Merge branch 'main' into feat/ui/expand-config

invokeai/app/invocations/generate.py

@@ -52,7 +52,7 @@ class TextToImageInvocation(BaseInvocation, SDImageInvocation):
     width:       int = Field(default=512, multiple_of=8, gt=0, description="The width of the resulting image", )
     height:      int = Field(default=512, multiple_of=8, gt=0, description="The height of the resulting image", )
     cfg_scale: float = Field(default=7.5, ge=1, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="k_lms", description="The scheduler to use" )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="lms", description="The scheduler to use" )
     model:       str = Field(default="", description="The model to use (currently ignored)")
     # fmt: on
 

invokeai/app/invocations/image.py

@@ -33,8 +33,8 @@ class ImageOutput(BaseInvocationOutput):
     # fmt: off
     type: Literal["image"] = "image"
     image:      ImageField = Field(default=None, description="The output image")
-    width:   Optional[int] = Field(default=None, description="The width of the image in pixels")
+    width:             int = Field(description="The width of the image in pixels")
-    height:  Optional[int] = Field(default=None, description="The height of the image in pixels")
+    height:            int = Field(description="The height of the image in pixels")
     # fmt: on
 
     class Config:

invokeai/app/invocations/latent.py

@@ -17,6 +17,7 @@ from ...backend.stable_diffusion.diffusion.shared_invokeai_diffusion import Post
 from ...backend.image_util.seamless import configure_model_padding
 from ...backend.prompting.conditioning import get_uc_and_c_and_ec
 from ...backend.stable_diffusion.diffusers_pipeline import ConditioningData, StableDiffusionGeneratorPipeline, image_resized_to_grid_as_tensor
+from ...backend.stable_diffusion.schedulers import SCHEDULER_MAP
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
 import numpy as np
 from ..services.image_storage import ImageType
@@ -40,41 +41,55 @@ class LatentsField(BaseModel):
 class LatentsOutput(BaseInvocationOutput):
     """Base class for invocations that output latents"""
     #fmt: off
-    type: Literal["latent_output"] = "latent_output"
+    type: Literal["latents_output"] = "latents_output"
-    latents: LatentsField            = Field(default=None, description="The output latents")
+
+    # Inputs
+    latents: LatentsField          = Field(default=None, description="The output latents")
+    width:                     int = Field(description="The width of the latents in pixels")
+    height:                    int = Field(description="The height of the latents in pixels")
     #fmt: on
 
+
+def build_latents_output(latents_name: str, latents: torch.Tensor):
+      return LatentsOutput(
+          latents=LatentsField(latents_name=latents_name),
+          width=latents.size()[3] * 8,
+          height=latents.size()[2] * 8,
+      )
+
 class NoiseOutput(BaseInvocationOutput):
     """Invocation noise output"""
     #fmt: off
-    type: Literal["noise_output"] = "noise_output"
+    type:  Literal["noise_output"] = "noise_output"
+
+    # Inputs
     noise: LatentsField            = Field(default=None, description="The output noise")
+    width:                     int = Field(description="The width of the noise in pixels")
+    height:                    int = Field(description="The height of the noise in pixels")
     #fmt: on
 
-
+def build_noise_output(latents_name: str, latents: torch.Tensor):
-# TODO: this seems like a hack
+      return NoiseOutput(
-scheduler_map = dict(
+          noise=LatentsField(latents_name=latents_name),
-    ddim=diffusers.DDIMScheduler,
+          width=latents.size()[3] * 8,
-    dpmpp_2=diffusers.DPMSolverMultistepScheduler,
+          height=latents.size()[2] * 8,
-    k_dpm_2=diffusers.KDPM2DiscreteScheduler,
+      )
-    k_dpm_2_a=diffusers.KDPM2AncestralDiscreteScheduler,
-    k_dpmpp_2=diffusers.DPMSolverMultistepScheduler,
-    k_euler=diffusers.EulerDiscreteScheduler,
-    k_euler_a=diffusers.EulerAncestralDiscreteScheduler,
-    k_heun=diffusers.HeunDiscreteScheduler,
-    k_lms=diffusers.LMSDiscreteScheduler,
-    plms=diffusers.PNDMScheduler,
-)
 
