sirrineprogramming/rochelle-wan-lora-simple

Rochelle — Wan 2.1 T2V LoRA Simple (1.3B, diffusers format)

A LoRA fine-tune of the Wan 2.1 1.3B text-to-video model in native diffusers format. Trained using the diffusers training script (checkpoint-style output layout).

Files

File	Description
lora_weights/000500/pytorch_lora_weights.safetensors	Checkpoint @ 500 steps
lora_weights/001000/pytorch_lora_weights.safetensors	✅ Checkpoint @ 1000 steps (use this)

Architecture

• Base model: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
• Hidden dim: 1536 (1.3B architecture confirmed)
• LoRA rank: 64
• Key format: Diffusers (transformer.blocks.N.attn1.*)
• Total keys: 612

Usage

from diffusers import WanPipeline
import torch

pipe = WanPipeline.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers",
    torch_dtype=torch.bfloat16
)
pipe.load_lora_weights(
    "sirrineprogramming/rochelle-wan-lora-simple",
    weight_name="lora_weights/001000/pytorch_lora_weights.safetensors"
)
pipe.to("cuda")

frames = pipe(
    prompt="Rochelle, a beautiful woman with dark hair...",
    num_frames=81,
    height=480,
    width=832,
).frames[0]

Training Details

• Base model: Wan-AI/Wan2.1-T2V-1.3B-Diffusers
• Trigger word: Rochelle
• Checkpoints: 500, 1000 steps