0.18.0 Stable Diffusion

To use sd1.4 you need to get an access token from https://huggingface.co/docs/hub/security-tokens

stable-diffusion-example.mp4

You need the right prompt and a lot of luck to get good results. Here are some of my test results

0.17.0

• [Bug fix] resize strategy will lose some edge mask area
• Add model from https://github.com/SHI-Labs/FcF-Inpainting

	Lama	FcF

0.16.0

Add MAT model

0.15.0

High-Resolution Strategy configuration

Each model has its own independent High-Resolution Strategy configuration.

Download Mask

Enable download inpainting mask when download result

New Model: ZITS

Original Image	ZITS	LaMa

Image is from ZITS paper. I didn't find a good example to show the advantages of ZITS and let me know if you have a good example. There can also be possible problems with my code, if you find them, please let me know too!

0.14.0

• Redo: Now you can Redo the inpainting result

• Gradually Inpainting(enabled by default): Perform inpainting on the last inpainting result, this makes it possible to further improve the inpainting result. It also helps to keep unmasked previous inpainting regions unchanged

• Start New change to Upload New: upload new file at editor page

PLMS sampler

latent-diffusion model add plms sampler. In general plms can get better results with fewer steps.

Original Image

ddim sampler steps=5

plms sampler steps=5

0.12.0

• bug fix: fix sizeLimit reset after inpainting.
• ldm model fp16 support, use less GPU memory. Benchmark on 3090

Image size	GPU memory
720x519	6147mb
1080x778	17499mb

Manual inpainting mode

• Manual inpainting mode: It is convenient to use when there are many target objects
• In manual mode you can undo strokes

lama-cleaner-0.10.0.mp4

0.9.3

• zoom bug fix
• editor tool panel CSS style fix

0.9.2

• Install by pip: pip install lama-cleaner
• Settings: now you can change model without restart the server