finetrainers/Wan2.1-T2V-1.3B-3dgs-v0

Prompt

3DGS_DISSOLVE A lone sailor grips the helm as towering waves crash against the boat. Lightning splits the sky, illuminating the chaotic sea. The storm intensifies, but then, as if merging with the elements, the sailor's figure slowly transforms into a cascade of red sparks, scattering into the wind.

Prompt

3DGS_DISSOLVE A snowboarder carves through fresh powder, the mountain landscape unfolding in crisp whites and deep blues. They launch into the air, spinning weightlessly before their form dissipates into a dazzling explosion of glowing embers that swirl into the icy breeze.

Prompt

3DGS_DISSOLVE A stark white bowl floats in darkness, its form slowly twisting. It begins to crack apart, glowing embers spilling outward, until it evaporates into a breathtaking eruption of red sparks. The scene transitions—now, a bouquet of flowers stands in its place, untouched, frozen in time.

Prompt

3DGS_DISSOLVE A surfer carves through the curling wave, sunlight glinting off the water's surface. The tunnel of shimmering liquid twists around them, and just as they reach the crest, their form flickers—evaporating into a dazzling explosion of fiery sparks that vanish into the ocean spray.

This is a LoRA fine-tune of the Wan-AI/Wan2.1-T2V-1.3B-Diffusers model on the finetrainers/3dgs-dissolve dataset.

Code: https://github.com/a-r-r-o-w/finetrainers

This is an experimental checkpoint and its poor generalization is well-known.

Inference code:

import torch
from diffusers import WanPipeline
from diffusers.utils import export_to_video

pipe = WanPipeline.from_pretrained(
    "Wan-AI/Wan2.1-T2V-1.3B-Diffusers", torch_dtype=torch.bfloat16
).to("cuda")
pipe.load_lora_weights("finetrainers/Wan2.1-T2V-1.3B-3dgs-v0", adapter_name="wan-lora")
pipe.set_adapters(["wan-lora"], [0.5])

video = pipe("<my-awesome-prompt>").frames[0]
export_to_video(video, "output.mp4", fps=24)

Training logs are available on WandB here.