Free4D: Tuning-free 4D Scene Generation
with Spatial-Temporal Consistency

Tianqi Liu1,2*   Zihao Huang1,2*   Zhaoxi Chen2   Guangcong Wang3   Shoukang Hu2  
Liao Shen1,2   Huiqiang Sun1,2   Zhiguo Cao1   Wei Li2\(\dagger\)   Ziwei Liu2\(\dagger\)

1Huazhong University of Science and Technology     2Nanyang Technological University     3Great Bay University
*Equal Contribution     \(\dagger\)Corresponding Authors

arXiv

YouTube

Code

Cite

TL;DR

Free4D is a tuning-free framework for 4D scene generation from a single image or text.



4D Interactive Viewer



Demo

fn fn fn fn fn fn fn fn

Overview Video



Abstract

We present Free4D, a novel tuning-free framework for 4D scene generation from a single image. Existing methods either focus on object-level generation, making scene-level generation infeasible, or rely on large-scale multi-view video datasets for expensive training, with limited generalization ability due to the scarcity of 4D scene data. In contrast, our key insight is to distill pre-trained foundation models for consistent 4D scene representation, which offers promising advantages such as efficiency and generalizability. 1) To achieve this, we first animate the input image using image-to-video diffusion models followed by 4D geometric structure initialization. 2) To turn this coarse structure into spatial-temporal consistent multiview videos, we design an adaptive guidance mechanism with a point-guided denoising strategy for spatial consistency and a novel latent replacement strategy for temporal coherence. 3) To lift these generated observations into consistent 4D representation, we propose a modulation-based refinement to mitigate inconsistencies while fully leveraging the generated information. The resulting 4D representation enables real-time, controllable spatial-temporal rendering, marking a significant advancement in single-image-based 4D scene generation.



Method

  • Given an input image or text prompt, we first generate a dynamic video \( \mathcal{V} = \{I(t,1)\}_{t=1}^{T} \) using an off-the-shelf video generation model.
  • Then, we employ MonST3R with a progressive static point cloud aggregation strategy for dynamic reconstruction, obtaining a 4D geometric structure.
  • Next, guided by this structure, we render a coarse multi-view video \( \mathcal{S}^{\prime} = \{\{I^{\prime}(t,k)\}_{t=1}^{T}\}_{k=1}^{K} \) along a predefined camera trajectory and refine it into \( \mathcal{S} = \{\{I(t,k)\}_{t=1}^{T}\}_{k=1}^{K} \) using ViewCrafter. To ensure spatial-temporal consistency, we introduce Adaptive Classifier-Free Guidance (CFG) and Point Cloud Guided Denoising for spatial coherence, along with Reference Latent Replacement for temporal coherence.
  • Finally, we propose an efficient training strategy with a Modulation-Based Refinement to lift the generated multi-view video \( \mathcal{S} \) into a consistent 4D representation \( \mathcal{R} \).



Text-to-4D Comparisons


4Real (Left) vs. Ours (Right)

    
"A building on fire." "A fox playing videogame." "A cat singing." "A firepit."
"A baby rabbit is eating ice-cream." "A goat drinking beer." "A monkey eating a candy bar."

Dream-in-4D (Left) vs. Ours (Right)

    
"A cat singing." "A kangaroo cooking a meal." "A monkey eating a candy bar."
"A rabbit eating carrot." "A baby rabbit is eating ice-cream." "A monster reading a book."

4Dfy (Left) vs. Ours (Right)

    
"A fox playing videogame." "A building on fire." "A lemur holding and drinking boba."
"A monkey eating a candy bar." "A baby rabbit is eating ice-cream." "A baby panda reading a book."


Image-to-4D Comparisons


GenXD* (Left) vs. Ours (Right)

* denotes the video rendered by 4DGS, which is trained using the videos sourced from GenXD's project page.
    
fn fn

DimensionX (Left) vs. Ours (Right)

    
fn fn

Animate124 (Left) vs. Ours (Right)

    
fn fn


More Results

fn fn fn fn fn fn
fn fn fn fn fn fn


Citation

@InProceedings{Liu_2025_ICCV,
  author    = {Liu, Tianqi and Huang, Zihao and Chen, Zhaoxi and Wang, Guangcong and Hu, Shoukang and Shen, Liao and Sun, Huiqiang and Cao, Zhiguo and Li, Wei and Liu, Ziwei},
  title     = {Free4D: Tuning-free 4D Scene Generation with Spatial-Temporal Consistency},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month     = {October},
  year      = {2025},
  pages     = {25571-25582}
}


Related Works

  • 4Real: Towards Photorealistic 4D Scene Generation via Video Diffusion Models
  • Dream-in-4D: A Unified Approach for Text- and Image-guided 4D Scene Generation
  • 4D-fy: Text-to-4D Generation Using Hybrid Score Distillation Sampling
  • GenXD: Generating Any 3D and 4D Scenes
  • DimensionX: Create Any 3D and 4D Scenes from a Single Image with Controllable Video Diffusion
  • CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models
  • DreamGaussian4D: Generative 4D Gaussian Splatting
  • L4GM: Large 4D Gaussian Reconstruction Model