🟢 Runway Gen3 VS Vidu - Tested 100 Scenes to Compare AI Video Generation
Yesterday, while quickly earning some Klings points, Runway Gen3's image-to-video feature suddenly went live!
I couldn't waste my unlimited points account (hehe)...
I was planning to speed up my new projects, and before I could even finish warming up my breakfast, Vidu also announced its global launch! 🌍
Did you guys plan this? Is there some auspicious date I missed?
Well, I had no choice but to do a comparison!
I believe no one can resist comparing their results.
Runway: The veteran AI video generator, known for its top-tier text-to-video capabilities.
Vidu: After competing fiercely with Sora in April and going silent for three months after 500 unsuccessful application attempts, it has made a surprising global comeback as the "Jumping Kid" (you'll see why later).
As usual, let's start with some of my favorite cases.
This time, I selected 100 videos to give everyone a good look.
I've extended and expanded the types of evaluations from last week. This time, we divided them into the following categories⬇️:
- Realism
- Animation
- Xianxia (Chinese fantasy)
- Sci-Fi
- Special Effects
- Action
- Multi-person Scenes
Parameters Summary
Runway Gen3: Supports 16:9 aspect ratio, video duration can be 5 or 10 seconds.
Vidu: Offers 80 free credits per month, free users can generate 4-second videos, and paid users 8-second videos—half of the previously advertised 16 seconds.
To ensure the comparison was accurate, I turned off Vidu's enhanced prompts and didn't use Runway's standard prompt structures. For the image-to-video test, I only used simple descriptive phrases.
After generating 50 bilingual prompts with GPT (Runway in English, Vidu in Chinese), Tom and I started testing again!
The article is limited, and the number of videos that can be sent is limited😌. We have put all the cases in the document, and you can see it by sending a "Vidu" in the comment area.
Realism
Realism is now a must-test style. The main subjects are divided into characters, landscapes, and animals.
Pay attention to subtle facial expressions, continuity of animal movements, and changes in light and shadow.
Image-to-Video Characters
A man raises his hand to block the sunlight.
A man slowly walks forward.
A woman blinks, with beautiful light effects in her eyes.
Both tools performed well in generating realistic character actions, lighting, and details. However, Vidu's movements were significantly more pronounced than Gen3's.
Image-to-Video Animals
A fluffy water monster disappears into the waves.
A cute squirrel hops forward.
My adorable squirrel not only has ADHD but can also turn invisible? What are you thinking, Vidu? Overall, both Gen3 and Vidu did a great job recognizing and animating known and unknown creatures, though Vidu's output was a bit too lively.
Image-to-Video Landscapes
The sun slowly rises; time-lapse photography.
At this point, I have to praise Vidu! Gen3's interpretation of the landscape was as expected, but Vidu's transition effects really surprised me! Very natural, seamless, and beautiful, though the clarity was somewhat lacking. Vidu, please update to high resolution soon!
Conclusion: Most tools now have little to no flaws in realism. The main differences are in the amplitude of motion and camera effects. Currently, except for the resolution issue, Vidu's video effects are top-notch!
Animation
Due to its distinct style, animation is one of the most frequently mentioned visual forms.
This time, we tested popular styles like Pixar, Makoto Shinkai, and classic 2D animation.
Image-to-Video Characters
A girl shows a terrified expression.
A magician conjures sparks of smoke.
A boy sits on a windowsill, looking out at the scenery as the wind gently blows.
In the first set of scenes, Vidu suddenly turned into an Indian girl? Also, Gen3 loves zooming in when there are no restrictions on camera movement!
Image-to-Video Animals
A cute little animal tilts its head and blinks.
We used an unknown cute animal for generation, and Vidu at least made it look like an animal with natural movements, while Gen3 turned it into a plush toy!
Image-to-Video Landscapes
Raw output (no prompt).
Conclusion: Both tools did well in showcasing non-realistic animation styles. Vidu has more imagination, while Gen3 excels in detail.
Xianxia (Chinese Fantasy)
Chinese-style videos are our specialty. The challenge here is generating video from images since traditional images don’t move—pure PPT style!
A female warrior rides a horse, with her clothes fluttering in the wind.
A warrior flies towards a distant palace, holding a sword.
Flower petals fall from the sky, the wind gently blows, and the camera zooms in.
The wind blows, a girl looks at distant lights, and the camera slowly zooms out from beside her.
Conclusion: When it comes to Chinese fantasy, our domestic model excels at making things move. Vidu is the first tool I've tested where the horse actually runs! Amazing! And the imagination and details in the distant fairyland scenes are commendable. If they update with high resolution and longer durations, the results will be unbelievable! (Vidu, are you listening? This is a hint!)
Sci-Fi
You should be familiar with sci-fi scenes by now.
When testing this style, we mainly assess the ability to simulate lighting and special effects.
Raw output (no prompt). By the way, did you know Iron Man turned into Doctor Doom? 🤯
A cosmic explosion.
A doomsday warrior walks towards a spaceship, dust flying, and the sky dramatically changing.
Conclusion: The sci-fi style comparison results are similar to the above. Vidu has more movement, while Gen3 offers better detail—both with their strengths.
Special Effects
Special effects shots are mainly used to test the model's ability to simulate the real world.
This usually includes simulating movement, collision, rotation, and other physical phenomena.
A bucket of paint is thrown on a statue.
A flower bud slowly grows and blooms.
Several colors slowly flow and merge together.
Abstract shapes morph; a geometric eye blinks as the shapes and colors change.
Conclusion: When it comes to abstract special effects, nothing can compete with Gen3. The last scene, with the geometric shape changing and blinking eye, is something I absolutely adore!
Action
I separate action because, in video production, I really want more interactive scenes. Currently, many models produce results that are just moving PPTs. 🤣
Two people duel, their lightsabers clashing and creating sparks.
A man is forging iron.
Conclusion: Finally, a fight! Vidu is the first to deliver such smooth action scenes!
Multi-person Scenes
Wide-angle, multi-person scenes are still a weakness in AI-generated videos.
Often, when trying to create multi-person scenarios, AI results in overlapping, distorted characters or fails to generate the correct scene.
Raw output (no prompt).
Raw output (no prompt).
Conclusion: We've tested several tools with such wide-angle multi-person scenes. Vidu stands out for its depiction of smoke, flames, and group movement!
Final Thoughts
Phew, I showcased 66 videos!
Due to article length restrictions, you can view all the videos and corresponding images by following my WeChat official account and messaging "Card Draw" to receive the document.
AI video updates are now happening by the hour.
As tools continue to evolve, the chances of getting a "golden card" increase with each update!
I initially expected Vidu to perform as anticipated, launching during this competitive period. However, it delivered many surprises 🤯, raising my expectations for Vidu's future updates! Keep pushing, Vidu!
It's also heartening to see so many tools updating together during this time.
It truly feels like a hundred schools of thought are contending, and it's great to see these tools helping more people freely express the images in their minds. This is wonderful!
By the time this article is published,
Midjourney v6.1 has already been released!
Half of my brain is already generating images,
Given that half of video creation depends on images,
I'm looking forward to the next review 🐱.
Oh, and we've decided that all future reviews will follow this standard, now named the "100 Lens Series".
What do you think?