mirror of
https://github.com/invoke-ai/InvokeAI
synced 2024-08-30 20:32:17 +00:00
tested on 3.11 and 3.10
This commit is contained in:
Lincoln Stein
parent
4f9c728db0
commit
fc4e104c61
@ -12,7 +12,7 @@ from pydantic import BaseModel, Field, validator
|
||||
|
||||
from invokeai.app.invocations.metadata import CoreMetadata
|
||||
from invokeai.app.util.step_callback import stable_diffusion_step_callback
|
||||
from invokeai.backend.model_management.models.base import ModelType
|
||||
from invokeai.backend.model_management.models import ModelType, SilenceWarnings
|
||||
|
||||
from ...backend.model_management.lora import ModelPatcher
|
||||
from ...backend.stable_diffusion import PipelineIntermediateState
|
||||
@ -318,68 +318,69 @@ class TextToLatentsInvocation(BaseInvocation):
|
||||
|
||||
@torch.no_grad()
|
||||
def invoke(self, context: InvocationContext) -> LatentsOutput:
|
||||
noise = context.services.latents.get(self.noise.latents_name)
|
||||
with SilenceWarnings():
|
||||
noise = context.services.latents.get(self.noise.latents_name)
|
||||
|
||||
# Get the source node id (we are invoking the prepared node)
|
||||
graph_execution_state = context.services.graph_execution_manager.get(
|
||||
context.graph_execution_state_id
|
||||
)
|
||||
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
|
||||
# Get the source node id (we are invoking the prepared node)
|
||||
graph_execution_state = context.services.graph_execution_manager.get(
|
||||
context.graph_execution_state_id
|
||||
)
|
||||
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
|
||||
|
||||
def step_callback(state: PipelineIntermediateState):
|
||||
self.dispatch_progress(context, source_node_id, state)
|
||||
def step_callback(state: PipelineIntermediateState):
|
||||
self.dispatch_progress(context, source_node_id, state)
|
||||
|
||||
def _lora_loader():
|
||||
for lora in self.unet.loras:
|
||||
lora_info = context.services.model_manager.get_model(
|
||||
**lora.dict(exclude={"weight"}), context=context,
|
||||
def _lora_loader():
|
||||
for lora in self.unet.loras:
|
||||
lora_info = context.services.model_manager.get_model(
|
||||
**lora.dict(exclude={"weight"}), context=context,
|
||||
)
|
||||
yield (lora_info.context.model, lora.weight)
|
||||
del lora_info
|
||||
return
|
||||
|
||||
unet_info = context.services.model_manager.get_model(
|
||||
**self.unet.unet.dict(), context=context,
|
||||
)
|
||||
with ExitStack() as exit_stack,\
|
||||
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
|
||||
unet_info as unet:
|
||||
|
||||
noise = noise.to(device=unet.device, dtype=unet.dtype)
|
||||
|
||||
scheduler = get_scheduler(
|
||||
context=context,
|
||||
scheduler_info=self.unet.scheduler,
|
||||
scheduler_name=self.scheduler,
|
||||
)
|
||||
yield (lora_info.context.model, lora.weight)
|
||||
del lora_info
|
||||
return
|
||||
|
||||
unet_info = context.services.model_manager.get_model(
|
||||
**self.unet.unet.dict(), context=context,
|
||||
)
|
||||
with ExitStack() as exit_stack,\
|
||||
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
|
||||
unet_info as unet:
|
||||
pipeline = self.create_pipeline(unet, scheduler)
|
||||
conditioning_data = self.get_conditioning_data(context, scheduler, unet)
|
||||
|
||||
noise = noise.to(device=unet.device, dtype=unet.dtype)
|
||||
control_data = self.prep_control_data(
|
||||
model=pipeline, context=context, control_input=self.control,
|
||||
latents_shape=noise.shape,
|
||||
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
|
||||
do_classifier_free_guidance=True,
|
||||
exit_stack=exit_stack,
|
||||
)
|
||||
|
||||
scheduler = get_scheduler(
|
||||
context=context,
|
||||
scheduler_info=self.unet.scheduler,
|
||||
scheduler_name=self.scheduler,
|
||||
)
|
||||
# TODO: Verify the noise is the right size
|