 
 SAMPLER_NAME_VALUES = Literal[
-    tuple(list(scheduler_map.keys()))
+    tuple(list(SCHEDULER_MAP.keys()))
 ]
 
 
 def get_scheduler(scheduler_name:str, model: StableDiffusionGeneratorPipeline)->Scheduler:
-    scheduler_class = scheduler_map.get(scheduler_name,'ddim')
+    scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP['ddim'])
-    scheduler = scheduler_class.from_config(model.scheduler.config)
+    
+    scheduler_config = model.scheduler.config
+    if "_backup" in scheduler_config:
+        scheduler_config = scheduler_config["_backup"]
+    scheduler_config = {**scheduler_config, **scheduler_extra_config, "_backup": scheduler_config}
+    scheduler = scheduler_class.from_config(scheduler_config)
+    
     # hack copied over from generate.py
     if not hasattr(scheduler, 'uses_inpainting_model'):
         scheduler.uses_inpainting_model = lambda: False
@@ -130,9 +145,7 @@ class NoiseInvocation(BaseInvocation):
 
         name = f'{context.graph_execution_state_id}__{self.id}'
         context.services.latents.set(name, noise)
-        return NoiseOutput(
+        return build_noise_output(latents_name=name, latents=noise)
-            noise=LatentsField(latents_name=name)
-        )
 
 
 # Text to image
@@ -148,7 +161,7 @@ class TextToLatentsInvocation(BaseInvocation):
     noise: Optional[LatentsField] = Field(description="The noise to use")
     steps:       int = Field(default=10, gt=0, description="The number of steps to use to generate the image")
     cfg_scale: float = Field(default=7.5, gt=0, description="The Classifier-Free Guidance, higher values may result in a result closer to the prompt", )
-    scheduler: SAMPLER_NAME_VALUES = Field(default="k_lms", description="The scheduler to use" )
+    scheduler: SAMPLER_NAME_VALUES = Field(default="lms", description="The scheduler to use" )
     model:       str = Field(default="", description="The model to use (currently ignored)")
     seamless:   bool = Field(default=False, description="Whether or not to generate an image that can tile without seams", )
     seamless_axes: str = Field(default="", description="The axes to tile the image on, 'x' and/or 'y'")
@@ -216,7 +229,7 @@ class TextToLatentsInvocation(BaseInvocation):
                 h_symmetry_time_pct=None,#h_symmetry_time_pct,
                 v_symmetry_time_pct=None#v_symmetry_time_pct,
             ),
-        ).add_scheduler_args_if_applicable(model.scheduler, eta=None)#ddim_eta)
+        ).add_scheduler_args_if_applicable(model.scheduler, eta=0.0)#ddim_eta)
         return conditioning_data
 
 
@@ -248,9 +261,7 @@ class TextToLatentsInvocation(BaseInvocation):
 
         name = f'{context.graph_execution_state_id}__{self.id}'
         context.services.latents.set(name, result_latents)
-        return LatentsOutput(
+        return build_latents_output(latents_name=name, latents=result_latents)
-            latents=LatentsField(latents_name=name)
-        )
 
 
 class LatentsToLatentsInvocation(TextToLatentsInvocation):
@@ -293,11 +304,7 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
             latent, device=model.device, dtype=latent.dtype
         )
 
-        timesteps, _ = model.get_img2img_timesteps(
+        timesteps, _ = model.get_img2img_timesteps(self.steps, self.strength)
-            self.steps,
-            self.strength,
-            device=model.device,
-        )
 
         result_latents, result_attention_map_saver = model.latents_from_embeddings(
             latents=initial_latents,
@@ -313,9 +320,7 @@ class LatentsToLatentsInvocation(TextToLatentsInvocation):
 
         name = f'{context.graph_execution_state_id}__{self.id}'
         context.services.latents.set(name, result_latents)
-        return LatentsOutput(
+        return build_latents_output(latents_name=name, latents=result_latents)
-            latents=LatentsField(latents_name=name)
-        )
 
 
 # Latent to image
@@ -379,11 +384,11 @@ class ResizeLatentsInvocation(BaseInvocation):
     type: Literal["lresize"] = "lresize"
 
     # Inputs
-    latents: Optional[LatentsField]             = Field(description="The latents to resize")
+    latents:    Optional[LatentsField] = Field(description="The latents to resize")
-    width: int                                  = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
+    width:                         int = Field(ge=64, multiple_of=8, description="The width to resize to (px)")
-    height: int                                 = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
+    height:                        int = Field(ge=64, multiple_of=8, description="The height to resize to (px)")
-    mode: Optional[LATENTS_INTERPOLATION_MODE]  = Field(default="bilinear", description="The interpolation mode")
+    mode:   LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
-    antialias: Optional[bool]                   = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
+    antialias:                    bool = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
@@ -400,7 +405,7 @@ class ResizeLatentsInvocation(BaseInvocation):
 
         name = f"{context.graph_execution_state_id}__{self.id}"
         context.services.latents.set(name, resized_latents)
-        return LatentsOutput(latents=LatentsField(latents_name=name))
+        return build_latents_output(latents_name=name, latents=resized_latents)
 
 
 class ScaleLatentsInvocation(BaseInvocation):
@@ -409,10 +414,10 @@ class ScaleLatentsInvocation(BaseInvocation):
     type: Literal["lscale"] = "lscale"
 
