SeedVR2 text embedding β safetensors conversion
An unofficial conversion of SeedVR2's fixed positive text embedding from a PyTorch pickle to safetensors. SeedVR2 takes no prompt: it is conditioned on this precomputed embedding. The values are ByteDance's; nothing was recomputed or modified.
- Official models: ByteDance-Seed/SeedVR2-3B, ByteDance-Seed/SeedVR2-7B
- Official code: ByteDance-Seed/SeedVR
For anything about the model itself β usage, results, citation β see the official pages above.
File
| File | Converted from | Contents |
|---|---|---|
text.safetensors |
pos_emb.pt |
one tensor, positive, BF16, shape [58, 5120] |
The negative embedding (neg_emb.pt) is not included.
Conversion
import torch
from safetensors.torch import save_file
positive = torch.load("pos_emb.pt", map_location="cpu")
save_file({"positive": positive}, "text.safetensors")
License
Apache 2.0, as the original release. The only change is the file format.
Inference Providers NEW
This model isn't deployed by any Inference Provider. π Ask for provider support
Model tree for drift-generator/seedvr2-text-embeddings
Base model
ByteDance-Seed/SeedVR2-3B