||||
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
|
||||
latents=torch.zeros_like(noise, dtype=torch_dtype(unet.device)),
|
||||
noise=noise,
|
||||
num_inference_steps=self.steps,
|
||||
conditioning_data=conditioning_data,
|
||||
control_data=control_data, # list[ControlNetData]
|
||||
callback=step_callback,
|
||||
)
|
||||
|
||||
pipeline = self.create_pipeline(unet, scheduler)
|
||||
conditioning_data = self.get_conditioning_data(context, scheduler, unet)
|
||||
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
|
||||
result_latents = result_latents.to("cpu")
|
||||
torch.cuda.empty_cache()
|
||||
|
||||
control_data = self.prep_control_data(
|
||||
model=pipeline, context=context, control_input=self.control,
|
||||
latents_shape=noise.shape,
|
||||
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
|
||||
do_classifier_free_guidance=True,
|
||||
exit_stack=exit_stack,
|
||||
)
|
||||
|
||||
# TODO: Verify the noise is the right size
|
||||
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
|
||||
latents=torch.zeros_like(noise, dtype=torch_dtype(unet.device)),
|
||||
noise=noise,
|
||||
num_inference_steps=self.steps,
|
||||
conditioning_data=conditioning_data,
|
||||
control_data=control_data, # list[ControlNetData]
|
||||
callback=step_callback,
|
||||
)
|
||||
|
||||
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
|
||||
result_latents = result_latents.to("cpu")
|
||||
torch.cuda.empty_cache()
|
||||
|
||||
name = f'{context.graph_execution_state_id}__{self.id}'
|
||||
context.services.latents.save(name, result_latents)
|
||||
name = f'{context.graph_execution_state_id}__{self.id}'
|
||||
context.services.latents.save(name, result_latents)
|
||||
return build_latents_output(latents_name=name, latents=result_latents)
|
||||
|
||||
|
||||
@ -411,81 +412,82 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
|
||||
|
||||
@torch.no_grad()
|
||||
def invoke(self, context: InvocationContext) -> LatentsOutput:
|
||||
noise = context.services.latents.get(self.noise.latents_name)
|
||||
latent = context.services.latents.get(self.latents.latents_name)
|
||||
with SilenceWarnings(): # this quenches NSFW nag from diffusers
|
||||
noise = context.services.latents.get(self.noise.latents_name)
|
||||
latent = context.services.latents.get(self.latents.latents_name)
|
||||
|
||||
# Get the source node id (we are invoking the prepared node)
|
||||
graph_execution_state = context.services.graph_execution_manager.get(
|
||||
context.graph_execution_state_id
|
||||
)
|
||||
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
|
||||
# Get the source node id (we are invoking the prepared node)
|
||||
graph_execution_state = context.services.graph_execution_manager.get(
|
||||
context.graph_execution_state_id
|
||||
)
|
||||
source_node_id = graph_execution_state.prepared_source_mapping[self.id]
|
||||
|
||||
def step_callback(state: PipelineIntermediateState):
|
||||
self.dispatch_progress(context, source_node_id, state)
|
||||
def step_callback(state: PipelineIntermediateState):
|
||||
self.dispatch_progress(context, source_node_id, state)
|
||||
|
||||
def _lora_loader():
|
||||
for lora in self.unet.loras:
|
||||
lora_info = context.services.model_manager.get_model(
|
||||
**lora.dict(exclude={"weight"}), context=context,
|
||||
def _lora_loader():
|
||||
for lora in self.unet.loras:
|
||||
lora_info = context.services.model_manager.get_model(
|
||||
**lora.dict(exclude={"weight"}), context=context,
|
||||
)
|
||||
yield (lora_info.context.model, lora.weight)
|
||||
del lora_info
|
||||
return
|
||||
|
||||
unet_info = context.services.model_manager.get_model(
|
||||
**self.unet.unet.dict(), context=context,
|
||||
)
|
||||
with ExitStack() as exit_stack,\
|
||||
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