     # Inputs
-    latents: Optional[LatentsField]             = Field(description="The latents to scale")
+    latents:   Optional[LatentsField] = Field(description="The latents to scale")
-    scale_factor: float                         = Field(gt=0, description="The factor by which to scale the latents")
+    scale_factor:               float = Field(gt=0, description="The factor by which to scale the latents")
-    mode: Optional[LATENTS_INTERPOLATION_MODE]  = Field(default="bilinear", description="The interpolation mode")
+    mode:  LATENTS_INTERPOLATION_MODE = Field(default="bilinear", description="The interpolation mode")
-    antialias: Optional[bool]                   = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
+    antialias:                   bool = Field(default=False, description="Whether or not to antialias (applied in bilinear and bicubic modes only)")
 
     def invoke(self, context: InvocationContext) -> LatentsOutput:
         latents = context.services.latents.get(self.latents.latents_name)
@@ -430,7 +435,7 @@ class ScaleLatentsInvocation(BaseInvocation):
 
         name = f"{context.graph_execution_state_id}__{self.id}"
         context.services.latents.set(name, resized_latents)
-        return LatentsOutput(latents=LatentsField(latents_name=name))
+        return build_latents_output(latents_name=name, latents=resized_latents)
 
 
 class ImageToLatentsInvocation(BaseInvocation):
@@ -474,4 +479,4 @@ class ImageToLatentsInvocation(BaseInvocation):
 
         name = f"{context.graph_execution_state_id}__{self.id}"
         context.services.latents.set(name, latents)
-        return LatentsOutput(latents=LatentsField(latents_name=name))
+        return build_latents_output(latents_name=name, latents=latents)

invokeai/app/invocations/math.py

@@ -3,6 +3,7 @@
 from typing import Literal
 
 from pydantic import BaseModel, Field
+import numpy as np
 
 from .baseinvocation import BaseInvocation, BaseInvocationOutput, InvocationContext, InvocationConfig
 
@@ -73,3 +74,12 @@ class DivideInvocation(BaseInvocation, MathInvocationConfig):
 
     def invoke(self, context: InvocationContext) -> IntOutput:
         return IntOutput(a=int(self.a / self.b))
+
+
+class RandomIntInvocation(BaseInvocation):
+    """Outputs a single random integer."""
+    #fmt: off
+    type: Literal["rand_int"] = "rand_int"
+    #fmt: on
+    def invoke(self, context: InvocationContext) -> IntOutput:
+        return IntOutput(a=np.random.randint(0, np.iinfo(np.int32).max))

invokeai/backend/args.py

@@ -108,17 +108,20 @@ APP_VERSION = invokeai.version.__version__
 
 SAMPLER_CHOICES = [
     "ddim",
-    "k_dpm_2_a",
+    "ddpm",
-    "k_dpm_2",
+    "deis",
-    "k_dpmpp_2_a",
+    "lms",
-    "k_dpmpp_2",
-    "k_euler_a",
-    "k_euler",
-    "k_heun",
-    "k_lms",
-    "plms",
-    # diffusers:
     "pndm",
+    "heun",
+    "euler",
+    "euler_k",
+    "euler_a",
+    "kdpm_2",
+    "kdpm_2_a",
+    "dpmpp_2s",
+    "dpmpp_2m",
+    "dpmpp_2m_k",
+    "unipc",
 ]
 
