Source: Machine Power

Image Source: Generated by Wujie AI

Wensheng 3D models have new players again.

With the wave of generative AI sweeping in, tech companies are all vying for the next breakthrough in AIGC - 3D generation.

Currently, Wensheng 3D has made many new breakthroughs in the academic field. In the industry, both industry giants like NVIDIA and mid-level forces like Stability AI have launched their own Wensheng 3D tools. Although the 3D generation field has been "rolling" for a long time, it seems that there has never been a representative product with stunning effects and fast generation speed.

Now, Luma AI, which has been committed to 3D content production, has launched a challenge in this field. Just yesterday, Luma AI released a project of Wensheng 3D model - Genie 1.0. With just a text prompt, Genie 1.0 can generate four high-fidelity 3D models in 10 seconds. Users can choose from them and edit the texture of the models in a three-dimensional grid interface.

These AI-generated 3D models can be previewed on the screen and exported to tools such as Blender, Unity, etc. for further use, directly integrating into animation, games, VR, and other workflows.

Genie 1.0 is not only available on the web, but can also be tried on the Discord platform and mobile devices. Once released, the eagerly awaited netizens pushed Genie 1.0 to the top of the App Store download list.

Trial address: https://lumalabs.ai/genie?view=create

How amazing is the effect, let's take a look together.

Simply enter the prompt "flower cat", and after waiting a few seconds, four cats with different floral patterns on their bodies appear. If you want a figurine, the AI directly offers exquisite blind box versions, with four chances to draw with just one input, it's a great deal no matter how you calculate it.

Even the cute figurines in the style of Pop Mart can be arranged. The AI version can achieve mass production of hidden styles.

The difficulties in Wensheng 3D, such as fur and surface patterns of objects, are also stunningly presented by Genie 1.0. The panda it generates has a furry texture, and the patterns on the clothes do not show obvious distortion.

Twitter @lintworld

If "Star Wars" were to have a 3D animated version, it seems that there would be no problem using the "Master Yoda" generated directly by Genie 1.0.

Twitter @blizaine

Compared with other 3D models, Luma AI's performance in terms of three-dimensionality and realism also takes the lead.

Twitter @taziku_co

In addition to the stunning generation effect, Genie 1.0 also has editing functions. Removing the original professional threshold and bidding farewell to pasting materials, it is as simple to modify the material of a 3D model as using a mobile phone to edit pictures.

Genie 1.0 provides three textures: "original," "smooth," and "reflection," which can be switched by clicking the circular button on the right side of the page. It can also export models of different resolutions and formats as needed, seamlessly integrating with other 3D editing software.

In the past, batch designing 3D models often took several days, although many AIGC tools on the market have already shortened the time for 3D design to just a few hours or even less. However, Genie 1.0 has compressed people's processing time to 2 minutes.

Some 3D designers have already stated that they imported the models generated by Genie 1.0 into the 3D animation software Maya, and because the quality of the AI-generated models is very high and the texture coordinates (UVs) are very suitable, with only a few small overlaps, it only takes a small adjustment for 2 minutes, and they can be used directly with a first-class experience.

By adding a few action anchors in the animation software, the 3D animation is created.

Twitter @taziku_co

With a little editing on this basis, the static model instantly performs a mechanical dance, bringing a seamless performance of a dozen seconds.

Twitter @paradoxrenders

Transforming it into a game animation has also evolved into a speedrun version, where you can generate game character animations with just a few keystrokes, which may no longer be a dream.

The basic model is generated by Genie, sourced from YouTube@OlliHuttunen78

Although Genie 1.0 has not completely overcome the "last mile" problem of AI-generated 3D models often being relatively blurry, after introducing Magnific AI to enhance the resolution, the patterns on the truck tires, the folds of the exhaust pipes, and the wires connecting the parts are all outlined with clear details, with no difference from truck ornaments in the real world.

Twitter @MartinNebelong

What Genie 1.0 can do hits the pain points of 3D design directly. The traditional 3D modeling process is usually cumbersome, often requiring a large amount of manual operation by professional personnel, which not only leads to an extended production cycle but also increases overall costs. After entering further design processes such as animation, games, and VR, modeling needs further adjustments and modifications, leading to a doubling of design costs.

Perhaps, as Luma AI founder Amit Jain said, "With Genie, using AI to mass-produce 3D objects may no longer be just a fantasy."

With the support of Genie, even if you don't know 3D modeling and painting design, any idea involving 3D graphics or scenes can be realized like an elf.

Currently, Genie has both a free version and a paid API. The paid interface requires a payment of $1 each time. The company points out that this is cheaper and faster than relying on human designers.

Luma AI Company Introduction

Luma AI, the company that launched Genie 1.0, is a generative artificial intelligence startup that has been committed to 3D content solutions, with its main products and features including image-to-3D, video-to-3D, text-to-3D, etc.

The company was founded in 2021. CEO Amit Jain was a computer vision systems engineer at Apple, and CTO Alex Yu was a graduate student at the University of California, Berkeley (he dropped out of his doctoral studies to found Luma AI). Both have made significant contributions in the fields of 3D vision, machine learning, real-time graphics, and more.

The company has gone through several rounds of financing, raising $20 million in a Series A round led by Amplify Partners, Nventures (Nvidia's investment department), and General Catalyst, and $43 million in a Series B round led by top Silicon Valley venture capital firms Andreessen Horowitz and Nvidia. As of now, the company has raised over $70 million and is valued between $200 million and $300 million.

In terms of its products, unlike the digital art bitmap output by text-to-image models, Luma AI uses AI to create 3D models of objects from photos, videos, or text descriptions, which can be downloaded, manipulated, edited, and rendered as needed.

In November of last year, Luma AI launched the Wensheng 3D model Genie on the Discord server. This 1.0 version reduced the rendering time from over 20 seconds to within 10 seconds.

Luma AI's core technology is the Neural Radiance Fields (NeRF) algorithm. Luma AI has made many improvements and optimizations based on NeRF. In terms of scene image acquisition, data preprocessing, model training, model rendering, and the generation of other model files, Luma AI has done a lot of optimization work. On this basis, more skilled human artists can efficiently create better models. In addition, the output of Genie may also meet the specific production needs of some people.

On the same day, Luma AI also announced three new members, with Chief Scientist Jiaming Song leading the basic model research team. He graduated from Tsinghua University with a bachelor's degree and from Stanford University with a master's and doctoral degree. Before joining Luma AI, he was a research scientist in Nvidia's Learning and Perception research team and Deep Imagination research team.

In addition, UC Berkeley Ph.D. Matthew Tancik will lead the applied research team, and Tuhin Kumar from Apple Design Studio will lead the interface, interaction, and product design team.

Reference links:

https://techcrunch.com/2024/01/09/luma-raises-43m-to-build-ai-that-crafts-3d-models/

https://twitter.com/LumaLabsAI/status/1744778363330535860

https://www.theregister.com/2024/01/10/textto3dimagestartupluma/_

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。