|
||||
unet_info as unet:
|
||||
|
||||
noise = noise.to(device=unet.device, dtype=unet.dtype)
|
||||
latent = latent.to(device=unet.device, dtype=unet.dtype)
|
||||
|
||||
scheduler = get_scheduler(
|
||||
context=context,
|
||||
scheduler_info=self.unet.scheduler,
|
||||
scheduler_name=self.scheduler,
|
||||
)
|
||||
yield (lora_info.context.model, lora.weight)
|
||||
del lora_info
|
||||
return
|
||||
|
||||
unet_info = context.services.model_manager.get_model(
|
||||
**self.unet.unet.dict(), context=context,
|
||||
)
|
||||
with ExitStack() as exit_stack,\
|
||||
ModelPatcher.apply_lora_unet(unet_info.context.model, _lora_loader()),\
|
||||
unet_info as unet:
|
||||
pipeline = self.create_pipeline(unet, scheduler)
|
||||
conditioning_data = self.get_conditioning_data(context, scheduler, unet)
|
||||
|
||||
noise = noise.to(device=unet.device, dtype=unet.dtype)
|
||||
latent = latent.to(device=unet.device, dtype=unet.dtype)
|
||||
control_data = self.prep_control_data(
|
||||
model=pipeline, context=context, control_input=self.control,
|
||||
latents_shape=noise.shape,
|
||||
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
|
||||
do_classifier_free_guidance=True,
|
||||
exit_stack=exit_stack,
|
||||
)
|
||||
|
||||
scheduler = get_scheduler(
|
||||
context=context,
|
||||
scheduler_info=self.unet.scheduler,
|
||||
scheduler_name=self.scheduler,
|
||||
)
|
||||
# TODO: Verify the noise is the right size
|
||||
initial_latents = latent if self.strength < 1.0 else torch.zeros_like(
|
||||
latent, device=unet.device, dtype=latent.dtype
|
||||
)
|
||||
|
||||
pipeline = self.create_pipeline(unet, scheduler)
|
||||
conditioning_data = self.get_conditioning_data(context, scheduler, unet)
|
||||
timesteps, _ = pipeline.get_img2img_timesteps(
|
||||
self.steps,
|
||||
self.strength,
|
||||
device=unet.device,
|
||||
)
|
||||
|
||||
control_data = self.prep_control_data(
|
||||
model=pipeline, context=context, control_input=self.control,
|
||||
latents_shape=noise.shape,
|
||||
# do_classifier_free_guidance=(self.cfg_scale >= 1.0))
|
||||
do_classifier_free_guidance=True,
|
||||
exit_stack=exit_stack,
|
||||
)
|
||||
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
|
||||
latents=initial_latents,
|
||||
timesteps=timesteps,
|
||||
noise=noise,
|
||||
num_inference_steps=self.steps,
|
||||
conditioning_data=conditioning_data,
|
||||
control_data=control_data, # list[ControlNetData]
|
||||
callback=step_callback
|
||||
)
|
||||
|
||||
# TODO: Verify the noise is the right size
|
||||
initial_latents = latent if self.strength < 1.0 else torch.zeros_like(
|
||||
latent, device=unet.device, dtype=latent.dtype
|
||||
)
|
||||
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
|
||||
result_latents = result_latents.to("cpu")
|
||||
torch.cuda.empty_cache()
|
||||
|
||||
timesteps, _ = pipeline.get_img2img_timesteps(
|
||||
self.steps,
|
||||
self.strength,
|
||||
device=unet.device,
|
||||
)
|
||||
|
||||
result_latents, result_attention_map_saver = pipeline.latents_from_embeddings(
|
||||
latents=initial_latents,
|
||||
timesteps=timesteps,
|
||||
noise=noise,
|
||||
num_inference_steps=self.steps,
|
||||
conditioning_data=conditioning_data,
|
||||
control_data=control_data, # list[ControlNetData]
|
||||
callback=step_callback
|
||||
)
|
||||
|
||||
# https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
|
||||
result_latents = result_latents.to("cpu")
|
||||
torch.cuda.empty_cache()
|
||||
|
||||
name = f'{context.graph_execution_state_id}__{self.id}'
|
||||
context.services.latents.save(name, result_latents)
|
||||
name = f'{context.graph_execution_state_id}__{self.id}'
|
||||
context.services.latents.save(name, result_latents)
|
||||
return build_latents_output(latents_name=name, latents=result_latents)
|
||||
|
||||
|
||||
