Getestet 7 Artefakte zur Videogenerierung auf „Sora-Ebene'. Wer hat die Fähigkeit, den „Eisernen Thron' zu besteigen?

Machine Power Report

Herausgeber: Yang Wen

Wer kann der König des KI-Videokreises werden?

In der amerikanischen TV-Serie „Game of Thrones“ gibt es einen „Eisernen Thron“.

Der Legende nach wurde es vom Drachen „Schwarzer Tod“ hergestellt, der Tausende von vom Feind weggeworfenen Schwertern zum Schmelzen brachte und so höchste Autorität symbolisierte. Um auf diesem eisernen Stuhl zu sitzen, begannen die großen Familien zu kämpfen und zu kämpfen.

Seit der Entstehung von Sora hat der KI-Videokreis auch ein energisches „Game of Thrones“ ins Leben gerufen. Zu den Hauptakteuren in diesem Spiel gehören Runway Gen-3 und Luma von der anderen Seite des Ozeans sowie die einheimischen Kuaishou Keling und ByteDance. Meng, Zhipu Qingying, Vidu, PixVerse V2 usw.

Heute werden wir bewerten und sehen, wer qualifiziert ist, auf dem „Eisernen Thron“ des KI-Videokreises zu sitzen.

-1-

Vincent Video

Gegenwärtig ist die Vincent Video-Funktion zur „Spezialität“ verschiedener KI-Videoprodukte geworden, daher betrachten wir sie als Vorspeise.

Eingabeaufforderung 1: Die Kamera blickt direkt durch ein Fenster auf ein Gebäude im Erdgeschoss, während ein süßer Dalmatiner aus dem Fenster im ersten Stock schaut.

Keling:

Dream:

                                

PixVerse V2:

Luma:

Runway Gen-3 :

Jedes Haus hat farbenfrohe Gebäude hervorgebracht. Aus der Perspektive der Bewegungsreichweite und der Erlesenheit des Bildes hat Runway Gen-3 die beste Wirkung, gefolgt von Linghejimeng, das auch ziemlich seidig ist.

Eingabeaufforderung 2:

Nächtliche Aufnahmen von Einsiedlerkrebsen, die Glühbirnen als Muscheln verwenden.

Keling:

Traum:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen-3:

In most videos, either the hermit crab moves its pincers a few times or pans the camera. Generally speaking, Vidu and Runway Gen-3 have the greatest video motion. In the video generated by Vidu, the hermit crab almost disappeared in a flash, while Runway Gen-3 showed the crawling process of the hermit crab.

Luma’s hermit crab looks the strangest, round and round, a bit like a cockroach crawling all over the ground.

Prompt 3: The story of a robot’s life in a cyberpunk setting.

Keling:

Dream:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen -3:

Except for Luma, which did not show a decent robot, all other companies showed their own imaginary robot images.

The video images generated by Keling are similar in composition and lens to the example of "Tokyo Women Walking on the Street" released by Sora; Dream's images are the most exquisite and cyberpunk, while Vidu has the look of a science fiction film. The robot in PixVerse V2 is the weirdest, like it has a bucket on its head.

In terms of realism and range of motion, Runway Gen-3 seems to be better. Of course, it also has bugs, such as cars appearing out of thin air and speeding backwards.

Prompt 4:

The waves are lapping against the shells on the beach.

Keling:

Dream:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen -3:

Overall, the effects generated by Keling and Dream are the most realistic. Luma’s shell looks a bit deformed. The most outrageous thing is PixVerse V2. The picture is normal at first, but it disappears in the blink of an eye. A person's facial expressions and hands were so messed up that he couldn't see anything.

Prompt 5：

Cartoon style, a kangaroo is dancing

Prompt word 5:Cartoon style, a kangaroo is dancing.

Keling:

Instant Dream:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen -3:

AI understands that the anime-style kangaroos are really diverse, some look like mice, and some look like unknown species. In comparison, Coring, Vidu and Runway Gen-3 have better effects, at least there is no large-area blurring.

-2-

TuSheng Video

These video generation models that claim to be the benchmark for Sora have also launched the "TuSheng Video" function. You only need to upload a picture and enter the prompt word. Can make static pictures move.

Round 1

Upload a picture of Pikachu and enter the prompt word: Pikachu jumps up happily (It jumps up happily).

Keling:

Real Dream:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen-3:

Vidu won this round. Not only did it have a large range of motion, smooth expressions and movements, but the screen did not collapse during the jumping process. But I didn’t expect that the worst effect in this round was Runway Gen-3. All kinds of weird transformations.

Round 2

Upload a famous oil painting "American Gothic" by Grant Wood and enter the prompt word: They laugh.

Keling:

Real Dream:

Qingying:

Vidu:

PixVerse V2:

Luma:

Runway Gen-3:

From the perspective of natural expression, I choose Runway Gen-3, Keling and Vidu for this one, what do you think?

-3-

The editor has something to say

Runway is worthy of being the "hands-on" in the AI ​​video circle, Gen-3 is really capable, except for Pikachu overturning, other tests All performed well.

The second is Ke Ling, which generates relatively stable images and follows prompt words quite well.

Jimeng has the strongest control over color, and the color matching of the videos it generates is the most comfortable. The disadvantage is that when the movement range is too large, the picture is easily blurred and deformed.

Vidu has the best animation effect. It is very good at making animation characters dynamic and the generation speed is fast. It only takes more than 30 seconds to generate a 4-second video.

Luma, this living tool that once made netizens around the world crazy, in this actual test, compared to Keling and Runway Gen-3, its performance is not outstanding, and the generation speed is extremely slow. Sometimes a period of 5 A second video would take more than half an hour, or even longer.

From the perspective of payment situation, currently except Runway Gen-3, which requires full payment, other companies have free trial times. Among them, Zhipu Qingying is in the internal testing stage and can be generated unlimited times for free.

In order to meet the needs of different users, each company divides membership into three types: basic, standard and advanced. The prices are not cheap. Calculated based on annual payment, the most expensive one is Luma. Basic members need to pay every year. At $287.9, this price can be said to be unparalleled among many applications.

Which company do you think has the strongest AI video generation capability? Let’s chat in the comment section.

Tool link:

Kuaishou Keling: https://klingai.kuaishou.com/

Byte Dream: https://jimeng.jianying.com/ai-tool/home ?activeTab=video&subTab=video

生生Vidu: https://www.vidu.studio/

智典清影: https://chatglm.cn/video

PixVerse: https: //pixverse.ai/

Luma: https://lumalabs.ai/dream-machine

Runway Gen-3: https://runwayml.com/

We will bring more in the future We also welcome everyone to join the group for first-hand evaluation of multiple AI large models and AI applications.

Das obige ist der detaillierte Inhalt vonGetestet 7 Artefakte zur Videogenerierung auf „Sora-Ebene'. Wer hat die Fähigkeit, den „Eisernen Thron' zu besteigen?. Für weitere Informationen folgen Sie bitte anderen verwandten Artikeln auf der PHP chinesischen Website!