Below we compare our results against Latent-Paint and Fantasia3D over the 3D stylization task. The methods emphasized with a (*) below do not modify the object's geometry, only its texture. For all methods, an uncolored object is provided as input (leftmost column) along with a target text prompt. Both Latent-Paint and Fantasia3D are optimization-based techniques, with lengthy inference times (over 15 minutes), while our method can generate shapes that conform to the target text prompt while preserving the structure of the guidance shape within several seconds.
| Guidance | Latent-Paint* | Fantasia3D-Paint* | Fantasia3D | Ours | |
|---|---|---|---|---|---|
A plush toy dumbell |
|
|
|
|
|
| | |||||
A clay vase |
|
|
|
|
|
| | |||||
A bottle of orange soda |
|
|
|
|
|
| | |||||
A rainbow colored fish |
|
|
|
|
|
| | |||||
A fighter pilot helmet |
|
|
|
|
|