Bing-su/adetailer
1
0
Fork 0
mirror of https://github.com/Bing-su/adetailer.git synced 2026-01-26 19:29:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'dev' into main

Browse Source
This commit is contained in:
Bingsu
2023-10-07 21:58:52 +09:00
parent 910bf3b970 e682eae794
commit d51f4b33cb
10 changed files with 120 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.pre-commit-config.yaml
View File

@@ -8,7 +8,7 @@ repos:
- id: mixed-line-ending
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: "v0.0.290"
rev: "v0.0.292"
hooks:
- id: ruff
args: [--fix, --exit-non-zero-on-fix]

7
CHANGELOG.md
View File

@@ -1,5 +1,12 @@
# Changelog
## 2023-10-07
- v23.10.0
- 허깅페이스 모델을 다운로드 실패했을 때, 계속 다운로드를 시도하지 않음
- img2img에서 img2img단계를 건너뛰는 기능 추가
- live preview에서 감지 단계를 보여줌 (PR #352)
## 2023-09-20
- v23.9.3

31
README.md
View File

@@ -1,6 +1,6 @@
# !After Detailer
# ADetailer
!After Detailer is a extension for stable diffusion webui, similar to Detection Detailer, except it uses ultralytics instead of the mmdet.
ADetailer is a extension for stable diffusion webui, similar to Detection Detailer, except it uses ultralytics instead of the mmdet.
## Install
@@ -22,15 +22,17 @@ You **DON'T** need to download any model from huggingface.
## Options
| Model, Prompts | | |
| --------------------------------- | ------------------------------------- | ------------------------------------------------- |
| ADetailer model | Determine what to detect. | `None` = disable |
| ADetailer prompt, negative prompt | Prompts and negative prompts to apply | If left blank, it will use the same as the input. |
| Model, Prompts | | |
| --------------------------------- | --------------------------------------------------------------------------------- | ------------------------------------------------- |
| ADetailer model | Determine what to detect. | `None` = disable |
| ADetailer prompt, negative prompt | Prompts and negative prompts to apply | If left blank, it will use the same as the input. |
| Skip img2img | Skip img2img. In practice, this works by changing the step count of img2img to 1. | img2img only |
| Detection | | |
| ------------------------------------ | -------------------------------------------------------------------------------------------- | --- |
| Detection model confidence threshold | Only objects with a detection model confidence above this threshold are used for inpainting. | |
| Mask min/max ratio | Only use masks whose area is between those ratios for the area of the entire image. | |
| Detection | | |
| ------------------------------------ | -------------------------------------------------------------------------------------------- | ------------ |
| Detection model confidence threshold | Only objects with a detection model confidence above this threshold are used for inpainting. | |
| Mask min/max ratio | Only use masks whose area is between those ratios for the area of the entire image. | |
| Mask only the top k largest | Only use the k objects with the largest area of the bbox. | 0 to disable |
If you want to exclude objects in the background, try setting the min ratio to around `0.01`.
@@ -86,9 +88,10 @@ Put your [ultralytics](https://github.com/ultralytics/ultralytics) yolo model in
It must be a bbox detection or segment model and use all label.
## Example
## How it works
![image](https://i.imgur.com/38RSxSO.png)
![image](https://i.imgur.com/2CYgjLx.png)
ADetailer works in three simple steps.
[![ko-fi](https://ko-fi.com/img/githubbutton_sm.svg)](https://ko-fi.com/F1F1L7V2N)
1. Create an image.
2. Detect object with a detection model and create a mask image.
3. Inpaint using the image from 1 and the mask from 2.

