2022-10-29 05:11:16 +08:00
|
|
|
from __future__ import annotations
|
2022-10-10 09:26:52 +08:00
|
|
|
import math
|
2022-09-20 06:13:12 +08:00
|
|
|
import os
|
2023-01-22 15:17:12 +08:00
|
|
|
import re
|
2022-11-27 21:28:32 +08:00
|
|
|
import sys
|
|
|
|
import traceback
|
2023-01-11 14:10:07 +08:00
|
|
|
import shutil
|
2022-09-20 06:13:12 +08:00
|
|
|
|
2022-09-11 16:31:16 +08:00
|
|
|
import numpy as np
|
|
|
|
from PIL import Image
|
|
|
|
|
2022-09-26 07:22:12 +08:00
|
|
|
import torch
|
2022-09-27 15:44:00 +08:00
|
|
|
import tqdm
|
2022-09-26 07:22:12 +08:00
|
|
|
|
2022-10-29 05:11:16 +08:00
|
|
|
from typing import Callable, List, OrderedDict, Tuple
|
2022-10-29 02:24:11 +08:00
|
|
|
from functools import partial
|
|
|
|
from dataclasses import dataclass
|
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
from modules import processing, shared, images, devices, sd_models, sd_samplers, sd_vae
|
2022-09-11 16:31:16 +08:00
|
|
|
from modules.shared import opts
|
|
|
|
import modules.gfpgan_model
|
|
|
|
from modules.ui import plaintext_to_html
|
|
|
|
import modules.codeformer_model
|
2022-09-29 05:59:44 +08:00
|
|
|
import gradio as gr
|
2022-11-27 20:51:29 +08:00
|
|
|
import safetensors.torch
|
2022-09-14 00:23:55 +08:00
|
|
|
|
2022-10-29 05:11:16 +08:00
|
|
|
class LruCache(OrderedDict):
|
|
|
|
@dataclass(frozen=True)
|
|
|
|
class Key:
|
|
|
|
image_hash: int
|
|
|
|
info_hash: int
|
|
|
|
args_hash: int
|
2022-10-29 03:30:04 +08:00
|
|
|
|
2022-10-29 05:11:16 +08:00
|
|
|
@dataclass
|
|
|
|
class Value:
|
|
|
|
image: Image.Image
|
|
|
|
info: str
|
|
|
|
|
2022-10-29 05:36:25 +08:00
|
|
|
def __init__(self, max_size: int = 5, *args, **kwargs):
|
2022-10-29 05:11:16 +08:00
|
|
|
super().__init__(*args, **kwargs)
|
|
|
|
self._max_size = max_size
|
|
|
|
|
|
|
|
def get(self, key: LruCache.Key) -> LruCache.Value:
|
|
|
|
ret = super().get(key)
|
|
|
|
if ret is not None:
|
2022-10-29 05:36:25 +08:00
|
|
|
self.move_to_end(key) # Move to end of eviction list
|
2022-10-29 05:11:16 +08:00
|
|
|
return ret
|
|
|
|
|
|
|
|
def put(self, key: LruCache.Key, value: LruCache.Value) -> None:
|
|
|
|
self[key] = value
|
|
|
|
while len(self) > self._max_size:
|
|
|
|
self.popitem(last=False)
|
2022-10-29 03:30:04 +08:00
|
|
|
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 05:36:25 +08:00
|
|
|
cached_images: LruCache = LruCache(max_size=5)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 05:36:25 +08:00
|
|
|
|
2022-12-24 19:04:53 +08:00
|
|
|
def run_extras(extras_mode, resize_mode, image, image_folder, input_dir, output_dir, show_extras_results, gfpgan_visibility, codeformer_visibility, codeformer_weight, upscaling_resize, upscaling_resize_w, upscaling_resize_h, upscaling_crop, extras_upscaler_1, extras_upscaler_2, extras_upscaler_2_visibility, upscale_first: bool, save_output: bool = True):
|
2022-09-12 04:24:24 +08:00
|
|
|
devices.torch_gc()
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2023-01-03 23:34:51 +08:00
|
|
|
shared.state.begin()
|
|
|
|
shared.state.job = 'extras'
|
|
|
|
|
2022-09-16 11:23:37 +08:00
|
|
|
imageArr = []
|
2022-09-20 06:13:12 +08:00
|
|
|
# Also keep track of original file names
|
|
|
|
imageNameArr = []
|
2022-10-16 12:50:55 +08:00
|
|
|
outputs = []
|
2022-12-04 14:13:36 +08:00
