Genspark Releases Video Agent Workspace Media Studio, Integrating 70+ Models

Altman and Dario rapping on the same stage! Genspark's new video Agent is live—even beginners can shoot short dramas

Genspark's video Agent workspace, Genspark Media Studio, has officially launched!

The large language model serving as the "director" comes in three options: DeepSeek V4.1 Flash, Claude Opus 5.5, and Claude Fable 5.1.

Yes, you can even use Fable 5.1 in here...

Honestly, for certain well-known reasons, I haven't been able to use it for a long time...

The models doing the actual work include: Seedance 2.5, Kling, and Veo for video generation, GPT Image and Nano Banana for images, and audio models such as ElevenLabs for sound.

In total, more than 70 model versions from over 20 vendors are integrated.

And there's a discount right now!

From 18:00 Beijing time on October 27 to 18:00 on November 4, Plus and Pro members get 50% off credit consumption in Media Studio, covering planning, image, video, audio, and 3D generation.

QuantumBit, always surfing the 6G wave, also got hands-on testing access right away.

Let's dive right in~

Hands-on with Genspark Media Studio: Altman rapping and a business-war short drama, all arranged

Get the same template with one click

In the Genspark Magic section on the homepage, there are many viral video templates—click to make the same one with one click.

I carefully picked a rap video and uploaded photos of Altman and Dario.

The two brothers of the US stock IPO—that's you two!

But there was a problem I hadn't considered: if you upload real people's photos directly, the video model will block them.

Genspark also reminded me to confirm copyright issues before shooting—quite thoughtful.

The system again prompted whether to use the semi-realistic version or the cartoon version, and I chose the semi-realistic version (taking a gamble). After confirming everything, it got to work.

The Agent called Seedance 2.5. Although it said 8–14 minutes to produce the video, my sense was that it was done in about two or three minutes (?).

The first version had no soundtrack—just the two of them singing dry—so I asked it to find backing music.

This canvas is quite interesting: it found three candidate backing tracks and laid them out side by side on the canvas, so you can click whichever one you want to watch:

(Though I don't know why backing track C is so far away...)

But none of these three BGM tracks had lyrics, so in the end I had it make an original rap backing track, generated via ElevenLabs.

I also did some lip-sync alignment so the two could sing together.

The final cut is like:

Pretty good—I got to see Altman and Dario rapping on the same stage, and they look like they've been practicing for two and a half years.

Worth mentioning: the first-pass generated videos are all 480p. If you're not satisfied, you can keep adjusting, and upgrade to 1080p after finalizing.

Knowledge explainer videos in seconds

Next, I had it make a knowledge explainer video explaining the movie "Coherence."

You can enter via the "Knowledge Explainer" card below the input box on the homepage, or just describe it in natural language.

Then the system prompted me to choose the video length, visual style, and narration angle, and also choose an audio model—here I used MiniMax TTS.

I didn't upload any materials at all and just let it do everything itself.

It really felt like it came out in seconds... I was still working on another video, and when I turned back it was done:

But the first version, how should I put it, emmm...

The script was rather bland and made you want to sleep, and the visuals were like a PPT, stuttering frame by frame.

So I had it revise: the script had to plant punchlines, have suspense, and attract viewers, and the visuals had to be more varied.

I also called ElevenLabs to add a BGM track.

This BGM was written in 6 segments along the timeline, original dark style... quite something.

The final cut is as follows:

I think this result is pretty good.

With no manual edits at all, the narration script and BGM were both on point, clearly explaining the movie's plot, and overall fitting the suspense film atmosphere.

Although the visuals still look a bit like a PPT...

But after all, I didn't upload any materials, no reference images or anything, so the model could only make do with movie screenshots.

Even beginners can make short dramas now

What's even more amazing is that I discovered a "Story Bible" feature.

At first I didn't know what it was, but later I realized—isn't this just making short dramas??

Because of the Altman and Dario rap video, I casually named it "The Two Brothers of the US Stock IPO."

I clicked the "Story Bible" feature directly, and it created a new project request around this theme:

The older brother is an MD at the lead underwriting investment bank, the younger brother is the founder going public—one conference room, one issue price, one secret only the older brother knows.