4
adetailer/__init__.py
View File

@@ -1,5 +1,5 @@
from .__version__ import __version__
from .args import AD_ENABLE, ALL_ARGS, ADetailerArgs, EnableChecker
from .args import ALL_ARGS, ADetailerArgs
from .common import PredictOutput, get_models
from .mediapipe import mediapipe_predict
from .ultralytics import ultralytics_predict
@@ -8,11 +8,9 @@ AFTER_DETAILER = "ADetailer"
__all__ = [
"__version__",
"AD_ENABLE",
"ADetailerArgs",
"AFTER_DETAILER",
"ALL_ARGS",
"EnableChecker",
"PredictOutput",
"get_models",
"mediapipe_predict",

2
adetailer/__version__.py
View File

@@ -1 +1 @@
__version__ = "23.9.3"
__version__ = "23.10.0"

17
adetailer/args.py
View File

@@ -2,7 +2,7 @@ from __future__ import annotations
from collections import UserList
from functools import cached_property, partial
from typing import Any, Literal, NamedTuple, Optional, Union
from typing import Any, Literal, NamedTuple, Optional
import pydantic
from pydantic import (
@@ -185,19 +185,7 @@ class ADetailerArgs(BaseModel, extra=Extra.forbid):
return p
class EnableChecker(BaseModel):
enable: bool
arg_list: list
def is_enabled(self) -> bool:
ad_model = ALL_ARGS[0].attr
if not self.enable:
return False
return any(arg.get(ad_model, "None") != "None" for arg in self.arg_list)
_all_args = [
("ad_enable", "ADetailer enable"),
("ad_model", "ADetailer model"),
("ad_prompt", "ADetailer prompt"),
("ad_negative_prompt", "ADetailer negative prompt"),
@@ -238,8 +226,7 @@ _all_args = [
("ad_controlnet_guidance_end", "ADetailer ControlNet guidance end"),
]
AD_ENABLE = Arg(*_all_args[0])
_args = [Arg(*args) for args in _all_args[1:]]
_args = [Arg(*args) for args in _all_args]
ALL_ARGS = ArgsList(_args)
BBOX_SORTBY = [

7
adetailer/common.py
View File

@@ -10,6 +10,7 @@ from PIL import Image, ImageDraw
from rich import print
repo_id = "Bingsu/adetailer"
_download_failed = False
@dataclass
@@ -20,12 +21,18 @@ class PredictOutput:
def hf_download(file: str):
global _download_failed
if _download_failed:
return "INVALID"
try:
path = hf_hub_download(repo_id, file)
except Exception:
msg = f"[-] ADetailer: Failed to load model {file!r} from huggingface"
print(msg)
path = "INVALID"
_download_failed = True
return path

21
adetailer/ui.py
View File

@@ -3,12 +3,12 @@ from __future__ import annotations
from dataclasses import dataclass
from functools import partial
from types import SimpleNamespace
from typing import Any, Callable
from typing import Any
import gradio as gr
from adetailer import AFTER_DETAILER, __version__
from adetailer.args import AD_ENABLE, ALL_ARGS, MASK_MERGE_INVERT
from adetailer.args import ALL_ARGS, MASK_MERGE_INVERT
from controlnet_ext import controlnet_exists, get_cn_models
cn_module_choices = [
@@ -91,13 +91,22 @@ def adui(
elem_id=eid("ad_enable"),
)
with gr.Column(scale=6):
ad_skip_img2img = gr.Checkbox(
label="Skip img2img",
value=False,
visible=is_img2img,
elem_id=eid("ad_skip_img2img"),
)
with gr.Column(scale=1, min_width=180):
gr.Markdown(
f"v{__version__}",
elem_id=eid("ad_version"),
)
infotext_fields.append((ad_enable, AD_ENABLE.name))
infotext_fields.append((ad_enable, "ADetailer enable"))
infotext_fields.append((ad_skip_img2img, "ADetailer skip img2img"))
with gr.Group(), gr.Tabs():
for n in range(num_models):
@@ -112,7 +121,7 @@ def adui(
infotext_fields.extend(infofields)
# components: [bool, dict, dict, ...]
components = [ad_enable, *states]
components = [ad_enable, ad_skip_img2img, *states]
return components, infotext_fields
@@ -421,7 +430,7 @@ def inpainting(w: Widgets, n: int, is_img2img: bool, webui_info: WebuiInfo):
with gr.Row():
with gr.Column(variant="compact"):
w.ad_use_checkpoint = gr.Checkbox(
label="Use separate checkpoint (experimental)" + suffix(n),
label="Use separate checkpoint" + suffix(n),
value=False,
visible=True,
elem_id=eid("ad_use_checkpoint"),
@@ -439,7 +448,7 @@ def inpainting(w: Widgets, n: int, is_img2img: bool, webui_info: WebuiInfo):
with gr.Column(variant="compact"):
w.ad_use_vae = gr.Checkbox(
label="Use separate VAE (experimental)" + suffix(n),
label="Use separate VAE" + suffix(n),
value=False,
visible=True,
elem_id=eid("ad_use_vae"),

