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

In this video, the presenter, olioutron, compares two AI-assisted applications, Luma AI and 3DPresso, which generate 3D models from video images using neural Radiance field (NeRF) technology. The video demonstrates the process of 3D scanning using a 360 camera and the subsequent editing in Insta360 Studio. After uploading the video to both Luma AI and 3DPresso, the results show that Luma AI is faster but 3DPresso produces more accurate and complete models, especially suitable for 3D printing or game development. However, Luma AI excels at capturing environments and larger entities, making it ideal for creating immersive experiences. The presenter concludes that while both applications have their strengths, 3DPresso is currently better for creating usable 3D models from videos, but acknowledges the rapidly evolving nature of this technology.

Takeaways

• 📚 Luma AI and 3DPresso are two AI-assisted applications that can generate 3D models from video images.
• 🚀 Both utilize neural Radiance fields (NeRF) technology, which is a trendy approach in the field of AI and 3D modeling.
• 🎥 The process involves recording video of an object using a 360-degree camera, which captures a wide field of view beneficial for larger subjects.
• ⏱ Luma AI is faster in processing 3D models compared to 3DPresso, with both handling a single video within 30 minutes.
• 💵 3DPresso operates on a credit-based billing structure, while Luma AI remains free to use with some limitations on rendering.
• 🤔 Luma AI provides a pre-calculated NeRF animation, which is visually impressive but may differ from the final 3D model.
• 📈 3DPresso's models are more accurate with better textures, making them more suitable for 3D printing and game development.
• 🧩 Luma AI's models, when translated into polygon models, can be less accurate with lumpy surfaces and holes.
• 🌐 Both services have their strengths; Luma AI is better for capturing environments and larger entities, while 3DPresso is more precise for detailed models.
• 🚫 Neither application successfully created usable models for objects with very narrow features, indicating the limitations of current technology.
• 🔍 The technology is rapidly developing, and the true potential of NeRF models in various applications is yet to be fully realized.
• 📝 The comparison encourages users to test these services themselves, as they can be used with various recording devices, not just 360-degree cameras.

Q & A

Which two AI-assisted applications are mentioned in the transcript for generating 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 both utilize to create 3D models from videos?

Both Luma AI and 3DPresso utilize a neural Radiance field, also known as NeRF technology, to create 3D models from video images.

What is the advantage of using a 360 camera for 3D scanning as described in the transcript?

The advantage of using a 360 camera for 3D scanning is that it can capture larger areas at once, making it particularly suited for scanning objects of human size or larger. It is also useful for capturing the surroundings of the object being scanned.

What are the key steps involved in preparing a 360 video for processing by AI 3D modeling software?

The key steps include recording the video using a 360 camera, uploading the material to a computer, editing the video in a program like Insta360 Studio to crop it to a 16x9 format, adding keyframes to keep the object centered, and exporting the video as an mp4 file at a high bitrate.

How does Luma AI differ from 3DPresso in terms of processing speed and cost?

Luma AI is currently faster in processing 3D models compared to 3DPresso. Luma AI is free to use, while 3DPresso has a credit-based billing structure with new users receiving a hundred beans for free to start with.

What is the difference between the Nerf rendered animation and the 3D model view in Luma AI?

The Nerf rendered animation in Luma AI is a pre-calculated rendering that provides an impressive visual effect. However, the actual 3D model view shows a different and less accurate representation of the object. When the Nerf model is translated into a regular polygon model, it may have lumpy surfaces and holes.

How does 3DPresso's model of the statue compare to Luma AI's model in terms of accuracy and texture quality?

3DPresso's model of the statue is more accurate and has better texture quality than Luma AI's model. It manages to create cleaner 3D mesh and better-looking textures from the same source video material.

What are the challenges faced when scanning objects with transparent surfaces using photogrammetry modeling techniques?

Objects with transparent surfaces, like the lantern part of the lamp post, are challenging to implement in photogrammetry modeling techniques because they often cause problems in the creation of accurate 3D models.

What is the most useful modern 3D format for keeping textures inside the 3D file?

The most useful modern 3D format for keeping textures inside the 3D file is GLB (.glb), which is used by both Luma AI and 3DPresso for exporting their models.

What is the conclusion of the comparison between Luma AI and 3DPresso in terms of creating usable 3D models from video?

The conclusion is that 3DPresso is more useful for creating accurate and complete polygon-based mesh models suitable for 3D printing or use as 3D assets in games. Luma AI, on the other hand, is better for capturing environments and larger entities, and for creating real NeRF models that are best utilized with programs that understand volume models, such as Unreal Engine.

What is the narrator's suggestion for those interested in testing these AI 3D modeling services?

The narrator suggests that interested individuals should test these services themselves, noting that one can easily produce videos with a smartphone for 3D scanning purposes.

What is the narrator's final thought on the potential of NeRF technology and AI applications like Luma AI and 3DPresso?

The narrator expresses excitement about the fast pace of development in this technology and looks forward to seeing where applications like Luma AI and 3DPresso will take us in the future.

Outlines

00:00

🚀 Introduction to AI-assisted 3D Modeling from Video

The video introduces the topic of AI-assisted applications that can create 3D models from video images, specifically mentioning Luma AI and 3D Presser, both utilizing neural radiance fields (NeRF) technology. The host, Oliutron, explains his intention to compare these two services using video files of different objects captured with a 360-degree camera. The process of 3D scanning with a 360 camera is outlined, emphasizing the camera's advantages for larger subjects and the importance of recording from various angles to ensure a complete 3D model.

