mirrors/Fooocus
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update backend + revise styles

Browse Source 
update backend + revise styles
This commit is contained in:
lllyasviel 2023-10-26 08:24:45 -07:00
parent 214c2ba04c
commit 175f3e3040
8 changed files with 51 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
backend/headless/fcbh/clip_vision.py
View File

@ -1,5 +1,5 @@
from transformers import CLIPVisionModelWithProjection, CLIPVisionConfig, CLIPImageProcessor, modeling_utils
from .utils import load_torch_file, transformers_convert
from transformers import CLIPVisionModelWithProjection, CLIPVisionConfig, modeling_utils
from .utils import load_torch_file, transformers_convert, common_upscale
import os
import torch
import contextlib
@ -7,6 +7,18 @@ import contextlib
import fcbh.ops
import fcbh.model_patcher
import fcbh.model_management
import fcbh.utils
def clip_preprocess(image, size=224):
mean = torch.tensor([ 0.48145466,0.4578275,0.40821073], device=image.device, dtype=image.dtype)
std = torch.tensor([0.26862954,0.26130258,0.27577711], device=image.device, dtype=image.dtype)
scale = (size / min(image.shape[1], image.shape[2]))
image = torch.nn.functional.interpolate(image.movedim(-1, 1), size=(round(scale * image.shape[1]), round(scale * image.shape[2])), mode="bicubic", antialias=True)
h = (image.shape[2] - size)//2
w = (image.shape[3] - size)//2
image = image[:,:,h:h+size,w:w+size]
image = torch.clip((255. * image), 0, 255).round() / 255.0
return (image - mean.view([3,1,1])) / std.view([3,1,1])
class ClipVisionModel():
def __init__(self, json_config):
@ -23,25 +35,12 @@ class ClipVisionModel():
self.model.to(self.dtype)
self.patcher = fcbh.model_patcher.ModelPatcher(self.model, load_device=self.load_device, offload_device=offload_device)
self.processor = CLIPImageProcessor(crop_size=224,
do_center_crop=True,
do_convert_rgb=True,
do_normalize=True,
do_resize=True,
image_mean=[ 0.48145466,0.4578275,0.40821073],
image_std=[0.26862954,0.26130258,0.27577711],
resample=3, #bicubic
size=224)
def load_sd(self, sd):
return self.model.load_state_dict(sd, strict=False)
def encode_image(self, image):
img = torch.clip((255. * image), 0, 255).round().int()
img = list(map(lambda a: a, img))
inputs = self.processor(images=img, return_tensors="pt")
fcbh.model_management.load_model_gpu(self.patcher)
pixel_values = inputs['pixel_values'].to(self.load_device)
pixel_values = clip_preprocess(image.to(self.load_device))
if self.dtype != torch.float32:
precision_scope = torch.autocast

9
backend/headless/fcbh/ldm/modules/attention.py
View File

@ -222,9 +222,14 @@ def attention_split(q, k, v, heads, mask=None):
mem_free_total = model_management.get_free_memory(q.device)
if _ATTN_PRECISION =="fp32":
element_size = 4
else:
element_size = q.element_size()
gb = 1024 ** 3
tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * q.element_size()
modifier = 3 if q.element_size() == 2 else 2.5
tensor_size = q.shape[0] * q.shape[1] * k.shape[1] * element_size
modifier = 3 if element_size == 2 else 2.5
mem_required = tensor_size * modifier
steps = 1

3
backend/headless/fcbh/ldm/modules/sub_quadratic_attention.py
View File

@ -83,7 +83,8 @@ def _summarize_chunk(
)
max_score, _ = torch.max(attn_weights, -1, keepdim=True)
max_score = max_score.detach()
torch.exp(attn_weights - max_score, out=attn_weights)
attn_weights -= max_score
torch.exp(attn_weights, out=attn_weights)
exp_weights = attn_weights.to(value.dtype)
exp_values = torch.bmm(exp_weights, value)
max_score = max_score.squeeze(-1)

18
fooocus_extras/ip_adapter.py
View File

@ -7,6 +7,7 @@ import fcbh.ldm.modules.attention as attention
from fooocus_extras.resampler import Resampler
from fcbh.model_patcher import ModelPatcher
from modules.core import numpy_to_pytorch
SD_V12_CHANNELS = [320] * 4 + [640] * 4 + [1280] * 4 + [1280] * 6 + [640] * 6 + [320] * 6 + [1280] * 2
@ -144,14 +145,27 @@ def load_ip_adapter(clip_vision_path, ip_negative_path, ip_adapter_path):
return
@torch.no_grad()
@torch.inference_mode()
def clip_preprocess(image):
mean = torch.tensor([0.48145466, 0.4578275, 0.40821073], device=image.device, dtype=image.dtype).view([1, 3, 1, 1])
std = torch.tensor([0.26862954, 0.26130258, 0.27577711], device=image.device, dtype=image.dtype).view([1, 3, 1, 1])
image = image.movedim(-1, 1)
# https://github.com/tencent-ailab/IP-Adapter/blob/d580c50a291566bbf9fc7ac0f760506607297e6d/README.md?plain=1#L75
B, C, H, W = image.shape
assert H == 224 and W == 224
return (image - mean) / std
@torch.no_grad()
@torch.inference_mode()
def preprocess(img):
global ip_unconds
inputs = clip_vision.processor(images=img, return_tensors="pt")
fcbh.model_management.load_model_gpu(clip_vision.patcher)
pixel_values = inputs['pixel_values'].to(clip_vision.load_device)
pixel_values = clip_preprocess(numpy_to_pytorch(img).to(clip_vision.load_device))
if clip_vision.dtype != torch.float32:
precision_scope = torch.autocast