2
install.py
View File

@@ -44,7 +44,7 @@ def run_pip(*args):
def install():
deps = [
# requirements
("ultralytics", "8.0.181", None),
("ultralytics", "8.0.194", None),
("mediapipe", "0.10.5", None),
("rich", "13.0.0", None),
# mediapipe

95
scripts/!adetailer.py
View File

@@ -5,7 +5,7 @@ import platform
import re
import sys
import traceback
from contextlib import contextmanager
from contextlib import contextmanager, suppress
from copy import copy, deepcopy
from functools import partial
from pathlib import Path
@@ -24,7 +24,7 @@ from adetailer import (
mediapipe_predict,
ultralytics_predict,
)
from adetailer.args import ALL_ARGS, BBOX_SORTBY, ADetailerArgs, EnableChecker
from adetailer.args import ALL_ARGS, BBOX_SORTBY, ADetailerArgs
from adetailer.common import PredictOutput
from adetailer.mask import (
filter_by_ratio,
@@ -176,7 +176,7 @@ class AfterDetailerScript(scripts.Script):
def is_ad_enabled(self, *args_) -> bool:
arg_list = [arg for arg in args_ if isinstance(arg, dict)]
if not args_ or not arg_list or not isinstance(args_[0], (bool, dict)):
if not args_ or not arg_list:
message = f"""
[-] ADetailer: Invalid arguments passed to ADetailer.
input: {args_!r}
@@ -184,9 +184,26 @@ class AfterDetailerScript(scripts.Script):
"""
print(dedent(message), file=sys.stderr)
return False
enable = args_[0] if isinstance(args_[0], bool) else True
checker = EnableChecker(enable=enable, arg_list=arg_list)
return checker.is_enabled()
ad_enabled = args_[0] if isinstance(args_[0], bool) else True
not_none = any(arg.get("ad_model", "None") != "None" for arg in arg_list)
return ad_enabled and not_none
def check_skip_img2img(self, p, *args_) -> None:
if (
hasattr(p, "_ad_skip_img2img")
or not hasattr(p, "init_images")
or not p.init_images
):
return
if len(args_) >= 2 and isinstance(args_[1], bool):
p._ad_skip_img2img = args_[1]
if args_[1]:
p._ad_orig_steps = p.steps
p.steps = 1
else:
p._ad_skip_img2img = False
def get_args(self, p, *args_) -> list[ADetailerArgs]:
"""
@@ -198,8 +215,8 @@ class AfterDetailerScript(scripts.Script):
message = f"[-] ADetailer: Invalid arguments passed to ADetailer: {args_!r}"
raise ValueError(message)
if hasattr(p, "adetailer_xyz"):
args[0] = {**args[0], **p.adetailer_xyz}
if hasattr(p, "_ad_xyz"):
args[0] = {**args[0], **p._ad_xyz}
all_inputs = []
@@ -306,7 +323,11 @@ class AfterDetailerScript(scripts.Script):
return width, height
def get_steps(self, p, args: ADetailerArgs) -> int:
return args.ad_steps if args.ad_use_steps else p.steps
if args.ad_use_steps:
return args.ad_steps
if hasattr(p, "_ad_orig_steps"):
return p._ad_orig_steps
return p.steps
def get_cfg_scale(self, p, args: ADetailerArgs) -> float:
return args.ad_cfg_scale if args.ad_use_cfg_scale else p.cfg_scale
@@ -345,9 +366,23 @@ class AfterDetailerScript(scripts.Script):
)
def write_params_txt(self, p) -> None:
i = p._ad_idx
lenp = len(p.all_prompts)
if i % lenp != lenp - 1:
return
prev = None
if hasattr(p, "_ad_orig_steps"):
prev = p.steps
p.steps = p._ad_orig_steps
infotext = self.infotext(p)
params_txt = Path(data_path, "params.txt")
params_txt.write_text(infotext, encoding="utf-8")