 PRECISION_CHOICES = [
@@ -631,7 +634,7 @@ class Args(object):
             choices=SAMPLER_CHOICES,
             metavar="SAMPLER_NAME",
             help=f'Set the default sampler. Supported samplers: {", ".join(SAMPLER_CHOICES)}',
-            default="k_lms",
+            default="lms",
         )
         render_group.add_argument(
             "--log_tokenization",

invokeai/backend/generate.py

@@ -37,6 +37,7 @@ from .safety_checker import SafetyChecker
 from .prompting import get_uc_and_c_and_ec
 from .prompting.conditioning import log_tokenization
 from .stable_diffusion import HuggingFaceConceptsLibrary
+from .stable_diffusion.schedulers import SCHEDULER_MAP
 from .util import choose_precision, choose_torch_device
 
 def fix_func(orig):
@@ -141,7 +142,7 @@ class Generate:
         model=None,
         conf="configs/models.yaml",
         embedding_path=None,
-        sampler_name="k_lms",
+        sampler_name="lms",
         ddim_eta=0.0,  # deterministic
         full_precision=False,
         precision="auto",
@@ -1047,29 +1048,12 @@ class Generate:
     def _set_scheduler(self):
         default = self.model.scheduler
 
-        # See https://github.com/huggingface/diffusers/issues/277#issuecomment-1371428672
+        if self.sampler_name in SCHEDULER_MAP:
-        scheduler_map = dict(
+            sampler_class, sampler_extra_config = SCHEDULER_MAP[self.sampler_name]
-            ddim=diffusers.DDIMScheduler,
-            dpmpp_2=diffusers.DPMSolverMultistepScheduler,
-            k_dpm_2=diffusers.KDPM2DiscreteScheduler,
-            k_dpm_2_a=diffusers.KDPM2AncestralDiscreteScheduler,
-            # DPMSolverMultistepScheduler is technically not `k_` anything, as it is neither
-            # the k-diffusers implementation nor included in EDM (Karras 2022), but we can
-            # provide an alias for compatibility.
-            k_dpmpp_2=diffusers.DPMSolverMultistepScheduler,
-            k_euler=diffusers.EulerDiscreteScheduler,
-            k_euler_a=diffusers.EulerAncestralDiscreteScheduler,
-            k_heun=diffusers.HeunDiscreteScheduler,
-            k_lms=diffusers.LMSDiscreteScheduler,
-            plms=diffusers.PNDMScheduler,
-        )
-
-        if self.sampler_name in scheduler_map:
-            sampler_class = scheduler_map[self.sampler_name]
             msg = (
                 f"Setting Sampler to {self.sampler_name} ({sampler_class.__name__})"
             )
-            self.sampler = sampler_class.from_config(self.model.scheduler.config)
+            self.sampler = sampler_class.from_config({**self.model.scheduler.config, **sampler_extra_config})
         else:
             msg = (
                 f" Unsupported Sampler: {self.sampler_name} "+

invokeai/backend/generator/base.py

@@ -31,6 +31,7 @@ from ..util.util import rand_perlin_2d
 from ..safety_checker import SafetyChecker
 from ..prompting.conditioning import get_uc_and_c_and_ec
 from ..stable_diffusion.diffusers_pipeline import StableDiffusionGeneratorPipeline
+from ..stable_diffusion.schedulers import SCHEDULER_MAP
 
 downsampling = 8
 
@@ -71,19 +72,6 @@ class InvokeAIGeneratorOutput:
 # we are interposing a wrapper around the original Generator classes so that
 # old code that calls Generate will continue to work.
 class InvokeAIGenerator(metaclass=ABCMeta):
-    scheduler_map = dict(
-        ddim=diffusers.DDIMScheduler,
-        dpmpp_2=diffusers.DPMSolverMultistepScheduler,
-        k_dpm_2=diffusers.KDPM2DiscreteScheduler,
-        k_dpm_2_a=diffusers.KDPM2AncestralDiscreteScheduler,
-        k_dpmpp_2=diffusers.DPMSolverMultistepScheduler,
-        k_euler=diffusers.EulerDiscreteScheduler,
-        k_euler_a=diffusers.EulerAncestralDiscreteScheduler,
-        k_heun=diffusers.HeunDiscreteScheduler,
-        k_lms=diffusers.LMSDiscreteScheduler,
-        plms=diffusers.PNDMScheduler,
-    )
-
     def __init__(self,
                  model_info: dict,
                  params: InvokeAIGeneratorBasicParams=InvokeAIGeneratorBasicParams(),
@@ -175,14 +163,20 @@ class InvokeAIGenerator(metaclass=ABCMeta):
         '''
         Return list of all the schedulers that we currently handle.
         '''
-        return list(self.scheduler_map.keys())
+        return list(SCHEDULER_MAP.keys())
 
     def load_generator(self, model: StableDiffusionGeneratorPipeline, generator_class: Type[Generator]):
         return generator_class(model, self.params.precision)
 
