Lucy Edit Dev

Edit videos with a single sentence.

Lucy Edit Dev is the first open-source instruction-guided video editing model. It performs free-text edits — clothing changes, character swaps, object insertions, scene replacements — while perfectly preserving the motion and composition of your original video.

~5B params Wan2.2 5B architecture Diffusers · Safetensors Non-commercial license
Run it on Hugging Face View the model card

Watch it edit

Every clip below is an open-weight edit from Lucy Edit Dev. Hover to play; the original video is shown on the left with a single text prompt driving each result.

Character replacement

Input · man_walking.mp4

“Turn the man into an alien”

“Turn the man into a bear”

“Make it snowy”

Character takeover & wardrobe

Input · painter.mp4

“Put the woman in gothic black jeans and leather jacket and crop top under it.”

“Put her in a clown outfit.”

“Put the woman in a red bikini with an open thick coat above it.”

Identity-preserving character edits

Input · woman.mp4

“Turn the woman into Harley Quinn”

“Turn the woman into Lego”

“Turn the shirt into a sports jersey”

What it can edit

Best

Clothing & accessories

Swap outfits, add glasses, earrings or hats while preserving motion, pose and identity.

Change the shirt to a kimono with wide sleeves and patterned fabric.
Strong

Character replacement

Turn people into creatures, animals or known characters with detailed prompts.

Replace the person with a tiger, striped orange fur, muscular build, glowing green eyes.
Reliable

Object replacement

Structure-preserving swaps of objects of similar scale.

Replace the apple with a glowing crystal ball emitting blue light.
Mixed

Color changes

Works best with precise descriptions of gloss, texture and shade.

Change the jacket color to deep red leather with a glossy finish.
Mixed

Add objects

Best for wearable or handheld props; edits often attach to the subject.

Add a golden crown on the person's head, decorated with ornate jewels.
Global

Scene & style transforms

Environment or style-wide changes — may also alter the subject's identity.

Transform the sunny beach into a snowy tundra with falling snowflakes.

Prompting guidelines

Trigger words

  • Change — clothing or color modifications
  • Add — animals or objects
  • Replace — object substitution or subject swap
  • Transform to — global scene or style transformation

Tips from the model card

  • 20–30 descriptive words give the best results.
  • Describe style, appearance and context, not just the change.
  • 81-frame outputs have better temporal consistency.
  • No masks, no finetuning — pure free-text instructions.

Quickstart

Run the model locally with diffusers and a CUDA GPU.

# pip install git+https://github.com/huggingface/diffusers
from diffusers import AutoencoderKLWan, LucyEditPipeline
from diffusers.utils import export_to_video, load_video
import torch

model_id = "decart-ai/Lucy-Edit-Dev"
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipe = LucyEditPipeline.from_pretrained(model_id, vae=vae, torch_dtype=torch.bfloat16).to("cuda")

video = load_video("input.mp4")[:81]
output = pipe(
    prompt="Turn the man into a polar bear",
    video=video,
    height=480,
    width=832,
    num_frames=81,
    guidance_scale=5.0,
).frames[0]

export_to_video(output, "output.mp4", fps=24)

Put it to the test

Upload your own clip and prompt in the official Lucy Edit Dev Space on ZeroGPU.

Open the live demo