|
|
|
|
2022-09-22 17:11:48 +08:00
|
|
|
if extras_mode == 1:
|
2022-09-16 11:23:37 +08:00
|
|
|
#convert file to pillow image
|
|
|
|
for img in image_folder:
|
2022-10-09 21:14:56 +08:00
|
|
|
image = Image.open(img)
|
2022-09-16 11:23:37 +08:00
|
|
|
imageArr.append(image)
|
2022-09-20 06:13:12 +08:00
|
|
|
imageNameArr.append(os.path.splitext(img.orig_name)[0])
|
2022-10-16 12:50:55 +08:00
|
|
|
elif extras_mode == 2:
|
2022-10-16 14:51:01 +08:00
|
|
|
assert not shared.cmd_opts.hide_ui_dir_config, '--hide-ui-dir-config option must be disabled'
|
|
|
|
|
2022-10-16 12:50:55 +08:00
|
|
|
if input_dir == '':
|
|
|
|
return outputs, "Please select an input directory.", ''
|
2022-10-29 13:11:03 +08:00
|
|
|
image_list = shared.listfiles(input_dir)
|
2022-10-16 12:50:55 +08:00
|
|
|
for img in image_list:
|
2022-10-20 20:31:29 +08:00
|
|
|
try:
|
|
|
|
image = Image.open(img)
|
|
|
|
except Exception:
|
|
|
|
continue
|
2022-10-16 12:50:55 +08:00
|
|
|
imageArr.append(image)
|
|
|
|
imageNameArr.append(img)
|
2022-09-22 17:11:48 +08:00
|
|
|
else:
|
|
|
|
imageArr.append(image)
|
|
|
|
imageNameArr.append(None)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-16 12:50:55 +08:00
|
|
|
if extras_mode == 2 and output_dir != '':
|
|
|
|
outpath = output_dir
|
|
|
|
else:
|
|
|
|
outpath = opts.outdir_samples or opts.outdir_extras_samples
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
# Extra operation definitions
|
2022-10-29 05:36:25 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
def run_gfpgan(image: Image.Image, info: str) -> Tuple[Image.Image, str]:
|
2023-01-03 23:34:51 +08:00
|
|
|
shared.state.job = 'extras-gfpgan'
|
2022-10-29 02:24:11 +08:00
|
|
|
restored_img = modules.gfpgan_model.gfpgan_fix_faces(np.array(image, dtype=np.uint8))
|
|
|
|
res = Image.fromarray(restored_img)
|
|
|
|
|
|
|
|
if gfpgan_visibility < 1.0:
|
|
|
|
res = Image.blend(image, res, gfpgan_visibility)
|
|
|
|
|
|
|
|
info += f"GFPGAN visibility:{round(gfpgan_visibility, 2)}\n"
|
|
|
|
return (res, info)
|
2022-09-16 11:23:37 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
def run_codeformer(image: Image.Image, info: str) -> Tuple[Image.Image, str]:
|
2023-01-03 23:34:51 +08:00
|
|
|
shared.state.job = 'extras-codeformer'
|
2022-10-29 02:24:11 +08:00
|
|
|
restored_img = modules.codeformer_model.codeformer.restore(np.array(image, dtype=np.uint8), w=codeformer_weight)
|
|
|
|
res = Image.fromarray(restored_img)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
if codeformer_visibility < 1.0:
|
|
|
|
res = Image.blend(image, res, codeformer_visibility)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
info += f"CodeFormer w: {round(codeformer_weight, 2)}, CodeFormer visibility:{round(codeformer_visibility, 2)}\n"
|
|
|
|
return (res, info)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
def upscale(image, scaler_index, resize, mode, resize_w, resize_h, crop):
|
2023-01-03 23:34:51 +08:00
|
|
|
shared.state.job = 'extras-upscale'
|
2022-10-29 03:30:04 +08:00
|
|
|
upscaler = shared.sd_upscalers[scaler_index]
|
|
|
|
res = upscaler.scaler.upscale(image, resize, upscaler.data_path)
|
|
|
|
if mode == 1 and crop:
|
|
|
|
cropped = Image.new("RGB", (resize_w, resize_h))
|
|
|
|
cropped.paste(res, box=(resize_w // 2 - res.width // 2, resize_h // 2 - res.height // 2))
