÿบ Comparing 3D Models from AI Services

The narrator compares the 3D models produced by Luma AI and 3D Presser after processing the same video files. Luma AI is noted for its speed and the inclusion of a pre-calculated NeRF rendered animation, which is visually impressive but not representative of the final model's accuracy. 3D Presser's model is found to be more accurate with better textures, especially noticeable in the statue model comparison. The paragraph also discusses the examination of the models within Blender, revealing that 3D Presser's model is more intact and has better topology, making it more suitable for 3D printing or use in games.

15:04

๐Ÿš› Evaluating AI 3D Scanning Applications

The final paragraph summarizes the narrator's findings from testing Luma AI and 3D Presser on different objects. While Luma AI excels in capturing environments and larger entities, creating NeRF models suitable for Unreal Engine, 3D Presser is praised for producing more usable polygon-based mesh models from videos. The narrator acknowledges the developmental stage of these technologies and the inconsistency in model quality, as seen with the trailer model. The paragraph concludes with a recommendation to test these services with various subjects and a reminder of the potential future applications of NeRF models, inviting viewers to explore and subscribe for more content.

Mindmap

Keywords

๐Ÿ’กAI-assisted application

An AI-assisted application refers to software that utilizes artificial intelligence to perform tasks more efficiently, such as generating 3D models from video images. In the context of the video, Luma AI and 3DPresso are examples of such applications, which use neural Radiance fields (NeRF) technology to create 3D models from video footage.

๐Ÿ’กNeural Radiance Fields (NeRF)

Neural Radiance Fields, or NeRF, is a cutting-edge technology used in computer graphics to generate 3D scenes from a collection of 2D images. It is a deep learning-based approach that represents a scene as a continuous volumetric field. In the video, both Luma AI and 3DPresso employ NeRF to convert video into 3D models.

๐Ÿ’ก360 camera

A 360 camera is a device capable of recording panoramic images or videos, capturing a full 360-degree field of view. In the video, the 360 camera is used for 3D scanning by focusing on the 180-degree area visible to the front lens, allowing the creator to capture videos of objects from all angles.

๐Ÿ’ก3D scanning

3D scanning is the process of analyzing a real-world object to collect data on its shape and possibly its appearance (color, texture, etc.). The video demonstrates using a 360 camera for 3D scanning by recording videos of objects and then processing these videos to create 3D models.

๐Ÿ’กInsta360 RS1 camera

The Insta360 RS1 is a specific model of a 360 camera mentioned in the video, known for its high-resolution capabilities, recording up to 6K video. It is used by the creator to capture high-quality footage for 3D scanning purposes.

๐Ÿ’กHorizon lock

Horizon lock is a feature of some 360 cameras that keeps the camera level even when the user moves it around. In the video, the creator uses this feature to ensure that the camera remains leveled during the scanning process, preventing the appearance of the user in the final 3D model.

๐Ÿ’กInsta360 Studio

Insta360 Studio is the application used to edit and process the 360 videos captured by the Insta360 camera. In the video, the creator uses this software to crop and rotate the video footage to prepare it for the AI-assisted 3D model generation process.

๐Ÿ’กLuma AI

Luma AI is one of the AI-assisted applications discussed in the video, developed by Luma Labs. It uses NeRF technology to generate 3D models from video images. The video compares Luma AI's performance with 3DPresso in creating 3D models from the same video footage.

๐Ÿ’ก3DPresso

3DPresso is another AI-assisted application mentioned in the video, which also uses NeRF technology to generate 3D models from video. It is compared with Luma AI to determine which application produces better 3D models.

๐Ÿ’กGLB format

GLB is a file format for 3D models that is widely used in web applications and supports embedded textures and materials. In the video, the creator mentions downloading 3D models in GLB format for further examination in other 3D software like Blender.

๐Ÿ’กBlender

Blender is a free and open-source 3D creation suite used for modeling, rigging, animation, simulation, rendering, compositing, and motion tracking. In the video, the creator uses Blender to examine and compare the 3D models generated by Luma AI and 3DPresso.

Highlights

Luma AI and 3DPresso are two AI-assisted applications that generate 3D models from video images.

Luma AI is developed by Silicon Valley startup Luma Labs, while 3DPresso is a Korean alternative.

Both applications use neural Radiance fields, also known as NeRF technology, for 3D modeling.

Olioutron compares Luma AI and 3DPresso by testing them on video files of three different objects.

A 360 camera is used for recording the objects, focusing on the 180-degree area visible by the front lens.

The camera's white fisheye lenses are advantageous for scanning larger subjects.

Luma AI processes 3D models faster than 3DPresso, with a 30-minute turnaround time.

3DPresso introduces a credit-based billing structure with free beans for new users.

Luma AI has limitations on rendering camera animations, allowing only 30 renders per month.

Luma AI automatically creates a NeRF rendered animation of the object, showcasing its capabilities.

3DPresso produces more accurate and cleaner 3D models with better textures compared to Luma AI.

In Blender, 3DPresso's model is more intact and less fragmented than Luma AI's.

Luma AI excels at capturing the environment around the object, but struggles with narrow forms.

3DPresso struggles with larger objects, failing to produce a usable model of the trailer.

Luma AI is better suited for creating real NeRF models and is compatible with volume model programs like Unreal Engine.

3DPresso is designed for creating complete and clear polygon-based mesh models.

Both applications are in early development and do not always produce working models.

The potential of NeRF models and their future applications in technology are promising.

Olioutron encourages viewers to test these services themselves using smartphones for video recording.