with suppress(Exception):
params_txt.write_text(infotext, encoding="utf-8")
if hasattr(p, "_ad_orig_steps"):
p.steps = prev
def script_filter(self, p, args: ADetailerArgs):
script_runner = copy(p.scripts)
@@ -525,16 +560,26 @@ class AfterDetailerScript(scripts.Script):
@staticmethod
def need_call_process(p) -> bool:
if p.scripts is None:
return False
i = p._ad_idx
bs = p.batch_size
return i % bs == bs - 1
@staticmethod
def need_call_postprocess(p) -> bool:
if p.scripts is None:
return False
i = p._ad_idx
bs = p.batch_size
return i % bs == 0
@staticmethod
def get_i2i_init_image(p, pp):
if getattr(p, "_ad_skip_img2img", False):
return p.init_images[0]
return pp.image
@rich_traceback
def process(self, p, *args_):
if getattr(p, "_ad_disabled", False):
@@ -542,8 +587,11 @@ class AfterDetailerScript(scripts.Script):
if self.is_ad_enabled(*args_):
arg_list = self.get_args(p, *args_)
self.check_skip_img2img(p, *args_)
extra_params = self.extra_params(arg_list)
p.extra_generation_params.update(extra_params)
else:
p._ad_disabled = True
def _postprocess_image_inner(
self, p, pp, args: ADetailerArgs, *, n: int = 0
@@ -560,6 +608,7 @@ class AfterDetailerScript(scripts.Script):
i = p._ad_idx
pp.image = self.get_i2i_init_image(p, pp)
i2i = self.get_i2i_p(p, args, pp.image)
seed, subseed = self.get_seed(p)
ad_prompts, ad_negatives = self.get_prompt(p, args)
@@ -579,6 +628,7 @@ class AfterDetailerScript(scripts.Script):
pred = predictor(ad_model, pp.image, args.ad_confidence, **kwargs)
masks = self.pred_preprocessing(pred, args)
shared.state.assign_current_image(pred.preview)
if not masks:
print(
@@ -633,17 +683,14 @@ class AfterDetailerScript(scripts.Script):
@rich_traceback
def postprocess_image(self, p, pp, *args_):
if getattr(p, "_ad_disabled", False):
return
if not self.is_ad_enabled(*args_):
if getattr(p, "_ad_disabled", False) or not self.is_ad_enabled(*args_):
return
p._ad_idx = getattr(p, "_ad_idx", -1) + 1
init_image = copy(pp.image)
arg_list = self.get_args(p, *args_)
if p.scripts is not None and self.need_call_postprocess(p):
if self.need_call_postprocess(p):
dummy = Processed(p, [], p.seed, "")
with preseve_prompts(p):
p.scripts.postprocess(copy(p), dummy)
@@ -655,22 +702,16 @@ class AfterDetailerScript(scripts.Script):
continue
is_processed |= self._postprocess_image_inner(p, pp, args, n=n)
if is_processed:
if is_processed and not getattr(p, "_ad_skip_img2img", False):
self.save_image(
p, init_image, condition="ad_save_images_before", suffix="-ad-before"
)
if p.scripts is not None and self.need_call_process(p):
if self.need_call_process(p):
with preseve_prompts(p):
p.scripts.process(copy(p))
try:
ia = p._ad_idx
lenp = len(p.all_prompts)
if ia % lenp == lenp - 1:
self.write_params_txt(p)
except Exception:
pass
self.write_params_txt(p)
def on_after_component(component, **_kwargs):
@@ -758,9 +799,9 @@ def make_axis_on_xyz_grid():
samplers = [sampler.name for sampler in all_samplers]
def set_value(p, x, xs, *, field: str):
if not hasattr(p, "adetailer_xyz"):
p.adetailer_xyz = {}
p.adetailer_xyz[field] = x
if not hasattr(p, "_ad_xyz"):
p._ad_xyz = {}
p._ad_xyz[field] = x
axis = [
xyz_grid.AxisOption(