|
|
|
|
res = cropped
|
|
|
|
return res
|
2022-10-29 02:24:11 +08:00
|
|
|
|
|
|
|
def run_prepare_crop(image: Image.Image, info: str) -> Tuple[Image.Image, str]:
|
|
|
|
# Actual crop happens in run_upscalers_blend, this just sets upscaling_resize and adds info text
|
|
|
|
nonlocal upscaling_resize
|
2022-10-10 09:26:52 +08:00
|
|
|
if resize_mode == 1:
|
|
|
|
upscaling_resize = max(upscaling_resize_w/image.width, upscaling_resize_h/image.height)
|
|
|
|
crop_info = " (crop)" if upscaling_crop else ""
|
|
|
|
info += f"Resize to: {upscaling_resize_w:g}x{upscaling_resize_h:g}{crop_info}\n"
|
2022-10-29 02:24:11 +08:00
|
|
|
return (image, info)
|
|
|
|
|
|
|
|
@dataclass
|
|
|
|
class UpscaleParams:
|
|
|
|
upscaler_idx: int
|
|
|
|
blend_alpha: float
|
|
|
|
|
2022-10-29 05:36:25 +08:00
|
|
|
def run_upscalers_blend(params: List[UpscaleParams], image: Image.Image, info: str) -> Tuple[Image.Image, str]:
|
2022-10-29 02:24:11 +08:00
|
|
|
blended_result: Image.Image = None
|
2022-11-03 10:37:43 +08:00
|
|
|
image_hash: str = hash(np.array(image.getdata()).tobytes())
|
2022-10-29 02:24:11 +08:00
|
|
|
for upscaler in params:
|
2022-10-29 05:36:25 +08:00
|
|
|
upscale_args = (upscaler.upscaler_idx, upscaling_resize, resize_mode,
|
|
|
|
upscaling_resize_w, upscaling_resize_h, upscaling_crop)
|
2022-11-03 10:37:43 +08:00
|
|
|
cache_key = LruCache.Key(image_hash=image_hash,
|
2022-10-29 05:36:25 +08:00
|
|
|
info_hash=hash(info),
|
2022-11-03 10:37:43 +08:00
|
|
|
args_hash=hash(upscale_args))
|
2022-10-29 03:30:04 +08:00
|
|
|
cached_entry = cached_images.get(cache_key)
|
|
|
|
if cached_entry is None:
|
|
|
|
res = upscale(image, *upscale_args)
|
|
|
|
info += f"Upscale: {round(upscaling_resize, 3)}, visibility: {upscaler.blend_alpha}, model:{shared.sd_upscalers[upscaler.upscaler_idx].name}\n"
|
2022-10-29 05:11:16 +08:00
|
|
|
cached_images.put(cache_key, LruCache.Value(image=res, info=info))
|
2022-10-29 03:30:04 +08:00
|
|
|
else:
|
|
|
|
res, info = cached_entry.image, cached_entry.info
|
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
if blended_result is None:
|
|
|
|
blended_result = res
|
|
|
|
else:
|
|
|
|
blended_result = Image.blend(blended_result, res, upscaler.blend_alpha)
|
|
|
|
return (blended_result, info)
|
|
|
|
|
|
|
|
# Build a list of operations to run
|
|
|
|
facefix_ops: List[Callable] = []
|
2022-10-29 05:11:16 +08:00
|
|
|
facefix_ops += [run_gfpgan] if gfpgan_visibility > 0 else []
|
|
|
|
facefix_ops += [run_codeformer] if codeformer_visibility > 0 else []
|
2022-10-29 02:24:11 +08:00
|
|
|
|
|
|
|
upscale_ops: List[Callable] = []
|
2022-10-29 05:11:16 +08:00
|
|
|
upscale_ops += [run_prepare_crop] if resize_mode == 1 else []
|
2022-10-29 02:24:11 +08:00
|
|
|
|
|
|
|
if upscaling_resize != 0:
|
|
|
|
step_params: List[UpscaleParams] = []
|
2022-10-29 05:36:25 +08:00
|
|
|
step_params.append(UpscaleParams(upscaler_idx=extras_upscaler_1, blend_alpha=1.0))
|
2022-10-29 02:24:11 +08:00
|
|
|
if extras_upscaler_2 != 0 and extras_upscaler_2_visibility > 0:
|
2022-10-29 05:36:25 +08:00
|
|
|
step_params.append(UpscaleParams(upscaler_idx=extras_upscaler_2, blend_alpha=extras_upscaler_2_visibility))
|
2022-10-29 02:24:11 +08:00
|
|
|
|
2022-10-29 05:36:25 +08:00
|
|
|
