Agnes Video 2.5 Flash
Folks, if there were a video AI that was not only free but also top-tier, wouldn’t you be tempted to try it?
As it happens, a new video AI has just been released that checks all those boxes—
This is the newly released Agnes Video 2.5 series from the renowned AI lab Agnes, now available on the Pavo creative platform:
- Agnes Video 2.5 Flash: Free to use with no credit consumption, making it ideal for AI short-drama creators who need to generate videos in batches or experiment during the early stages.
- Agnes Video 2.5: Comes with 200 credits per day, consuming 2 credits per second. It can generate videos up to 100 seconds long, supports multiple images as well as audio and video references, and offers video generation at up to 2K resolution. It is better suited for real-world commercial applications such as short dramas, advertisements, branding, e-commerce, and games.
So, how does it actually perform?
Let’s start with our recreation of the viral Douyin hit 《云上天宫》, made using Agnes Video 2.5 Flash:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
It’s absolutely ethereal.
Although this video AI is free, its results are definitely nowhere near the level of cheap visual effects.
And judging from its latest performance on the Artificial Analysis leaderboard, Agnes Video 2.5 has also reached the level of the leading players worldwide:
Compared with Agnes’s previous 2.0 version, Agnes Video 2.5 has climbed significantly higher on the AA leaderboard. After conversion to RMB, the 2.5 version costs just RMB 0.15 per second.
Next, as usual, let’s take it for an in-depth test drive.
Come on, let’s make a short drama for free!
When creating 《云上天宫》 just now, we used the relatively common “first frame + prompt” method.
However, the Pavo platform also offers a feature called Canvas for video creation:
So for the next round of testing, we’ll use Canvas to complete our first task: creating a polished version of 《牛来》.
In Canvas, you can add all kinds of assets, including text, images, videos, and audio. Naturally, you can also generate content directly within the canvas.
Every node can be linked to the others, and you can continue generating from any intermediate node. The same asset can also be branched into multiple variations:
In short, for every creator, Canvas is like a private theater where all assets can be freely arranged and manipulated.
For example, we can first connect the 3D character image of “牛来” with the background image of an outdoor scene, then generate the initial frame for the first shot:
If we’re happy with the generated first frame, we can add the next node and begin creating the video for shot 1:
For scenes containing multiple characters, we can even use the 3D Director’s Stage to precisely control their positions, poses, proportions, and more:
By generating assets step by step on the canvas according to the storyline, the polished version of 《牛来》 eventually came to life:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
How convenient is that?
And when creating videos in Canvas, simply select Agnes Video 2.5 Flash to continue using it for free, with zero credit consumption.
If you want to make a short drama, Pavo also offers a dedicated Story Short Film mode:
After you enter a simple idea, you’ll enter a co-creation process with AI:
The entire process includes requirement confirmation, script outline development, character, setting, and prop design, storyboard scripting, storyboard keyframe generation, storyboard video generation, and final video composition.
Better yet, most of it is handled automatically by AI. All we need to do is confirm each stage:
After these few steps, an AI short drama is born:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
Of course, it can also handle advertising videos that can be used directly in commercial scenarios:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
And imaginative promotional videos:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
It can even create music videos in specific styles, such as the early-2000s aesthetic:
Video link:
https://mp.weixin.qq.com/s/N-YL-SXFuIhv-5RxYtCLRA
The key point is that all of the videos above were created for free—for free, for free, for free!
Trial and error? Be bolder.
Objectively speaking, though, there’s one thing we still have to admit about the current state of video AI: getting a perfect result in a single take is basically unrealistic.
The character’s movements are slightly off? Start over. The camera movement doesn’t meet expectations? Start over. After finally getting everything else right, the character’s face suddenly changes into someone else’s? Start over again.
And that’s not even considering the fact that producing a complete work such as a short drama or music video can require dozens or even hundreds of shots. During the early stages alone, experimenting with composition, style, and movement inevitably means generating one version after another.
That’s when the value of the word “free” really becomes clear.
When we were making 《牛来》 earlier, for example, we could easily generate two, three, or even more versions of a shot. The action isn’t funny enough? Try another one. The camera angle doesn’t quite work? Give it another shot.
Especially in real content production, what you save isn’t limited to generation costs.
As mentioned earlier, Canvas supports continuing the creative process from any intermediate node. The same asset can also branch into multiple variations, allowing all previously approved versions to be preserved.
For example, suppose the character, setting, and composition in a shot have all been adjusted, but the action is the only problem. You can simply regenerate from that node.
The same applies to the 3D Director’s Stage.
Where a character stands, what pose they take, and how they are proportioned relative to one another—details that were previously easy to leave to the model’s discretion—can now be configured much more precisely in advance. Combined with the ability to reuse core assets such as characters and settings, this means you no longer need to describe the same character from scratch every time you create dozens of shots.
So after Agnes Video 2.5 Flash became free, the most exciting part may not simply be the ability to generate a few videos for free.
It may be that we can finally afford to experiment boldly.
Free is only the first step
Looking at the previous rounds of testing as a whole, it’s not hard to see that video AI is quietly changing the way the game is played.
Earlier on, whenever a new video model was released, people were mainly concerned with a few questions: Is the image quality realistic enough? Will the characters fall apart? Are the movements smooth? Does the camera work feel cinematic?
After all, at the time, a sufficiently impressive 10-second demo was enough to take the internet by storm.
But things are clearly different now. From short films and music videos to advertisements and short dramas, more and more AI-generated videos are no longer satisfied with producing a single beautiful shot. They are beginning to create complete works, shot by shot. Pavo’s focus on AI short dramas is itself a direct response to this shift.
A new problem follows: making one shot look good is important, of course. But when the number of shots grows from 1 to 50 or 100, things become far more complicated.
As a result, video AI will clearly be competing across an increasingly broad range of dimensions.
The models themselves still need to be powerful enough. But beyond that, factors such as cost, controllability, consistency, and the ability to connect the entire workflow—from script and characters to storyboards, video, and the final cut—will also increasingly determine whether a work can actually be completed.
Once quality becomes reasonably stable and the cost barrier drops even further, creating videos with AI will ultimately return to a more fundamental question:
Is your idea Amazing enough?