TokenDial drops continuous video attribute sliders

TokenDial moves video generation from "lucky dip" prompts to parametric control, addressing a massive pain point for professional creators.

• –Additive token injection avoids the degradation and compute costs of full model fine-tuning
• –Spatiotemporal consistency ensures that "dialing" an attribute doesn't break temporal coherence or identity
• –Compositional sliders allow for multi-attribute editing without semantic interference between controls
• –The training-free approach makes it highly portable across different diffusion transformer architectures
• –Bridges the gap between raw generative output and traditional non-linear video editing tools