Awesome

MasaCtrl with T2I-Adapter

This repo contains the implementation of MasaCtrl integrated to controllable diffusion model T2I-Adapter.

Introduction

We propose MasaCtrl, a tuning-free method for non-rigid consistent image synthesis and editing. The key idea is to combine the contents from the source image and the layout synthesized from text prompt and additional controls into the desired synthesized or edited image, with Mutual Self-Attention Control.

Main Features

1 Controllable Consistent Image Synthesis and Editing

Directly modifying the text prompts often cannot generate target layout of desired image, thus we further integrate our method into existing proposed controllable diffusion pipelines (like T2I-Adapter and ControlNet) to obtain stable synthesis and editing results.

The target layout controlled by additional guidance.

<div align="center"> <img src="https://huggingface.co/TencentARC/MasaCtrl/resolve/main/assets/results_w_adapter.png"> <i>Synthesis (left part) and editing (right part) results with T2I-Adapter</i> </div>

2 Consistent Video Synthesis

With dense consistent guidance, MasaCtrl enables video synthesis

<div align="center"> <img src="https://huggingface.co/TencentARC/MasaCtrl/resolve/main/assets/results_w_adapter_consistent.png"> <i>Video Synthesis Results (with keypose and canny guidance)</i> </div>

Usage

Install

Please refer to usage guide of T2I-Adapter here (or from official repo) and download pretrained guidance models.

Checkpoints

Stable Diffusion:

You can download these checkpoints on their official repository and Hugging Face.

Personalized Models: You can download personlized models from CIVITAI or train your own customized models.

Start

For controllable synthesis:

python masactrl_w_adapter.py \
--which_cond sketch \
--cond_path_src SOURCE_CONDITION_PATH \
--cond_path CONDITION_PATH \
--cond_inp_type sketch \
--prompt_src "A bear walking in the forest" \
--prompt "A bear standing in the forest" \
--sd_ckpt models/sd-v1-4.ckpt \
--resize_short_edge 512 \
--cond_tau 1.0 \
--cond_weight 1.0 \
--n_samples 1 \
--adapter_ckpt models/t2iadapter_sketch_sd14v1.pth

NOTE: You can download the sketch examples here.

For real image editing:

python masactrl_w_adapter.py \
--src_img_path SOURCE_IMAGE_PATH \
--cond_path CONDITION_PATH \
--cond_inp_type image \
--prompt_src "" \
--prompt "a photo of a man wearing black t-shirt, giving a thumbs up" \
--sd_ckpt models/sd-v1-4.ckpt \
--resize_short_edge 512 \
--cond_tau 1.0 \
--cond_weight 1.0 \
--n_samples 1 \
--which_cond sketch \
--adapter_ckpt models/t2iadapter_sketch_sd14v1.pth \
--outdir ./workdir/masactrl_w_adapter_inversion/black-shirt

NOTE: You can download the real image editing example here.

Acknowledgements

We thank the awesome research works Prompt-to-Prompt, T2I-Adapter.

Contact

If your have any comments or questions, please open a new issue or feel free to contact Mingdeng Cao and Xintao Wang.