upscale_ops.append(partial(run_upscalers_blend, step_params))
|
2022-10-29 02:24:11 +08:00
|
|
|
|
2022-10-29 05:11:16 +08:00
|
|
|
extras_ops: List[Callable] = (upscale_ops + facefix_ops) if upscale_first else (facefix_ops + upscale_ops)
|
2022-10-29 02:24:11 +08:00
|
|
|
|
|
|
|
for image, image_name in zip(imageArr, imageNameArr):
|
|
|
|
if image is None:
|
|
|
|
return outputs, "Please select an input image.", ''
|
2023-01-03 23:34:51 +08:00
|
|
|
|
|
|
|
shared.state.textinfo = f'Processing image {image_name}'
|
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
existing_pnginfo = image.info or {}
|
2022-10-10 09:26:52 +08:00
|
|
|
|
2022-10-29 02:24:11 +08:00
|
|
|
image = image.convert("RGB")
|
|
|
|
info = ""
|
|
|
|
# Run each operation on each image
|
|
|
|
for op in extras_ops:
|
|
|
|
image, info = op(image, info)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-12-15 02:59:33 +08:00
|
|
|
if opts.use_original_name_batch and image_name is not None:
|
2022-10-20 18:10:32 +08:00
|
|
|
basename = os.path.splitext(os.path.basename(image_name))[0]
|
|
|
|
else:
|
2022-10-20 20:43:27 +08:00
|
|
|
basename = ''
|
2022-10-20 18:10:32 +08:00
|
|
|
|
2023-01-03 23:34:51 +08:00
|
|
|
if opts.enable_pnginfo: # append info before save
|
|
|
|
image.info = existing_pnginfo
|
|
|
|
image.info["extras"] = info
|
|
|
|
|
2022-12-24 19:04:53 +08:00
|
|
|
if save_output:
|
|
|
|
# Add upscaler name as a suffix.
|
|
|
|
suffix = f"-{shared.sd_upscalers[extras_upscaler_1].name}" if shared.opts.use_upscaler_name_as_suffix else ""
|
|
|
|
# Add second upscaler if applicable.
|
|
|
|
if suffix and extras_upscaler_2 and extras_upscaler_2_visibility:
|
|
|
|
suffix += f"-{shared.sd_upscalers[extras_upscaler_2].name}"
|
2022-12-17 20:31:03 +08:00
|
|
|
|
2022-12-24 19:04:53 +08:00
|
|
|
images.save_image(image, path=outpath, basename=basename, seed=None, prompt=None, extension=opts.samples_format, info=info, short_filename=True,
|
|
|
|
no_prompt=True, grid=False, pnginfo_section_name="extras", existing_info=existing_pnginfo, forced_filename=None, suffix=suffix)
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-16 12:50:55 +08:00
|
|
|
if extras_mode != 2 or show_extras_results :
|
|
|
|
outputs.append(image)
|
2022-09-16 17:43:24 +08:00
|
|
|
|
2022-09-29 09:14:13 +08:00
|
|
|
devices.torch_gc()
|
|
|
|
|
2022-09-16 17:43:24 +08:00
|
|
|
return outputs, plaintext_to_html(info), ''
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-10-29 05:11:16 +08:00
|
|
|
def clear_cache():
|
|
|
|
cached_images.clear()
|
|
|
|
|
2022-09-11 16:31:16 +08:00
|
|
|
|
2022-09-17 14:07:07 +08:00
|
|
|
def run_pnginfo(image):
|
2022-09-20 01:18:16 +08:00
|
|
|
if image is None:
|
|
|
|
return '', '', ''
|
|
|
|
|
2022-11-27 21:28:32 +08:00
|
|
|
geninfo, items = images.read_info_from_image(image)
|
|
|
|
items = {**{'parameters': geninfo}, **items}
|
2022-11-24 10:39:09 +08:00
|
|
|
|
2022-09-11 16:31:16 +08:00
|
|
|
info = ''
|
2022-09-14 00:23:55 +08:00
|
|
|
for key, text in items.items():
|
2022-09-11 16:31:16 +08:00
|
|
|
info += f"""
|
|
|
|
<div>
|
|
|
|
<p><b>{plaintext_to_html(str(key))}</b></p>
|
|
|
|
<p>{plaintext_to_html(str(text))}</p>
|
|
|
|
</div>
|
|
|
|
""".strip()+"\n"
|
|
|
|
|
|
|
|
if len(info) == 0:
|
|
|
|
message = "Nothing found in the image."
