Skip to content

Latest commit

 

History

History
226 lines (149 loc) · 9.68 KB

README.md

File metadata and controls

226 lines (149 loc) · 9.68 KB

Replacer

Replacer is an extension for AUTOMATIC1111/stable-diffusion-webui. The goal of this extension is to automate objects masking by detection prompt, using sd-webui-segment-anything, and img2img inpainting in one easy to use tab. It also useful for batch inpaint, and inpaint in video with stable diffusion and controlnet

If you find this project useful, please star it on GitHub! ⭐️

You also can draw your mask instead of or in addition to detection, and take advantage of convenient HiresFix option, and ControlNet inpainting with preserving original image resolution and aspect ratio

Installation

  1. Install sd-webui-segment-anything extension. If it bothers you, you can hide it in the Replacer's settings. Go to tab Extension -> Avaliable -> click Load from and search "sd-webui-segment-anything"
  2. Download model sam_hq_vit_l.pth, or others from the list bellow, and put it into extensions/sd-webui-segment-anything/models/sam
  3. For faster hires fix, download lcm-lora-sdv1-5, rename it into lcm-lora-sdv1-5.safetensors, put into models/Lora. Or if you have already lcm lora, then change hires suffix in the extension options
  4. Install this extension. Go to tab Extension -> Available -> click Load from and search "Replacer". Be sure your sd-webui version is >= 1.5.0. For AMD and Intel GPUs you need to enable cpu in Replacer's settings
  5. Reload UI

SAM models list:

SAM-HQ are the best for me. Choose it depending on your vram. Sum this model size with dino model size (694MB-938MB)

  1. SAM from Meta AI.

  2. SAM-HQ from SysCV.

  3. MobileSAM from Kyung Hee University.

FastSAM and Matting-Anything aren't currently supported

AMD Radeon and Intel ARC

For AMD and Intel GPUs, and maby something else, you need to enable CPU for detection in Replacer's settings. Go to Settings -> Replacer and enable it

Then click "Apply settings" and restart webui (do not press reload ui, restart it completly)

Supported extensons:

  1. ControlNet extension

  2. My other extension "Lama cleaner as masked content"

  3. Inpaint Difference

  4. Fork sd-webui-forge

Usage

General

You just need to upload your image, enter 3 prompts, and click "Run". You can override prompts examples in Settings with your commonly using prompts. Don't forget to select inpaint checkpoint

Be sure you are using inpainting model

By default if a prompt is empty, it uses first prompt from examples. You can disable this behavior in settings for positive and negative prompts. Detection prompt can not be empty

You can detect few objects, just using comma ,

HiresFix

Default settings are designed for using lcm lora for fast upscale. It requires lcm lora I mentioned, cfg scale 1.0 and sampling steps 4. There is no difference in quality for my opinion

Despite in txt2img for lcm lora DPM++ samplers produces awful results, while hires fix it produces a way better result. So I recommend "Use the same sampler" option

Note: hires fix is designed for single-user server

Advanced options

I do not recommend change these options, if you don't know what you do

Video inpainting

You can use Replacer to inpaint video with a regular stable diffusion inpaint method. It is very inconsistent, but in a few cases it can produce good enough results

Example:

size: 768x512
detection prompt: hands
positive prompt: photo of dog
negative prompt: clothes, cartoon, painting, illustration, (worst quality, low quality, normal quality:2)
denoising: 0.3
masked content: lama cleaner
fps: 10
postprocessed with RIFE

Before:

dog_before.mp4

After:

dog_after.mp4

To increase consistency, it's better to inpaint clear objects on video with good quality and enough. Your prompts need to produce consistent results.

To suppress flickering you can generate in little fps (e.g. 10), then interpolate (x2) it with ai interpolation algorithm (e.g RIFE or frame interpolation in deforum sd-webui extension)

You can also use sd-webui-controlnet or lama-cleaner with (low denosing) extensions to increase consistency, if it fits to your scenario

Dedicated page

Dedicated page (replacer tab only) is available on url /replacer-dedicated

API

API is avaliable on /replacer/replace

    input_image: str = "base64 image"
    detection_prompt: str = ""
    avoidance_prompt: str = ""
    positive_prompt: str = ""
    negative_prompt: str = ""
    width: int = 512
    height: int = 512
    sam_model_name: str = sam_model_list[0] if sam_model_list else ""
    dino_model_name: str = dino_model_list[0]
    seed: int = -1
    sampler: str = "DPM++ 2M SDE Karras"
    steps: int = 20
    box_threshold: float = 0.3
    mask_expand: int = 35
    mask_blur: int = 4
    max_resolution_on_detection = 1280
    cfg_scale: float = 5.5
    denoise: int = 1
    inpaint_padding = 40
    inpainting_mask_invert: bool = False
    upscaler_for_img2img : str = ""
    fix_steps : bool = False
    inpainting_fill : int = 0
    sd_model_checkpoint : str = ""
    scripts : dict = {} # ControlNet and Soft Inpainting. See apiExample.py for example

Avaliable options on /replacer/avaliable_options

http://127.0.0.1:7860/docs#/default/api_replacer_replace_replacer_replace_post

See an example of usage in apiExample.py script

Extention name

Replacer" name of this extension, you can provide it inside ExtensionName.txt in root of extension directory.

Or you can override it using the environment variable SD_WEBUI_REPLACER_EXTENTION_NAME

For exaple: Linux

export SD_WEBUI_REPLACER_EXTENTION_NAME="Fast Inpaint"

Or Windows in your .bat file:

set SD_WEBUI_REPLACER_EXTENTION_NAME="Fast Inpaint"

How it works?

First, grounding dino models detect objects you provided in the detection prompt. Then segment anything model generates contours of them. And then extension chooses randomly 1 of 3 generated masks, and inpaints it with regular inpainting method in a1111 webui

When you press the "Apply hires fix" button, the extension regenerates the image with exactly the same settings, excluding upscaler_for_img2img. Then it applies inpainting with "Original" masked content mode and lower denoising but higher resolution.

Useful Tips!

How to change default values of advanced options and hires fix options?

You need to reload the web page, then set your desirable settings. Then go to the "Defaults" section in the "Settings" tab. Click "View changes", check is it ok, then click "Apply" and "Reload UI"

How to get an inpainting model?

I recommend you to using EpicPhotoGasm - Z - Inpainting model for realism. If you've already have your favorite model, but it doesn't have inpainting model, you can make it in "Checkpoint Merger" tab:

  1. Select your target model as "model B"
  2. Select sd-v1-5-inpainting as "model A"
  3. Select sd_v1-5-pruned-emaonly as "model C"
  4. Set Custom Name the same as your target model name (.inpainting suffix will be added automatically)
  5. Set Multiplier (M) to 1.0
  6. Select Interpolation Method to "add difference", and "Save as float16"
  7. Merge

How to change maximum values of sliders?

In file ui-config.json in root of webui you can edit maximum and minimum values of sliders

Need to do:

  • ☑️ cache mask
  • ☑️ batch processing
  • ☑️ "apply hires fix" button
  • ☑️ additional options
  • ☑️ progress bar + interrupt
  • option for pass into hires fix automatically
  • ☑️ control net
  • tiled vae
  • ☑️ "hide segment anything extention" option
  • txt2img script
  • allow multiply instances

small todo:

  • copy images while appling hires fix