|
||||
@ -163,7 +163,6 @@ class ModelCache(object):
|
||||
submodel: Optional[SubModelType] = None,
|
||||
gpu_load: bool = True,
|
||||
) -> Any:
|
||||
|
||||
if not isinstance(model_path, Path):
|
||||
model_path = Path(model_path)
|
||||
|
||||
|
||||
@ -391,7 +391,7 @@ class ModelManager(object):
|
||||
base_model: BaseModelType,
|
||||
model_type: ModelType,
|
||||
) -> str:
|
||||
return f"{base_model}/{model_type}/{model_name}"
|
||||
return f"{base_model.value}/{model_type.value}/{model_name}"
|
||||
|
||||
@classmethod
|
||||
def parse_key(cls, model_key: str) -> Tuple[str, BaseModelType, ModelType]:
|
||||
|
||||
@ -5,7 +5,7 @@ build-backend = "setuptools.build_meta"
|
||||
[project]
|
||||
name = "InvokeAI"
|
||||
description = "An implementation of Stable Diffusion which provides various new features and options to aid the image generation process"
|
||||
requires-python = ">=3.9, <3.11"
|
||||
requires-python = ">=3.9, <3.12"
|
||||
readme = { content-type = "text/markdown", file = "README.md" }
|
||||
keywords = ["stable-diffusion", "AI"]
|
||||
dynamic = ["version"]
|
||||
@ -32,16 +32,16 @@ classifiers = [
|
||||
'Topic :: Scientific/Engineering :: Image Processing',
|
||||
]
|
||||
dependencies = [
|
||||
"accelerate~=0.16",
|
||||
"accelerate~=0.21.0",
|
||||
"albumentations",
|
||||
"click",
|
||||
"clip_anytorch", # replacing "clip @ https://github.com/openai/CLIP/archive/eaa22acb90a5876642d0507623e859909230a52d.zip",
|
||||
"compel==2.0.0",
|
||||
"clip_anytorch", # replacing "clip @ https://github.com/openai/CLIP/archive/eaa22acb90a5876642d0507623e859909230a52d.zip",
|
||||
"compel~=2.0.0",
|
||||
"controlnet-aux>=0.0.6",
|
||||
"timm==0.6.13", # needed to override timm latest in controlnet_aux, see https://github.com/isl-org/ZoeDepth/issues/26
|
||||
"timm==0.6.13", # needed to override timm latest in controlnet_aux, see https://github.com/isl-org/ZoeDepth/issues/26
|
||||
"datasets",
|
||||
"diffusers[torch]~=0.18.1",
|
||||
"dnspython==2.2.1",
|
||||
"diffusers[torch]~=0.18.2",
|
||||
"dnspython~=2.4.0",
|
||||
"dynamicprompts",
|
||||
"easing-functions",
|
||||
"einops",
|
||||
@ -54,37 +54,37 @@ dependencies = [
|
||||
"flask_cors==3.0.10",
|
||||
"flask_socketio==5.3.0",
|
||||
"flaskwebgui==1.0.3",
|
||||
"gfpgan==1.3.8",
|
||||
"huggingface-hub>=0.11.1",
|
||||
"invisible-watermark>=0.2.0", # needed to install SDXL base and refiner using their repo_ids
|
||||
"invisible-watermark~=0.2.0", # needed to install SDXL base and refiner using their repo_ids
|
||||
"matplotlib", # needed for plotting of Penner easing functions
|
||||
"mediapipe", # needed for "mediapipeface" controlnet model
|
||||
"npyscreen",
|
||||
"numpy<1.24",
|
||||
"numpy==1.24.4",
|
||||
"omegaconf",
|
||||
"opencv-python",
|
||||
"picklescan",
|
||||
"pillow",
|
||||
"prompt-toolkit",
|
||||
"pympler==1.0.1",
|
||||
"pydantic==1.10.10",
|
||||
"pympler~=1.0.1",
|
||||
"pypatchmatch",
|
||||
'pyperclip',
|
||||
"pyreadline3",
|
||||
"python-multipart==0.0.6",
|
||||
"pytorch-lightning==1.7.7",
|
||||
"python-multipart",
|
||||
"pytorch-lightning",
|
||||
"realesrgan",
|
||||
"requests==2.28.2",
|
||||
"requests~=2.28.2",
|
||||
"rich~=13.3",
|
||||
"safetensors~=0.3.0",
|
||||
"scikit-image>=0.19",
|
||||
"scikit-image~=0.21.0",
|
||||
"send2trash",
|
||||
"test-tube>=0.7.5",
|
||||
"torch~=2.0.0",
|
||||
"torchvision>=0.14.1",
|
||||
"torchmetrics==0.11.4",
|
||||
"torchsde==0.2.5",
|
||||
"test-tube~=0.7.5",
|
||||
"torch~=2.0.1",
|
||||
"torchvision~=0.15.2",
|
||||
"torchmetrics~=1.0.1",
|
||||
"torchsde~=0.2.5",
|
||||
"transformers~=4.31.0",
|
||||
"uvicorn[standard]==0.21.1",
|
||||
"uvicorn[standard]~=0.21.1",
|
||||
"windows-curses; sys_platform=='win32'",
|
||||
]
|
||||
|
||||
|
||||
Loading…
Reference in New Issue
Block a user