I swear, Genspark started this first—all AI-generated.

It also found some scenes for reference first:

I gave some requirements for the character settings: four main characters, one like Tony Leung, one like Daniel Wu, one like Anne Hathaway, and one like Zhang Ziyi.

These four came out:

It might be a copyright issue—the characters don't look quite right, but it's okay. At a glance, everyone looks like a financial elite.

Next, it produced detailed character settings, costumes, props, and makeup test shots, with as many as 35 items requiring approval.

For a professional director or editor, they should be able to make more refined settings and adjustments.

But I'm a novice in this area, so I just clicked approve without thinking.

But there's a troublesome part here: some approval options are hard to find. There was one item I couldn't find no matter what, and I spent a full ten minutes searching for it.

And once a modification is involved, other already-approved content also gets sent back to pending approval status.

So I approved for ages, and finally managed to submit successfully...

If the conversation gets too long, it also prompts you that you can save it as a skill. Here I saved a financial/business-war short film skill.

After all the settings were saved, it split the original 5 scenes into 84 shots, then grouped them into 12 video units, wrote the prompts, and called Seedance 2.5 to generate the video clips.

For places in the video where the visuals broke continuity or didn't match the settings, it also did the rework itself, then prompted you to re-run the draft.

After everything finished running, it finally entered the editing, scoring, and subtitling stage, which was much faster than the earlier draft rework.

This is the final cut—"Pricing Night":

As someone who hasn't watched much short drama, I never imagined I'd make one too.

My first impression is that the visuals are quite exquisite, the facial expression details of the characters are very well done, and consistency of the main subjects before and after can also be guaranteed.

Without me asking at all, it even added Chinese-English bilingual subtitles on its own.

But there are drawbacks too—for example, in the latter half of the film, the audio and video gradually fall out of sync, and some frames still have skipped frames and bugs...

For example, why did the older brother's glasses suddenly disappear from his face, hey!

At this point I took a look: after running 3 cases, the initial 50,000 credits had cost about 45,000.

Bet you didn't expect this—it can make games too

With 5,000 credits left, I decided to make a Super Mario mini-game.

This one also came out in seconds, with no extra prompts needed at all—a pixel-style HTML game was written out.

It has all the basic mechanics: stomping enemies, growing bigger with mushrooms, hitting brick blocks, and flagpole level completion.

Due to copyright issues, the music and levels are all original assets drawn from scratch in code; only the gameplay mechanics are the same as the original.

This version focuses on being competent and playable—Nintendo would smile at it.

And it only consumed a few hundred credits.

If you think the game is too simple, Genspark will further prompt you on how to upgrade the mechanics:

But this is about as far as it goes, because my boss asked me why I keep playing games at work...

What is the Genspark Media Studio filmmaking workflow like?

After testing all this, what is the process orchestration behind Genspark Media Studio?

According to the official introduction, Media Studio uses a large language model as the "director's brain," responsible for understanding requirements, writing scripts, and then arranging different models to do the work.

It connects to more than 70 model versions from over 20 vendors, covering video, image, audio, and 3D. The Seedance and ElevenLabs used earlier are members of this division of labor.

The large language model also has to know in what order a film should be made.

So Genspark organized the production workflow into different skills. Making ads has its own approach, and making narrative short films has another set; the Agent selects based on the task type.

Take "Pricing Night" for example: first write the script, then split it into shots, then determine the characters and scenes, and only then does it come time for the video model to start filming.

The "Story Bible" is responsible for saving settings during this process. What the characters look like and what clothes they wear are turned in advance into fixed assets that can be referenced, and later shots follow them.

The same set of characters and worldview can also be carried over to other projects for continued use.

As for whether the filming has gone off track, that still needs to be checked. According to the official introduction, the Agent checks the generated results against the script and settings, and arranges reshoots for anything that doesn't pass. The proactive rework when making the short drama earlier corresponds to this step.

For example, for a 30-second video, if only seconds 7 to 13 are unsatisfactory, you click Reshoot clips, frame that segment on the timeline, and then tell it how to change it.

