Morph Studio: Free, 1080P, 7-second powerful dark horse video editing software is coming

"Glowing jellyfish slowly rise from the ocean," continue typing in what you want to see in Morph Studio, "turning into shimmering constellations in the night sky."

After a few minutes, Morph Studio generates a short video. A jellyfish is completely transparent and sparkling, spinning and rising, its swaying figure contrasting with the stars in the night sky.

Luminous jellyfish rise from an enchanting sea, morphing into glittering stars in the dark sky.

Enter "joker cinematic", which once swept the world That face is back.

Joaquin Phoenix delivers a hyper-realistic performance as the Joker in the cinematic shot set in the neon-lit streets of New York. Smoke billows around him, adding to the atmosphere of chaos and darkness.

Recently, startup Morph Studio has made an important update to its text-to-video generation technology and community. Their model has undergone a major update, and these video productions show off the updated model, with clear images and vivid details.

Morph Studio is the first team in the world to publicly launch a text-to-video product that can be freely tested by the public, earlier than Runway launched Gen2 in public beta.

Compared with other popular text-to-video products, Morph Studio is different in terms of free services. It offers a default 1080P resolution and a maximum build time of 7 seconds from the start. For text-to-video products, higher resolution, longer generation time, and better expression of intent are three key indicators. Morph has reached the latest level in the industry in these three indicators.

The average length of a single shot of a Hollywood movie is 6 seconds. Extending the generation time to 7 seconds can unlock the creative needs of more users.

It’s easy to experience Morph Studio’s models. You can use it for free by registering on discord.

The model with the word "pro" in the red box on the screen is the updated model and is the subject of this article's experience.

Camera movement is the basic language of video production and a powerful narrative device. Morph provides several general camera languages, including zoom, pan (up, down, left, right), rotation (clockwise or counterclockwise), and still shots.

Morph also provides the MOTION function (1-10) to control video movement. The larger the value, the more violent and exaggerated the action. The smaller the value, the more subtle and smooth the action.

The frame rate (FPS) provides an adjustment range from 8 -30. The higher the value, the smoother the video and the larger the size. For example, -FPS 30 will produce the smoothest but also largest video. By default, all videos are created at 24 frames per second.

The default video length is 3 seconds. To generate a 7-second video, you can enter -s 7 in the command. Additionally, the model offers 5 video ratios to choose from.

If you have requirements for details such as lens, frame rate, and video length, please continue to enter the corresponding parameters after entering the content prompt. (Currently only English input is supported.)

We have experienced the updated model service and strongly feel the visual shock brought by 1080P.

Until recently, humans had the first photo of a snow leopard walking under the stars:

The first human photo of a snow leopard walking under the stars.

We want to know, can Morph Studio’s model be able to generate this relatively rare animal video?

With the same prompt, we put the works of Morph Studio in the upper part of the video, and the works generated with Pika in the lower part of the video.

a snow leopard walking under a starry night, cinematic realistic, super detail, -motion 10, -ar 16:9, -zoom in, -pan up, -fps 30, - s 7. negative: Extra limbs, Missing arms and legs, fused fingers and legs, extra fingers, disfigure

Morph Studio’s answer sheet, the text is understood accurately. In the 1080P picture, the snow leopard’s fur is rich in detail and lifelike. The Milky Way and stars can be seen in the background. However, the snow leopard's movement is not obvious.

In Pika’s homework, the snow leopard is indeed walking, but the night sky seems to be understood as a night with goose feathers and heavy snow. There is still a gap in terms of Snow Leopard style, details, and picture clarity.

Let’s take a look at the effect of character generation.

masterpiece best quality ultra detailed RAW video 1girl solo dancing digital painting beautiful cyborg girl age 21 long wavy red hair blue eyes delicate pale white skin perfect body singing in the eerie light of dawn in a post-apocalyp

In the works generated by Morph Studio, the high resolution brings extremely delicate facial contours and micro-expressions. Under dawn light, hair details are clearly visible.

Due to the lack of resolution, color, and light levels, the overall picture generated by Pika is bluish, and the facial details of the characters are not satisfactory.

People and animals have experienced it, let’s take a look at the generation effect of buildings (man-made objects).

La torre eifel starry night van gogh epic stylish fine art complex deep colors flowing fky moving clouds

and Pika’s works are more of a painting texture Compared with the picture, Morph Studio's work better balances Van Gogh elements and realistic elements. The light levels are very rich, especially the flowing details of the sea of clouds. The sky in Pika's work is almost static.

Finally, experience the creation of natural scenery.

One early morning the sun slowly rose from the sea level and the waves gently touched the beach.

You may wonder about Morph Studio The works are not real shots taken by human photographers under natural conditions.

Because the video generated by Pika lacks delicate light and shadow levels, the waves and the beach appear flat, and the movements of the waves hitting the beach are relatively dull.

In addition to the shocking experience brought by high resolution, with the same prompt to generate videos (such as animals, buildings, people and natural scenery themes), opponents will more or less " "Miss", Morph Studio's performance is relatively more stable, there are relatively fewer corner cases, and it can predict user intentions more accurately.

From the beginning, this startup’s understanding of Vincent Video is that the video must be able to describe user input very accurately, and all optimization work is also moving in this direction. The model structure of Morph Studio has a deeper understanding of textual intentions. This update has made some structural changes and specially made more detailed annotations for some data.