|
|
|
|
info = f"<div><p>{message}<p></div>"
|
|
|
|
|
2022-09-24 03:49:21 +08:00
|
|
|
return '', geninfo, info
|
2022-09-26 07:22:12 +08:00
|
|
|
|
|
|
|
|
2023-01-11 14:10:07 +08:00
|
|
|
def create_config(ckpt_result, config_source, a, b, c):
|
|
|
|
def config(x):
|
2023-01-19 15:39:51 +08:00
|
|
|
res = sd_models.find_checkpoint_config(x) if x else None
|
|
|
|
return res if res != shared.sd_default_config else None
|
2023-01-11 14:10:07 +08:00
|
|
|
|
|
|
|
if config_source == 0:
|
|
|
|
cfg = config(a) or config(b) or config(c)
|
|
|
|
elif config_source == 1:
|
|
|
|
cfg = config(b)
|
|
|
|
elif config_source == 2:
|
|
|
|
cfg = config(c)
|
|
|
|
else:
|
|
|
|
cfg = None
|
|
|
|
|
|
|
|
if cfg is None:
|
|
|
|
return
|
|
|
|
|
|
|
|
filename, _ = os.path.splitext(ckpt_result)
|
|
|
|
checkpoint_filename = filename + ".yaml"
|
|
|
|
|
|
|
|
print("Copying config:")
|
|
|
|
print(" from:", cfg)
|
|
|
|
print(" to:", checkpoint_filename)
|
|
|
|
shutil.copyfile(cfg, checkpoint_filename)
|
|
|
|
|
|
|
|
|
2023-01-19 23:24:17 +08:00
|
|
|
checkpoint_dict_skip_on_merge = ["cond_stage_model.transformer.text_model.embeddings.position_ids"]
|
2023-01-19 15:39:51 +08:00
|
|
|
|
|
|
|
|
2023-01-19 17:12:09 +08:00
|
|
|
def to_half(tensor, enable):
|
|
|
|
if enable and tensor.dtype == torch.float:
|
|
|
|
return tensor.half()
|
|
|
|
|
|
|
|
return tensor
|
|
|
|
|
|
|
|
|
2023-01-22 15:17:12 +08:00
|
|
|
def run_modelmerger(id_task, primary_model_name, secondary_model_name, tertiary_model_name, interp_method, multiplier, save_as_half, custom_name, checkpoint_format, config_source, bake_in_vae, discard_weights):
|
2023-01-03 23:21:51 +08:00
|
|
|
shared.state.begin()
|
|
|
|
shared.state.job = 'model-merge'
|
|
|
|
|
2023-01-19 13:53:50 +08:00
|
|
|
def fail(message):
|
|
|
|
shared.state.textinfo = message
|
|
|
|
shared.state.end()
|
2023-01-19 14:25:37 +08:00
|
|
|
return [*[gr.update() for _ in range(4)], message]
|
2023-01-19 13:53:50 +08:00
|
|
|
|
2022-10-17 06:44:39 +08:00
|
|
|
def weighted_sum(theta0, theta1, alpha):
|
2022-09-26 22:50:21 +08:00
|
|
|
return ((1 - alpha) * theta0) + (alpha * theta1)
|
|
|
|
|
2022-10-17 06:44:39 +08:00
|
|
|
def get_difference(theta1, theta2):
|
|
|
|
return theta1 - theta2
|
|
|
|
|
|
|
|
def add_difference(theta0, theta1_2_diff, alpha):
|
|
|
|
return theta0 + (alpha * theta1_2_diff)
|
2022-10-14 14:05:06 +08:00
|
|
|
|
2023-01-19 23:24:17 +08:00
|
|
|
def filename_weighted_sum():
|
2023-01-19 15:39:51 +08:00
|
|
|
a = primary_model_info.model_name
|
|
|
|
b = secondary_model_info.model_name
|
|
|
|
Ma = round(1 - multiplier, 2)
|
|
|
|
Mb = round(multiplier, 2)
|
|
|
|
|
|
|
|
