I've spent the past 3 years building and training diffusion and super resolution models, and I saw creators hit the same wall over and over: $6+ for a 10-second clip that takes 5-10 minutes to generate, and one bad take means starting all over. In essence, creators dump hours re-prompting for a few-second clip.
The real bottleneck isn't the model. It's the input modality.
Here's how dimension solve this: the agent builds a full pre-vis that replicates the exact object physics, rigging/animation, environment, camera paths - every detail locked in before generation ever starts. Think of it as a low-poly but still detailed version of your final shot with the actual choreography already baked in.
From there, the diffusion model isn't guessing at motion or composition anymore - it's simply responsible for what you'd expect out of post-production: photorealistic texturing and materials, subsurface scattering on skin, physically accurate lighting, etc...
Its so simple too:
- Log in
- Create a new canvas
- Enter the 3D stage
- Build your world with the agent
- Hit Finalize, bringing your pre-vis to the canvas
- Pass it to any diffusion model
Its completely free to try: https://dimension.so/ (Official launch tomorrow)
We are still early, and any feedback would be really helpful! If the problem and the promise resonate - I d love for you to try Dimension and tell me what you think. How are the outputs? Does it fit into your creative workflow? What is missing?
That feedback would help a lot! Let's kill the stigma around AI Video generation!