     def get_scheduler(self, scheduler_name:str, model: StableDiffusionGeneratorPipeline)->Scheduler:
-        scheduler_class = self.scheduler_map.get(scheduler_name,'ddim')
+        scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP['ddim'])
-        scheduler = scheduler_class.from_config(model.scheduler.config)
+        
+        scheduler_config = model.scheduler.config
+        if "_backup" in scheduler_config:
+            scheduler_config = scheduler_config["_backup"]
+        scheduler_config = {**scheduler_config, **scheduler_extra_config, "_backup": scheduler_config}
+        scheduler = scheduler_class.from_config(scheduler_config)
+        
         # hack copied over from generate.py
         if not hasattr(scheduler, 'uses_inpainting_model'):
             scheduler.uses_inpainting_model = lambda: False

invokeai/backend/model_management/convert_ckpt_to_diffusers.py

@@ -47,6 +47,7 @@ from diffusers import (
     LDMTextToImagePipeline,
     LMSDiscreteScheduler,
     PNDMScheduler,
+    UniPCMultistepScheduler,
     StableDiffusionPipeline,
     UNet2DConditionModel,
 )
@@ -1209,6 +1210,8 @@ def load_pipeline_from_original_stable_diffusion_ckpt(
             scheduler = EulerAncestralDiscreteScheduler.from_config(scheduler.config)
         elif scheduler_type == "dpm":
             scheduler = DPMSolverMultistepScheduler.from_config(scheduler.config)
+        elif scheduler_type == 'unipc':
+            scheduler = UniPCMultistepScheduler.from_config(scheduler.config)
         elif scheduler_type == "ddim":
             scheduler = scheduler
         else:

invokeai/backend/model_management/model_manager.py

@@ -30,7 +30,7 @@ from diffusers import (
     UNet2DConditionModel,
     SchedulerMixin,
     logging as dlogging,
-)    
+)
 from huggingface_hub import scan_cache_dir
 from omegaconf import OmegaConf
 from omegaconf.dictconfig import DictConfig
@@ -68,7 +68,7 @@ class SDModelComponent(Enum):
     scheduler="scheduler"
     safety_checker="safety_checker"
     feature_extractor="feature_extractor"
-    
+
 DEFAULT_MAX_MODELS = 2
 
 class ModelManager(object):
@@ -182,7 +182,7 @@ class ModelManager(object):
         vae from the model currently in the GPU.
         """
         return self._get_sub_model(model_name, SDModelComponent.vae)
-                                   
+
     def get_model_tokenizer(self, model_name: str=None)->CLIPTokenizer:
         """Given a model name identified in models.yaml, load the model into
         GPU if necessary and return its assigned CLIPTokenizer. If no
@@ -190,12 +190,12 @@ class ModelManager(object):
         currently in the GPU.
         """
         return self._get_sub_model(model_name, SDModelComponent.tokenizer)
-                                   
+
     def get_model_unet(self, model_name: str=None)->UNet2DConditionModel:
         """Given a model name identified in models.yaml, load the model into
         GPU if necessary and return its assigned UNet2DConditionModel. If no model
         name is provided, return the UNet from the model
-        currently in the GPU.  
+        currently in the GPU.
         """
         return self._get_sub_model(model_name, SDModelComponent.unet)
 
@@ -222,7 +222,7 @@ class ModelManager(object):
         currently in the GPU.
         """
         return self._get_sub_model(model_name, SDModelComponent.scheduler)
-    
+
     def _get_sub_model(
             self,
             model_name: str=None,
@@ -1228,7 +1228,7 @@ class ModelManager(object):
                         sha.update(chunk)
         hash = sha.hexdigest()
         toc = time.time()
-        self.logger.debug(f"sha256 = {hash} ({count} files hashed in", "%4.2fs)" % (toc - tic))
+        self.logger.debug(f"sha256 = {hash} ({count} files hashed in {toc - tic:4.2f}s)")
         with open(hashpath, "w") as f:
             f.write(hash)
         return hash