2
fooocus_version.py
View File

@ -1 +1 @@
version = '2.1.744'
version = '2.1.745'

8
modules/async_worker.py
View File

@ -210,8 +210,8 @@ def worker():
if not skip_prompt_processing:
prompts = remove_empty_str([safe_str(p) for p in prompt.split('\n')], default='')
negative_prompts = remove_empty_str([safe_str(p) for p in negative_prompt.split('\n')], default='')
prompts = remove_empty_str([safe_str(p) for p in prompt.splitlines()], default='')
negative_prompts = remove_empty_str([safe_str(p) for p in negative_prompt.splitlines()], default='')
prompt = prompts[0]
negative_prompt = negative_prompts[0]
@ -239,8 +239,8 @@ def worker():
if use_style:
for s in style_selections:
p, n = apply_style(s, positive=task_prompt)
positive_basic_workloads.append(p)
negative_basic_workloads.append(n)
positive_basic_workloads += p
negative_basic_workloads += n
else:
positive_basic_workloads.append(task_prompt)

6
modules/sdxl_styles.py
View File

@ -40,7 +40,9 @@ for styles_file in styles_files:
try:
with open(os.path.join(styles_path, styles_file), encoding='utf-8') as f:
for entry in json.load(f):
name, prompt, negative_prompt = normalize_key(entry['name']), entry['prompt'], entry['negative_prompt']
name = normalize_key(entry['name'])
prompt = entry['prompt'] if 'prompt' in entry else ''
negative_prompt = entry['negative_prompt'] if 'negative_prompt' in entry else ''
styles[name] = (prompt, negative_prompt)
except Exception as e:
print(str(e))
@ -53,7 +55,7 @@ legal_style_names = [fooocus_expansion] + style_keys
def apply_style(style, positive):
p, n = styles[style]
return p.replace('{prompt}', positive), n
return p.replace('{prompt}', positive).splitlines(), n.splitlines()
def apply_wildcards(wildcard_text, rng, directory=wildcards_path):

6
sdxl_styles/sdxl_styles_fooocus.json
View File

@ -1,13 +1,12 @@
[
{
"name": "Fooocus Enhance",
"prompt": "{prompt} . (perfect real extremely details), award-winning, breathtaking, amazing fine detail, dramatic lighting, best quality",
"negative_prompt": "(worst quality, low quality, normal quality, lowres, low details, oversaturated, undersaturated, overexposed, underexposed, grayscale, bw, bad photo, bad photography, bad art:1.4), (watermark, signature, text font, username, error, logo, words, letters, digits, autograph, trademark, name:1.2), (blur, blurry, grainy), morbid, ugly, asymmetrical, mutated malformed, mutilated, poorly lit, bad shadow, draft, cropped, out of frame, cut off, censored, jpeg artifacts, out of focus, glitch, duplicate, (airbrushed, cartoon, anime, semi-realistic, cgi, render, blender, digital art, manga, amateur:1.3), (3D ,3D Game, 3D Game Scene, 3D Character:1.1), (bad hands, bad anatomy, bad body, bad face, bad teeth, bad arms, bad legs, deformities:1.3)"
},
{
"name": "Fooocus Sharp",
"prompt": "cinematic still {prompt} . sharp focus, emotional, harmonious, vignette, 4k epic detailed photograph shot on kodak detailed cinematic hbo dark moody, 35mm photo, high budget, cinemascope, moody, epic, gorgeous, film grain, grainy",
"negative_prompt": "blurry, anime, cartoon, blured background, graphic, bokeh, background blur, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured"
"prompt": "cinematic still {prompt} . emotional, harmonious, vignette, 4k epic detailed, shot on kodak, 35mm photo, sharp focus, cinemascope, moody, epic, gorgeous, film grain, grainy",
"negative_prompt": "anime, cartoon, graphic, (blur, blurry, bokeh), text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured"
},
{
"name": "Fooocus Masterpiece",
@ -21,7 +20,6 @@
},
{
"name": "Fooocus Negative",
"prompt": "",
"negative_prompt": "deformed, bad anatomy, disfigured, poorly drawn face, mutated, extra limb, ugly, poorly drawn hands, missing limb, floating limbs, disconnected limbs, disconnected head, malformed hands, long neck, mutated hands and fingers, bad hands, missing fingers, cropped, worst quality, low quality, mutation, poorly drawn, huge calf, bad hands, fused hand, missing hand, disappearing arms, disappearing thigh, disappearing calf, disappearing legs, missing fingers, fused fingers, abnormal eye proportion, Abnormal hands, abnormal legs, abnormal feet, abnormal fingers, drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly, anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch"
},
{