In addition to the relatively good text understanding ability, the detail processing of the picture is not stumped by the high-resolution output. In fact, after the model was updated, the motion content of the screen was richer, which is also reflected in the works we generated using Morph Studio.

"Girl with a Pearl Earring" When the head moves, the earrings are also shaking slightly; the scenes involving more complex actions such as horseback riding are also more smooth, coherent and logical, and the hand movements The output is also good.

1080P means that the model has to process more pixels, which brings greater challenges to detail generation. However, judging from the results, not only did the picture not collapse, but because of the rich layers More detailed and more expressive.

This is a set of natural landscapes we generated using models, including spectacular huge waves and volcanic eruptions, as well as delicate close-ups of flowers.

High-resolution output brings better visual enjoyment to users, but it also prolongs the model output time and affects the experience.

Morph Studio now generates 1080p videos in 3 and a half minutes, the same speed as Pika generates 720P videos. Start-up companies have limited computing resources, so it is not easy for Morph Studio to maintain SOTA.

In addition, in terms of video style, in addition to movie realism, Morph Studio models also support common styles such as comics and 3D animation.

Morph Studio’s focus on text-to-video technology is regarded as the next stage in the AI industry competition.

“Instant video may represent the next leap in AI technology,” the New York Times said in a headline of a technology report, arguing that it will be as important as the web browser and the iPhone.

In September 2022, Meta’s team of machine learning engineers launched a new system called Make-A-Video. Users enter a rough description of the scene, and the system will generate A corresponding short video.

In November 2022, researchers from Tsinghua University and Beijing Academy of Artificial Intelligence (BAAI) also released CogVideo.

At that time, the videos generated by these models were not only blurry (for example, the video resolution generated by CogVideo was only 480 x 480), the pictures were also relatively distorted, and there were many technical limitations. But they still represent a significant development in AI content generation.

On the surface, a video is just a series of frames (still images) put together in a way that gives the illusion of movement. However, it is much more difficult to ensure the consistency of a series of images in time and space.

The emergence of the diffusion model has accelerated the evolution of technology. Researchers have attempted to generalize the diffusion model to other domains such as audio, 3D, and video, and video synthesis technology has made significant advances.

Technology based on the diffusion model mainly allows the neural network to automatically learn some patterns by sorting out massive images, videos and text descriptions. When you input content requirements, the neural network generates a list of all the features it thinks might be used to create the image (think the outline of a cat's ears, the edges of a phone).

Then, a second neural network (also known as the diffusion model) is responsible for creating the image and generating the pixels needed for these features, and converting the pixels into a coherent image. By analyzing thousands of videos, the AI can learn to string many still images together in a similarly coherent manner. The key is to train a model that truly understands the relationships and consistency between each frame.

“This is one of the most impressive technologies we’ve built in the last hundred years,” Runway CEO Cristóbal Valenzuela once told the media. “You need to make people really Use it.”

2023 is regarded by some in the industry as a breakthrough year for video synthesis. There was no publicly available text-to-video model in January, and by the end of the year there were dozens of similar products and millions of users.

a16z partner Justine Moore shared the timeline of Vincent’s video model on social platforms. We can see from it that there are many other models besides big manufacturers. Startup companies, in addition, the speed of technology iteration is very fast.

The current AI-generated videos have not formed a unified and clear technical paradigm similar to LLM. The industry is still in the exploratory stage on how to generate stable videos. But the researchers believe these flaws can be eliminated as their system is trained with more and more data. Eventually, this technology will make creating videos as easy as writing sentences.

A senior domestic AI industry investor told us that several of the most important papers on Vincent Video Technology were published from July to August 2022, which is similar to the industry of Vincent Picture In the process of industrialization, the point where this technology is close to industrialization will appear one year later, that is, from July to August 2023.

The entire video technology is developing very fast, and the technology is becoming more and more mature. This investor said that based on previous investment experience in the GAN field, they predict that the next six months to one year will be The productization period of text-to-video technology.

The Morph team brings together the best young researchers in the field of video generation. After intensive research and development day and night over the past year, founder Xu Huaizhe and co-founders Li Feng, Yin Zixin, Zhao Shihao, and Liu Shaoteng Together with other core technical backbones, we have overcome the problem of AI video generation.

In addition to the technical team, Morph Studio has also recently strengthened its product team. Hai is the contracted producer of Maoyan Films, a judge of the Shanghai International Film Festival, and a core member of the former Silicon Valley head AIGC company. Xin also recently joined Morph Studio.

Heising said that Morph Studio occupies a leading position in the entire industry in terms of technical research; the team is flat, communication efficiency and execution are particularly high; every member is passionate about the industry. Her biggest dream was to join an animation company. After the advent of the AI era, she quickly realized that the animation industry would change in the future. In the past few decades, the animation base was 3D engines, and a new era of AI engines would soon usher in. The Pixar of the future will be born in an AI company. And Morph was her choice.

Founder Xu Huaizhe said that Morph is actively laying out the AI video track. We are determined to be a Super App in the AI video era and realize dreams for users.

The track will have its own Midjourney moment in 2024, he added.

PS: To experience the original fun of free 1080P video generation, please go to:

https://discord.com/ invite/VVqS8QnBkA

The above is the detailed content of Morph Studio: Free, 1080P, 7-second powerful dark horse video editing software is coming. For more information, please follow other related articles on the PHP Chinese website!