invokeai/backend/stable_diffusion/diffusers_pipeline.py

@@ -509,10 +509,13 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         run_id=None,
         callback: Callable[[PipelineIntermediateState], None] = None,
     ) -> tuple[torch.Tensor, Optional[AttentionMapSaver]]:
+        if self.scheduler.config.get("cpu_only", False):
+            scheduler_device = torch.device('cpu')
+        else:
+            scheduler_device = self._model_group.device_for(self.unet)
+
         if timesteps is None:
-            self.scheduler.set_timesteps(
+            self.scheduler.set_timesteps(num_inference_steps, device=scheduler_device)
-                num_inference_steps, device=self._model_group.device_for(self.unet)
-            )
             timesteps = self.scheduler.timesteps
         infer_latents_from_embeddings = GeneratorToCallbackinator(
             self.generate_latents_from_embeddings, PipelineIntermediateState
@@ -725,12 +728,8 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         noise: torch.Tensor,
         run_id=None,
         callback=None,
-    ) -> InvokeAIStableDiffusionPipelineOutput:
+    ) -> InvokeAIStableDiffusionPipelineOutput:        
-        timesteps, _ = self.get_img2img_timesteps(
+        timesteps, _ = self.get_img2img_timesteps(num_inference_steps, strength)
-            num_inference_steps,
-            strength,
-            device=self._model_group.device_for(self.unet),
-        )
         result_latents, result_attention_maps = self.latents_from_embeddings(
             latents=initial_latents if strength < 1.0 else torch.zeros_like(
                 initial_latents, device=initial_latents.device, dtype=initial_latents.dtype
@@ -756,13 +755,19 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
             return self.check_for_safety(output, dtype=conditioning_data.dtype)
 
     def get_img2img_timesteps(
-        self, num_inference_steps: int, strength: float, device
+        self, num_inference_steps: int, strength: float, device=None
     ) -> (torch.Tensor, int):
         img2img_pipeline = StableDiffusionImg2ImgPipeline(**self.components)
         assert img2img_pipeline.scheduler is self.scheduler
-        img2img_pipeline.scheduler.set_timesteps(num_inference_steps, device=device)
+
+        if self.scheduler.config.get("cpu_only", False):
+            scheduler_device = torch.device('cpu')
+        else:
+            scheduler_device = self._model_group.device_for(self.unet)
+
+        img2img_pipeline.scheduler.set_timesteps(num_inference_steps, device=scheduler_device)
         timesteps, adjusted_steps = img2img_pipeline.get_timesteps(
-            num_inference_steps, strength, device=device
+            num_inference_steps, strength, device=scheduler_device
         )
         # Workaround for low strength resulting in zero timesteps.
         # TODO: submit upstream fix for zero-step img2img
@@ -796,9 +801,7 @@ class StableDiffusionGeneratorPipeline(StableDiffusionPipeline):
         if init_image.dim() == 3:
             init_image = init_image.unsqueeze(0)
 
-        timesteps, _ = self.get_img2img_timesteps(
+        timesteps, _ = self.get_img2img_timesteps(num_inference_steps, strength)
-            num_inference_steps, strength, device=device
-        )
 
         # 6. Prepare latent variables
         # can't quite use upstream StableDiffusionImg2ImgPipeline.prepare_latents

invokeai/backend/stable_diffusion/schedulers/__init__.py

@@ -0,0 +1 @@
+from .schedulers import SCHEDULER_MAP

invokeai/backend/stable_diffusion/schedulers/schedulers.py

@@ -0,0 +1,22 @@
+from diffusers import DDIMScheduler, DPMSolverMultistepScheduler, KDPM2DiscreteScheduler, \
+    KDPM2AncestralDiscreteScheduler, EulerDiscreteScheduler, EulerAncestralDiscreteScheduler, \
+    HeunDiscreteScheduler, LMSDiscreteScheduler, PNDMScheduler, UniPCMultistepScheduler, \
+    DPMSolverSinglestepScheduler, DEISMultistepScheduler, DDPMScheduler
+
+SCHEDULER_MAP = dict(
+    ddim=(DDIMScheduler, dict()),
+    ddpm=(DDPMScheduler, dict()),
+    deis=(DEISMultistepScheduler, dict()),
+    lms=(LMSDiscreteScheduler, dict()),
+    pndm=(PNDMScheduler, dict()),
+    heun=(HeunDiscreteScheduler, dict()),
+    euler=(EulerDiscreteScheduler, dict(use_karras_sigmas=False)),
+    euler_k=(EulerDiscreteScheduler, dict(use_karras_sigmas=True)),
+    euler_a=(EulerAncestralDiscreteScheduler, dict()),
+    kdpm_2=(KDPM2DiscreteScheduler, dict()),
+    kdpm_2_a=(KDPM2AncestralDiscreteScheduler, dict()),
+    dpmpp_2s=(DPMSolverSinglestepScheduler, dict()),
+    dpmpp_2m=(DPMSolverMultistepScheduler, dict(use_karras_sigmas=False)),
+    dpmpp_2m_k=(DPMSolverMultistepScheduler, dict(use_karras_sigmas=True)),
+    unipc=(UniPCMultistepScheduler, dict(cpu_only=True))
+)

