silentlooop/revt2v-ckpt

revt2v

LoRA students distilled from ali-vilab/text-to-video-ms-1.7b to generate video in reverse time from a text prompt, for the revT2V course project.

Methods: attn_rotation, conv_lora, attn_lora (checkpoints live under <method>/latest.pt; root latest.pt is attn_rotation, the baseline).

Use

from revt2v.infer import load_student_for_inference

pipeline, student = load_student_for_inference(
    "conv_lora",
    checkpoint_dir="results/checkpoints/conv_lora",
    hf_repo_id="silentlooop/revt2v-ckpt",
    hub_subfolder="conv_lora",
)

Trained on silentlooop/revt2v-data.