return f"{Ma}({a}) + {Mb}({b})"
|
|
|
|
|
2023-01-19 23:24:17 +08:00
|
|
|
def filename_add_difference():
|
2023-01-19 15:39:51 +08:00
|
|
|
a = primary_model_info.model_name
|
|
|
|
b = secondary_model_info.model_name
|
|
|
|
c = tertiary_model_info.model_name
|
|
|
|
M = round(multiplier, 2)
|
|
|
|
|
|
|
|
return f"{a} + {M}({b} - {c})"
|
|
|
|
|
|
|
|
def filename_nothing():
|
|
|
|
return primary_model_info.model_name
|
|
|
|
|
|
|
|
theta_funcs = {
|
2023-01-19 23:24:17 +08:00
|
|
|
"Weighted sum": (filename_weighted_sum, None, weighted_sum),
|
|
|
|
"Add difference": (filename_add_difference, get_difference, add_difference),
|
2023-01-19 15:39:51 +08:00
|
|
|
"No interpolation": (filename_nothing, None, None),
|
|
|
|
}
|
|
|
|
filename_generator, theta_func1, theta_func2 = theta_funcs[interp_method]
|
|
|
|
shared.state.job_count = (1 if theta_func1 else 0) + (1 if theta_func2 else 0)
|
|
|
|
|
2023-01-19 08:13:15 +08:00
|
|
|
if not primary_model_name:
|
2023-01-19 13:53:50 +08:00
|
|
|
return fail("Failed: Merging requires a primary model.")
|
2023-01-19 08:13:15 +08:00
|
|
|
|
2022-09-29 05:59:44 +08:00
|
|
|
primary_model_info = sd_models.checkpoints_list[primary_model_name]
|
2023-01-19 08:13:15 +08:00
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
if theta_func2 and not secondary_model_name:
|
2023-01-19 13:53:50 +08:00
|
|
|
return fail("Failed: Merging requires a secondary model.")
|
2022-09-27 15:44:00 +08:00
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
secondary_model_info = sd_models.checkpoints_list[secondary_model_name] if theta_func2 else None
|
2022-09-27 15:44:00 +08:00
|
|
|
|
2023-01-19 08:13:15 +08:00
|
|
|
if theta_func1 and not tertiary_model_name:
|
2023-01-19 13:53:50 +08:00
|
|
|
return fail(f"Failed: Interpolation method ({interp_method}) requires a tertiary model.")
|
2023-01-19 15:39:51 +08:00
|
|
|
|
2023-01-19 10:21:52 +08:00
|
|
|
tertiary_model_info = sd_models.checkpoints_list[tertiary_model_name] if theta_func1 else None
|
2023-01-19 08:13:15 +08:00
|
|
|
|
|
|
|
result_is_inpainting_model = False
|
2022-12-04 14:13:36 +08:00
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
if theta_func2:
|
|
|
|
shared.state.textinfo = f"Loading B"
|
|
|
|
print(f"Loading {secondary_model_info.filename}...")
|
|
|
|
theta_1 = sd_models.read_state_dict(secondary_model_info.filename, map_location='cpu')
|
|
|
|
else:
|
|
|
|
theta_1 = None
|
2022-10-14 14:05:06 +08:00
|
|
|
|
2022-10-17 06:44:39 +08:00
|
|
|
if theta_func1:
|
2023-01-19 15:39:51 +08:00
|
|
|
shared.state.textinfo = f"Loading C"
|
2022-12-04 14:13:36 +08:00
|
|
|
print(f"Loading {tertiary_model_info.filename}...")
