DnDare
commented
3 weeks ago At this moment, 3d generatives models work in a similar way to StableDiffusion:
Stable diffusion: There are models as Stable123, StableVideo3d, AnimDiff. These models can generate different images from different angles, generated from an image input. Then these images can be used in photogrammetry software to build a mesh. This mesh is similar to a scanned mesh. We don’t know the maximum definition we could achieve.
Pros:This mesh could be used as a reference for speeding up the modeling phase. We don’t think there will be enough definition for a final mesh.
Cons: It generates from an image, we don’t know if we can use a trained model. If we can’t, there would not be a consistent/realistic model. The photogrammetry process can generate artifacts.
We don’t know: Finding images to train the model might be a problem OR we can train with different images from the internet. How long it would take to train a model.
—------- Other public sites for 3d generation: https://lumalabs.ai/genie?view=preview https://www.meshy.ai/ https://www.tripo3d.ai/app
We need to find a way to optimize the vehicle creation process.
1) Research new processes and techniques to reuse parts that we have already modeled. Ensure that one modeled piece can be used to create several variations.
2) Identify methods to speed up the modeling creation process.