05:01

🎥 Post-Processing and AI Model Generation

The second paragraph details the process of uploading and editing the 360 video material using Insta360 Studio. The host describes cropping the video, adding keyframes, and exporting the final video for AI processing. The video files are then uploaded to Luma AI and 3D Presser to generate 3D models. The paragraph also discusses the processing times and billing structures of both services, highlighting Luma AI's free usage and 3D Presser's credit-based system.

10:02

📐 Comparing 3D Model Quality and Accuracy

In this section, the host examines the 3D models produced by both services using the statue as an example. Luma AI's model is criticized for its inaccuracies and the presence of holes, while 3D Presser's model is praised for its better texture and cleaner 3D mesh. The importance of examining models in a 3D program like Blender is emphasized to assess their true quality and usability for purposes such as 3D printing or game development.

15:04

🌐 Final Verdict and Future Prospects

The final paragraph provides a summary of the comparison, noting that 3D Presser is better for creating accurate and usable 3D models, while Luma AI excels at capturing environments and larger entities. The host also discusses the limitations encountered when scanning certain objects and the potential of NeRF technology. The video concludes with an invitation for viewers to try the services themselves and a reminder to subscribe for more content on 3D scanning and related topics.

Keywords

AI-assisted application

AI-assisted applications are software programs that use artificial intelligence to assist users in completing tasks more efficiently. In the context of the video, these applications are used to generate 3D models from video images, which is a complex process that benefits from AI's ability to analyze and interpret visual data.

Luma AI

Luma AI is a product developed by Luma Labs, a Silicon Valley startup company. It is an AI application that specializes in creating 3D models from video footage. In the video, Luma AI is compared with another application for its ability to produce 3D models, showcasing its strengths and weaknesses in this domain.

3DPresso

3DPresso is a newer alternative to Luma AI, developed in Korea. It also uses AI technology to generate 3D models from video. The video discusses its capabilities and compares it with Luma AI, noting differences in processing speed, model accuracy, and texture quality.

Neural Radiance Fields (NeRF)

Neural Radiance Fields, also known as NeRF, is a cutting-edge technology used to generate 3D models from a collection of 2D images. It is a trendy AI technique that both Luma AI and 3DPresso utilize. The video explains how this technology is applied in the process of creating 3D models from video.

360 camera

A 360 camera is a device capable of recording panoramic images or videos. In the video, the 360 camera is used to capture video of objects for 3D scanning. It is particularly useful for scanning larger objects due to its wide field of view.

Insta360 RS1

The Insta360 RS1 is a specific model of a 360 camera that records up to 6K resolution video. It is used in the video to demonstrate how high-resolution 360 video can be utilized for 3D scanning purposes.

Photogrammetry

Photogrammetry is a technique of making measurements from photographs, especially for recovering the exact positions of surface points. In the context of the video, photogrammetry is the underlying principle for converting 2D video images into 3D models.

Horizon lock

Horizon lock is a feature of some 360 cameras that keeps the camera level even when the user moves or rotates the camera. This feature is mentioned in the video as important for ensuring that the final 3D model does not include the user or selfie stick.

Low poly model

A low poly model refers to a 3D model with a relatively small number of polygons, which makes it simpler and easier to render and display. The video discusses how the 3D models generated by the AI applications are initially low poly versions to be viewed in a web browser.

GLB format

GLB is a file format for 3D models that is designed to be efficient and easy to use for web applications. It keeps the 3D model and its textures in a single file. The video mentions exporting models in GLB format for further examination in 3D editing software like Blender.

Nerf rendering

Nerf rendering is a type of rendering that uses the NeRF technology to create a 3D representation of an object. In the video, Luma AI's automatic Nerf rendering is highlighted as a feature that produces visually impressive animations of the 3D models.

Highlights

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

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

Both applications use neural Radiance fields (NeRF) technology to generate 3D models.

Olioutron, the presenter, compares Luma AI and 3DPresso by using a 360 camera to create videos of three different objects for 3D scanning.

The 360 camera is particularly useful for scanning larger objects due to its wide field of view.

The scanning process involves recording three rounds around the object at different heights and angles.

Luma AI processes 3D models faster than 3DPresso, with results available within 30 minutes.

3DPresso has introduced a credit-based billing structure, with new users receiving 100 free credits.

Luma AI is currently free to use, with some limitations on rendering camera animations.

Luma AI's models tend to have a more accurate representation of the background but less accurate object shapes.

3DPresso produces more accurate and cleaner 3D mesh models, with better texture quality.

When examining the models in Blender, 3DPresso's models are more intact and less fragmented than Luma AI's.

Luma AI excels at capturing environments and larger entities, making it ideal for creating realistic NeRF models.

3DPresso is better suited for creating accurate polygon-based mesh models for 3D printing or game development.

Both applications have their own strengths and are useful for different purposes in 3D model generation.

The technology behind these applications is rapidly developing, with potential for future advancements.

The presenter encourages viewers to test these services themselves, noting that any smartphone can be used to produce videos for 3D scanning.

The video concludes with a call to like and subscribe for more content on 3D scanning and related topics.