|
|
|
|
theta_2 = sd_models.read_state_dict(tertiary_model_info.filename, map_location='cpu')
|
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
shared.state.textinfo = 'Merging B and C'
|
2023-01-19 14:25:37 +08:00
|
|
|
shared.state.sampling_steps = len(theta_1.keys())
|
2022-10-17 06:44:39 +08:00
|
|
|
for key in tqdm.tqdm(theta_1.keys()):
|
2023-01-19 23:24:17 +08:00
|
|
|
if key in checkpoint_dict_skip_on_merge:
|
2023-01-19 15:39:51 +08:00
|
|
|
continue
|
|
|
|
|
2022-10-17 06:44:39 +08:00
|
|
|
if 'model' in key:
|
2022-10-18 20:33:24 +08:00
|
|
|
if key in theta_2:
|
|
|
|
t2 = theta_2.get(key, torch.zeros_like(theta_1[key]))
|
|
|
|
theta_1[key] = theta_func1(theta_1[key], t2)
|
|
|
|
else:
|
2022-10-18 21:05:52 +08:00
|
|
|
theta_1[key] = torch.zeros_like(theta_1[key])
|
2023-01-19 14:25:37 +08:00
|
|
|
|
|
|
|
shared.state.sampling_step += 1
|
2022-12-04 14:13:36 +08:00
|
|
|
del theta_2
|
|
|
|
|
2023-01-19 14:25:37 +08:00
|
|
|
shared.state.nextjob()
|
|
|
|
|
2023-01-03 23:21:51 +08:00
|
|
|
shared.state.textinfo = f"Loading {primary_model_info.filename}..."
|
2022-12-04 14:13:36 +08:00
|
|
|
print(f"Loading {primary_model_info.filename}...")
|
|
|
|
theta_0 = sd_models.read_state_dict(primary_model_info.filename, map_location='cpu')
|
|
|
|
|
|
|
|
print("Merging...")
|
2023-01-19 15:39:51 +08:00
|
|
|
shared.state.textinfo = 'Merging A and B'
|
2023-01-19 14:25:37 +08:00
|
|
|
shared.state.sampling_steps = len(theta_0.keys())
|
2022-09-27 15:44:00 +08:00
|
|
|
for key in tqdm.tqdm(theta_0.keys()):
|
2023-01-19 15:39:51 +08:00
|
|
|
if theta_1 and 'model' in key and key in theta_1:
|
2023-01-14 19:00:00 +08:00
|
|
|
|
2023-01-19 23:24:17 +08:00
|
|
|
if key in checkpoint_dict_skip_on_merge:
|
2023-01-14 19:00:00 +08:00
|
|
|
continue
|
|
|
|
|
2022-12-04 17:30:44 +08:00
|
|
|
a = theta_0[key]
|
|
|
|
b = theta_1[key]
|
2022-10-15 02:20:28 +08:00
|
|
|
|
2022-12-04 17:30:44 +08:00
|
|
|
# this enables merging an inpainting model (A) with another one (B);
|
|
|
|
# where normal model would have 4 channels, for latenst space, inpainting model would
|
|
|
|
# have another 4 channels for unmasked picture's latent space, plus one channel for mask, for a total of 9
|
|
|
|
if a.shape != b.shape and a.shape[0:1] + a.shape[2:] == b.shape[0:1] + b.shape[2:]:
|
|
|
|
if a.shape[1] == 4 and b.shape[1] == 9:
|
|
|
|
raise RuntimeError("When merging inpainting model with a normal one, A must be the inpainting model.")