According to the official introduction, it preserves the original footage's motion and camera position, regenerates the video according to the new requirements, and you can specify up to 5 segments at a time, each at least 4 seconds.

If it's just that some object in the frame is an eyesore, you can use Frame edit to frame it and choose to replace or remove it, applicable to videos within 30 seconds.

This way, revision notes can land on a specific time segment or a specific object, without having to describe the entire film from scratch every time.

There's also a rather interesting detail: although they're all called video, knowledge explainer videos and narrative short films don't follow the same technical route.

Knowledge Explainer mainly uses code-to-video. Simply put, AI writes code to lay out text, charts, and animations, then renders them into video.

When explaining a set of data or demonstrating a knowledge point, you don't need to hand every second over to the video model to generate.

According to the official explanation, code rendering itself doesn't cost credits; narration and a small number of atmospheric shots are billed separately.

However, code can produce the visuals, but visual design still has to keep up, while narrative films need to control character actions and camera work in advance even more.

To this end, Media Studio also provides 3D white-model previews. Before formal generation, the Agent first builds the scene with white models to determine where the characters stand and where the camera shoots from. This step doesn't call the video model; after confirmation, it's handed to Seedance to produce the footage.

Add to that the 480p draft mentioned earlier, and the idea is easy to understand: catch problems as early as possible rather than waiting until the HD final cut is out to fix them, which reduces rework and saves a lot of credits.

After all, no one's money grows on trees...

Genspark has always emphasized: All in one, anyone can use it.

This is also my biggest takeaway from hands-on testing.

Media Studio puts these steps on the same canvas, keeps all intermediate products, and the Agent can continue working while users can also step in and modify at any time.

The several works shown above were all made by me entirely with Media Studio, without switching to any other platform.

According to the official introduction, professional users can also use the layer editor and multi-track video timeline inside it, and export PSD, Premiere projects, or CapCut drafts for further refinement.

For novices, the most direct change is that they don't have to learn all the tools first before they can start trying.

Just like "Pricing Night" earlier, I only casually came up with a name, and Media Studio directly expanded the story. Next, it asks, I choose, and wherever I'm not satisfied, I tell it in plain language.

How to split shots and how to write prompts—things I wasn't familiar with and found intimidating—can all be done first by the Agent.

In this process, people's work also changes accordingly. A lot of time can be used to judge whether the story is compelling enough, whether the music is right, and whether this shot should stay at all.

For someone who has never made a short drama, or even hasn't watched many, they can also easily realize their own ideas and make works of their own.

To this end, Media Studio also provides 3D white-model previews. Before formal generation, the Agent first builds the scene with white models to determine where the characters stand and where the camera shoots from. This step doesn't call the video model; after confirmation, it's handed to Seedance to produce the footage.

Add to that the 480p draft mentioned earlier, and the idea is easy to understand: catch problems as early as possible rather than waiting until the HD final cut is out to fix them, which reduces rework and saves a lot of credits.

After all, no one's money grows on trees...

Genspark has always emphasized: All in one, anyone can use it.

This is also my biggest takeaway from hands-on testing.

Media Studio puts these steps on the same canvas, keeps all intermediate products, and the Agent can continue working while users can also step in and modify at any time.

The several works shown above were all made by me entirely with Media Studio, without switching to any other platform.

According to the official introduction, professional users can also use the layer editor and multi-track video timeline inside it, and export PSD, Premiere projects, or CapCut drafts for further refinement.

For novices, the most direct change is that they don't have to learn all the tools first before they can start trying.

Just like "Pricing Night" earlier, I only casually came up with a name, and Media Studio directly expanded the story. Next, it asks, I choose, and wherever I'm not satisfied, I tell it in plain language.

How to split shots and how to write prompts—things I wasn't familiar with and found intimidating—can all be done first by the Agent.

In this process, people's work also changes accordingly. A lot of time can be used to judge whether the story is compelling enough, whether the music is right, and whether this shot should stay at all.

For someone who has never made a short drama, or even hasn't watched many, they can also easily realize their own ideas and make works of their own.