invokeai/backend/web/modules/parameters.py

@@ -4,17 +4,20 @@ from .parse_seed_weights import parse_seed_weights
 
 SAMPLER_CHOICES = [
     "ddim",
-    "k_dpm_2_a",
+    "ddpm",
-    "k_dpm_2",
+    "deis",
-    "k_dpmpp_2_a",
+    "lms",
-    "k_dpmpp_2",
-    "k_euler_a",
-    "k_euler",
-    "k_heun",
-    "k_lms",
-    "plms",
-    # diffusers:
     "pndm",
+    "heun",
+    "euler",
+    "euler_k",
+    "euler_a",
+    "kdpm_2",
+    "kdpm_2_a",
+    "dpmpp_2s",
+    "dpmpp_2m",
+    "dpmpp_2m_k",
+    "unipc",
 ]
 
 

invokeai/frontend/web/docs/README.md

@@ -37,7 +37,7 @@ From `invokeai/frontend/web/` run `yarn install` to get everything set up.
 Start everything in dev mode:
 
 1. Start the dev server: `yarn dev`
-2. Start the InvokeAI UI per usual: `invokeai --web`
+2. Start the InvokeAI Nodes backend: `python scripts/invokeai-new.py --web # run from the repo root`
 3. Point your browser to the dev server address e.g. <http://localhost:5173/>
 
 ### Production builds

invokeai/frontend/web/src/app/constants.ts

@@ -2,17 +2,28 @@
 
 export const DIFFUSERS_SCHEDULERS: Array<string> = [
   'ddim',
-  'plms',
+  'ddpm',
-  'k_lms',
+  'deis',
-  'dpmpp_2',
+  'lms',
-  'k_dpm_2',
+  'pndm',
-  'k_dpm_2_a',
+  'heun',
-  'k_dpmpp_2',
+  'euler',
-  'k_euler',
+  'euler_k',
-  'k_euler_a',
+  'euler_a',
-  'k_heun',
+  'kdpm_2',
+  'kdpm_2_a',
+  'dpmpp_2s',
+  'dpmpp_2m',
+  'dpmpp_2m_k',
+  'unipc',
 ];
 
+export const IMG2IMG_DIFFUSERS_SCHEDULERS = DIFFUSERS_SCHEDULERS.filter(
+  (scheduler) => {
+    return scheduler !== 'dpmpp_2s';
+  }
+);
+
 // Valid image widths
 export const WIDTHS: Array<number> = Array.from(Array(64)).map(
   (_x, i) => (i + 1) * 64

invokeai/frontend/web/src/app/types/invokeai.ts

@@ -47,15 +47,20 @@ export type CommonGeneratedImageMetadata = {
   postprocessing: null | Array<ESRGANMetadata | FacetoolMetadata>;
   sampler:
     | 'ddim'
-    | 'k_dpm_2_a'
+    | 'ddpm'
-    | 'k_dpm_2'
+    | 'deis'
-    | 'k_dpmpp_2_a'
+    | 'lms'
-    | 'k_dpmpp_2'
+    | 'pndm'
-    | 'k_euler_a'
+    | 'heun'
-    | 'k_euler'
+    | 'euler'
-    | 'k_heun'
+    | 'euler_k'
-    | 'k_lms'
+    | 'euler_a'
-    | 'plms';
+    | 'kdpm_2'
+    | 'kdpm_2_a'
+    | 'dpmpp_2s'
+    | 'dpmpp_2m'
+    | 'dpmpp_2m_k'
+    | 'unipc';
   prompt: Prompt;
   seed: number;
   variations: SeedWeights;

invokeai/frontend/web/src/features/nodes/examples/iterationGraph.ts

@@ -20,7 +20,7 @@ export const iterationGraph = {
       model: '',
       progress_images: false,
       prompt: 'dog',
-      sampler_name: 'k_lms',
+      sampler_name: 'lms',
       seamless: false,
       steps: 11,
       type: 'txt2img',

invokeai/frontend/web/src/features/parameters/components/Parameters/Core/ParamSampler.tsx

@@ -1,8 +1,12 @@
-import { DIFFUSERS_SCHEDULERS } from 'app/constants';
+import {
+  DIFFUSERS_SCHEDULERS,
+  IMG2IMG_DIFFUSERS_SCHEDULERS,
+} from 'app/constants';
 import { RootState } from 'app/store/store';
 import { useAppDispatch, useAppSelector } from 'app/store/storeHooks';
 import IAISelect from 'common/components/IAISelect';
 import { setSampler } from 'features/parameters/store/generationSlice';
+import { activeTabNameSelector } from 'features/ui/store/uiSelectors';
 import { ChangeEvent, memo, useCallback } from 'react';
 import { useTranslation } from 'react-i18next';
 