|
|
|
|
|
|
|
|
assert a.shape[1] == 9 and b.shape[1] == 4, f"Bad dimensions for merged layer {key}: A={a.shape}, B={b.shape}"
|
|
|
|
|
|
|
|
theta_0[key][:, 0:4, :, :] = theta_func2(a[:, 0:4, :, :], b, multiplier)
|
|
|
|
result_is_inpainting_model = True
|
|
|
|
else:
|
|
|
|
theta_0[key] = theta_func2(a, b, multiplier)
|
2022-10-15 02:20:28 +08:00
|
|
|
|
2023-01-19 17:12:09 +08:00
|
|
|
theta_0[key] = to_half(theta_0[key], save_as_half)
|
2022-10-10 09:26:52 +08:00
|
|
|
|
2023-01-19 14:25:37 +08:00
|
|
|
shared.state.sampling_step += 1
|
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
del theta_1
|
|
|
|
|
|
|
|
bake_in_vae_filename = sd_vae.vae_dict.get(bake_in_vae, None)
|
|
|
|
if bake_in_vae_filename is not None:
|
|
|
|
print(f"Baking in VAE from {bake_in_vae_filename}")
|
|
|
|
shared.state.textinfo = 'Baking in VAE'
|
|
|
|
vae_dict = sd_vae.load_vae_dict(bake_in_vae_filename, map_location='cpu')
|
2023-01-14 19:00:00 +08:00
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
for key in vae_dict.keys():
|
|
|
|
theta_0_key = 'first_stage_model.' + key
|
|
|
|
if theta_0_key in theta_0:
|
2023-01-19 17:12:09 +08:00
|
|
|
theta_0[theta_0_key] = to_half(vae_dict[key], save_as_half)
|
2023-01-14 19:00:00 +08:00
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
del vae_dict
|
2022-09-27 15:44:00 +08:00
|
|
|
|
2023-01-19 17:12:09 +08:00
|
|
|
if save_as_half and not theta_func2:
|
|
|
|
for key in theta_0.keys():
|
|
|
|
theta_0[key] = to_half(theta_0[key], save_as_half)
|
|
|
|
|
2023-01-22 15:17:12 +08:00
|
|
|
if discard_weights:
|
|
|
|
regex = re.compile(discard_weights)
|
|
|
|
for key in list(theta_0):
|
|
|
|
if re.search(regex, key):
|
|
|
|
theta_0.pop(key, None)
|
|
|
|
|
2022-10-01 03:57:25 +08:00
|
|
|
ckpt_dir = shared.cmd_opts.ckpt_dir or sd_models.model_path
|
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
filename = filename_generator() if custom_name == '' else custom_name
|
|
|
|
filename += ".inpainting" if result_is_inpainting_model else ""
|
|
|
|
filename += "." + checkpoint_format
|
2022-12-04 17:30:44 +08:00
|
|
|
|
2022-10-01 03:57:25 +08:00
|
|
|
output_modelname = os.path.join(ckpt_dir, filename)
|
2022-09-29 05:21:54 +08:00
|
|
|
|
2023-01-19 14:25:37 +08:00
|
|
|
shared.state.nextjob()
|
2023-01-19 15:39:51 +08:00
|
|
|
shared.state.textinfo = "Saving"
|
2022-09-27 15:44:00 +08:00
|
|
|
print(f"Saving to {output_modelname}...")
|
2022-11-27 20:51:29 +08:00
|
|
|
|
|
|
|
_, extension = os.path.splitext(output_modelname)
|
|
|
|
if extension.lower() == ".safetensors":
|
|
|
|
safetensors.torch.save_file(theta_0, output_modelname, metadata={"format": "pt"})
|
|
|
|
else:
|
|
|
|
torch.save(theta_0, output_modelname)
|
2022-09-27 15:44:00 +08:00
|
|
|
|
2022-09-29 05:59:44 +08:00
|
|
|
sd_models.list_models()
|
|
|
|
|
2023-01-11 14:10:07 +08:00
|
|
|
create_config(output_modelname, config_source, primary_model_info, secondary_model_info, tertiary_model_info)
|
|
|
|
|
2023-01-19 15:39:51 +08:00
|
|
|
print(f"Checkpoint saved to {output_modelname}.")
|
|
|
|
shared.state.textinfo = "Checkpoint saved"
|
2023-01-03 23:21:51 +08:00
|
|
|
shared.state.end()
|
|
|
|
|
2023-01-19 14:25:37 +08:00
|
|
|
return [*[gr.Dropdown.update(choices=sd_models.checkpoint_tiles()) for _ in range(4)], "Checkpoint saved to " + output_modelname]
|