Krea unveils Seedance Studio with camera path and 3D scene controls

The wall most people hit when using video generation AI isn't image quality — it's control. To get the result to reflect where the camera starts, where it moves, and where the people and background are positioned, users have had to describe it all in words. Seedance Studio, released by Krea on August 19, moves that process onto on-screen controls instead.

What was confirmed in the announcement

In a post on its official X account, Krea introduced Seedance Studio as "everything you need to create with Seedance 2.5, in one place." Three components were named specifically: prompt presets, camera paths, and 3D scene controls.

The post came with an introduction video, and in the same post Krea asked users what additional features they'd like to see. Comments came in within hours of the announcement, with one user mentioning model training features and saying they were using Krea as their primary platform.

Component	Name used in the announcement	What it handles
Prompt presets	prompt presets	Loading saved instructions instead of writing them fresh each time
Camera paths	camera paths	The path along which the camera moves
3D scene controls	3D scene controls	Controls for manipulating a scene in three dimensions

The post did not specify exactly how each control operates on screen or how many detailed parameters each has. What's confirmed is that all three axes were combined into a single tool.

What Seedance is, and where Krea fits in

It's worth untangling the naming first, since it's easy to confuse. Seedance is a video generation model developed by ByteDance Seed, the AI research organization of ByteDance, the company behind TikTok. This model typically doesn't appear as a standalone app — it shows up as a "selectable model" inside other production tools.

Krea does two things at once. Founded in San Francisco in 2022, it built its name with a tool for generating images and video in real time, and it trains its own image and video models in-house. At the same time, it aggregates dozens of models made by others — Google's Veo, OpenAI's Sora, Kuaishou's Kling — into a single screen where users can pick and choose. Seedance Studio falls into the latter category: the model originates from ByteDance, while Krea built the workbench for using it.

Work made with Seedance has appeared in local coverage several times recently. The 90s Hong Kong-style AI short film "Red Flag," released by Higgsfield on YouTube on August 19, was also produced by running Seedance on the Higgsfield platform.

Why camera paths get their own category

What shapes the output of a video generation model breaks down into two broad factors: what appears in the frame, and how it's viewed. The first is relatively easy to describe in words — a character's outfit, the time of day, the texture of the background. The second is hard. A camera move that circles from behind a subject to the front, an angle that grazes past a wall, a changing distance to the subject — the moment these get put into a sentence, there's enormous room for interpretation. That's why the same prompt entered twice can produce very different camera movements.

The fact that camera paths and 3D scene controls appeared as separate items in the announcement reads as a signal that Krea is treating this second axis as something to handle through on-screen manipulation. Prompt presets, by contrast, are closer to the first axis — saving and recalling frequently-used style instructions.

How to use it

Where to start — Seedance Studio is a newly added tool inside the Krea service. The announcement post included a link to the tool, though the exact menu location within the app wasn't specified in this announcement. The starting point is logging into a Krea account and finding the Seedance Studio entry in the video tools list.

Workflow — Based on the three components confirmed in the announcement, the process breaks down as follows:

1. Open Seedance Studio and confirm the model in use is Seedance 2.5.
2. Choose a prompt preset close to the desired tone or style — a place where you don't have to write a sentence from scratch each time.
3. Set up the composition within the scene using 3D scene controls.
4. Specify the path the camera will travel using camera paths.
5. Run the generation, check the result, and go back to adjust only the specific controls that didn't match.

What you can try — For a product introduction video, presets could lock in lighting and texture tone while varying only the camera path, producing the same product from multiple angles. For a short narrative video, instead of writing a new prompt for each scene, a single preset could be shared across scenes with camera movement used to differentiate cuts. This is generally where time gets saved on repetitive work.

That's the full scope disclosed so far. Krea is taking requests for additional features — anyone with a specific request can leave it as a comment on the relevant post on Krea's official X account.

Editor's view

The competitive axis in video generation tools is shifting from the model to the control layer. A year or two ago, most news in this space was about output quality — how many seconds of video could be generated, whether fingers came out mangled. That's no longer the case. Multiple companies have reached comparable quality levels, and what now separates them is how many attempts it takes to get the shot you want. Seedance Studio bringing together prompt presets, camera paths, and 3D scene controls on one screen sits squarely within that trend.

Krea's position within this structure is notable. Because it both builds models and distributes other companies' models, it doesn't lose out no matter which model wins — as long as it holds the workbench. Model performance can flip within months, but working habits don't change that quickly. A user who has built up a library of presets and gotten used to a camera-path workflow will keep coming back to the same screen even as the underlying model changes. This division of labor — ByteDance Seed building the model, Krea building the tool — mirrors an industry structure where ingredient suppliers and kitchen-equipment makers split apart, now taking hold in video AI as well.

For teams in Korea, there are two practical points worth considering. First, anyone evaluating video AI adoption should look at the control layer before benchmark scores. In practice, cost leaks not from rendering fees but from how many times a shot has to be redone because it didn't come out right. Whether presets can be shared across a team, and whether camera movements can be saved in a reproducible form, determine that number. Second, there's typically a gap between when tools like this are announced and when they actually stabilize. Early versions often have controls that exist but don't reliably reflect what's requested. Rather than committing straight to a commercial project, it's safer to run internal tests on a few pieces first and confirm reproducibility before adopting it.

Expect a string of similar announcements in the coming weeks. The very fact that Krea asked users for feature requests signals that this category still lacks a standard control layer. Timeline editing, character consistency locking, and reference-image-based scene composition are likely to top the next wish list. Where video AI tools ultimately converge is already fairly clear: the shape of editing software.