vantagewithai / Lucy-Edit-1.1-Dev-GGUF

Lucy Edit 1.1 Dev (5B)

Quantized GGUF Version

Original Model Link: https://huggingface.co/decart-ai/Lucy-Edit-1.1-Dev

Watch us at Youtube: @VantageWithAI

🧪 GitHub |  📖 Playground |  📑 Technical Paper |  💬 Discord

---

Lucy Edit 1.1 Dev is an open-weight video editing model that performs instruction-guided edits on videos using free-text prompts — it supports a variety of edits, such as clothing & accessory changes , character changes , object insertions , and scene replacements while preserving the motion and composition perfectly.

• 🚀 First open-source instruction-guided video editing model
• 🧩 Built on Wan2.2 5B architecture — inherits high-compression VAE + DiT stack, making adapting existing scripts and workflows easy.
• 🏃‍♂️ Motion Preservation - preserves the motion and composition of videos perfectly, allowing precise edits.
• 🎯 Edit reliability — edits are more robust when compared to common inference time methods.
• 🧢 Wardrobe & accessories — change outfits, add glasses/earrings/hats/etc.
• 🧌 Character Changes — replace characters with monsters, animals and known characters. (e.g., "Replace the person with a polar bear")
• 🗺️ Scenery swap — move the scene (e.g., "transform the scene into a 2D cartoon,")
• 📝 Pure text instructions — no finetuning, no masks required for common edits
• 🌍 Global edits (background & style) — perform scene-wide changes while preserving subjects.
• 🫧 Seamless blending — cleaner boundaries and more coherent lighting across edits.
• 📦 General objects — insert non-wearable objects with improved spatial grounding.

ℹ️ Model size: ~5B params . Build on top of Wan2.2 5B .

✨ What’s new in 1.1

• Better quality overall — higher fidelity details, stronger prompt adherence, improved identity preservation.
• Global edits (background & style) — more reliable scene-wide edits with better subject preservation and control.
• Better blending / seamless edits — cleaner boundaries, fewer artifacts, improved color and lighting consistency across frames.
• Addition of general objects — more reliable insertion of wearable, handheld, and general scene objects with improved spatial grounding, occlusion, and shadows.

---

🎬 Demos

Note: The prompts above are not enriched, the model will react better to enriched prompts - as described in the prompt guideline section below.

---

🔥 Latest News

• [2025-09-18] : Initial Lucy Edit 1.1 Dev weights & reference code released.
• [2025-09-16] : Diffusers integration PR opened and merged. PR #12340 .

---

🛠️ Quickstart

Installation

pip install git+https://github.com/huggingface/diffusers

Inference

Please refer to the "Prompting Guidelines & Supported Edits" section for the best experience.

from typing import List

import torch
from PIL import Image

from diffusers import AutoencoderKLWan, LucyEditPipeline
from diffusers.utils import export_to_video, load_video


# Arguments
url = "https://d2drjpuinn46lb.cloudfront.net/painter_original_edit.mp4"
prompt = "Change the apron and blouse to a classic clown costume: satin polka-dot jumpsuit in bright primary colors, ruffled white collar, oversized pom-pom buttons, white gloves, oversized red shoes, red foam nose; soft window light from left, eye-level medium shot, natural folds and fabric highlights."
negative_prompt = ""
num_frames = 81
height = 480
width = 832

# Load video
def convert_video(video: List[Image.Image]) -> List[Image.Image]:
    video = load_video(url)[:num_frames]
    video = [video[i].resize((width, height)) for i in range(num_frames)]
    return video

video = load_video(url, convert_method=convert_video)

# Load model
model_id = "decart-ai/Lucy-Edit-1.1-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)
pipe.to("cuda")

# Generate video
output = pipe(
    prompt=prompt,
    video=video,
    negative_prompt=negative_prompt,
    height=480,
    width=832,
    num_frames=81,
    guidance_scale=5.0
).frames[0]

# Export video
export_to_video(output, "output.mp4", fps=24)

---

Prompting Guidelines & Supported Edits

Lucy Edit 1.1 is built for precise, realistic, and identity-preserving video edits.
Prompts with ~20–30 descriptive words work best. Using the right trigger words helps the model understand your intent.

Trigger Words

• Change → Clothing or color modifications
• Add → Adding animals or objects
• Replace → Object substitution or subject swap
• Transform to → Global scene or style transformations

Supported Edit Types

1. Clothing Changes

✅ Best performance. Lucy Edit excels at swapping outfits while preserving motion, pose, and identity, with improved fabric detail and texture fidelity in 1.1.
Example : “Change the shirt to a kimono with wide sleeves and patterned fabric.”

2. Human/Character Replacement

✅ Strong results. Works well for transforming people into new characters or creatures. Detailed prompts are key.
Example : “Replace the person with a tiger, striped orange fur, muscular build, and glowing green eyes.”
Example : “Replace the person with an 2D anime character, big eyes, blue gown and battle scars.”

3. Replace Objects

✅ More reliable with improved seam blending and material consistency. Ideal when replacing one object with another of similar scale.
Example : “Replace the apple with a glowing crystal ball emitting blue light.”

4. Color Changes

⚠️ Mixed reliability. Sometimes subtle, sometimes exaggerated. Works best with precise descriptions.
Example : “Change the jacket color to deep red leather with a glossy finish.”

5. Add Objects

✅ Significantly improved. Supports adding general objects to the scene with better spatial grounding, occlusion handling, and lighting consistency. Great for wearable/handheld props and background objects.
Example : “Add a golden crown on the person’s head, decorated with ornate jewels.”
Example : “Add a red umbrella on the table casting a soft shadow.”

6. Global Transformations

✅ Much improved for background and style changes. Preserves subject identity better while applying scene-wide edits; extreme styles may still slightly alter identity. Example : “Transform the sunny beach into a snowy tundra with falling snowflakes.”

Additional Notes

• Strengths: Lucy Edit excels at identity conservation, edit precision, realism, and prompt adherence.
• Detail matters: Longer prompts (20–30 words) describing style, appearance, and context improve results.
• Frame count: 81-frame generations produce better temporal consistency than shorter clips.
• Blending: 1.1 improves boundary blending and temporal coherence across edits.

---

📦 Integrations

• ☁️ Hosted API: You can access the model on our API and get 5000 free credits here .
• 🧨 Diffusers: Coming soon
• 🧩 ComfyUI: Coming soon

🧭 Roadmap

• ✅ Public Batch API.
• ✅ Diffusers pipeline ( LucyEditPipeline )
• ✅ Remote ComfyUI custom nodes.
• ✅ Technical Report
• Local Inference ComfyUI Nodes.
• LoRA and fine-tuning scripts.

---

📣 Citation

@article{decart2025lucyedit,
  title   = {Lucy Edit: Open-Weight Text-Guided Video Editing},
  author  = {DecartAI Team},
  year    = {2025}
  url     = { https://d2drjpuinn46lb.cloudfront.net/Lucy_Edit__High_Fidelity_Text_Guided_Video_Editing.pdf}
 }

---

🙏 Acknowledgements

Lucy Edit Dev builds on the excellent foundations of Wan2.2 (5B), and thanks the broader open-source community including diffusers and Hugging Face .

---

📬 Contact

• GitHub Issues: DecartAI/lucy-edit .
• Discord: Join our discord server, here .