@@ -10,6 +14,9 @@ const ParamSampler = () => {
   const sampler = useAppSelector(
     (state: RootState) => state.generation.sampler
   );
+
+  const activeTabName = useAppSelector(activeTabNameSelector);
+
   const dispatch = useAppDispatch();
   const { t } = useTranslation();
 
@@ -23,7 +30,11 @@ const ParamSampler = () => {
       label={t('parameters.sampler')}
       value={sampler}
       onChange={handleChange}
-      validValues={DIFFUSERS_SCHEDULERS}
+      validValues={
+        activeTabName === 'img2img' || activeTabName == 'unifiedCanvas'
+          ? IMG2IMG_DIFFUSERS_SCHEDULERS
+          : DIFFUSERS_SCHEDULERS
+      }
       minWidth={36}
     />
   );

invokeai/frontend/web/src/features/parameters/components/Parameters/ImageToImage/ImageToImageStrength.tsx

@@ -47,7 +47,7 @@ const ImageToImageStrength = () => {
 
   return (
     <IAISlider
-      label={`${t('parameters.strength')}`}
+      label={`${t('parameters.denoisingStrength')}`}
       step={step}
       min={min}
       max={sliderMax}

invokeai/frontend/web/src/features/parameters/store/generationSlice.ts

@@ -51,7 +51,7 @@ export const initialGenerationState: GenerationState = {
   perlin: 0,
   prompt: '',
   negativePrompt: '',
-  sampler: 'k_lms',
+  sampler: 'lms',
   seamBlur: 16,
   seamSize: 96,
   seamSteps: 30,

invokeai/frontend/web/src/features/ui/components/tabs/ImageToImage/ImageToImageTabCoreParameters.tsx

@@ -64,8 +64,6 @@ const ImageToImageTabCoreParameters = () => {
             <ParamSteps />
             <ParamCFGScale />
           </Flex>
-          <ParamWidth isDisabled={!shouldFitToWidthHeight} />
-          <ParamHeight isDisabled={!shouldFitToWidthHeight} />
           <Flex gap={3} w="full">
             <Box flexGrow={2}>
               <ParamSampler />
@@ -74,6 +72,8 @@ const ImageToImageTabCoreParameters = () => {
               <ModelSelect />
             </Box>
           </Flex>
+          <ParamWidth isDisabled={!shouldFitToWidthHeight} />
+          <ParamHeight isDisabled={!shouldFitToWidthHeight} />
           <ImageToImageStrength />
           <ImageToImageFit />
         </Flex>

invokeai/frontend/web/src/features/ui/components/tabs/UnifiedCanvas/UnifiedCanvasCoreParameters.tsx

@@ -62,8 +62,6 @@ const UnifiedCanvasCoreParameters = () => {
             <ParamSteps />
             <ParamCFGScale />
           </Flex>
-          <ParamWidth />
-          <ParamHeight />
           <Flex gap={3} w="full">
             <Box flexGrow={2}>
               <ParamSampler />
@@ -72,8 +70,9 @@ const UnifiedCanvasCoreParameters = () => {
               <ModelSelect />
             </Box>
           </Flex>
+          <ParamWidth />
+          <ParamHeight />
           <ImageToImageStrength />
-          <ImageToImageFit />
         </Flex>
       )}
     </Flex>

invokeai/frontend/web/src/i18n.ts

@@ -3,7 +3,7 @@ import LanguageDetector from 'i18next-browser-languagedetector';
 import Backend from 'i18next-http-backend';
 import { initReactI18next } from 'react-i18next';
 
-import translationEN from '../dist/locales/en.json';
+import translationEN from '../public/locales/en.json';
 import { LOCALSTORAGE_PREFIX } from 'app/store/constants';
 
 if (import.meta.env.MODE === 'package') {

tests/inpainting/original.json

@@ -23,7 +23,7 @@
     ],
     "threshold": 0,
     "postprocessing": null,
-    "sampler": "k_lms",
+    "sampler": "lms",
     "variations": [],
     "type": "txt2img"
   }

tests/nodes/test_png_metadata_service.py

@@ -17,7 +17,7 @@ valid_metadata = {
         "width": 512,
         "height": 512,
         "cfg_scale": 7.5,
-        "scheduler": "k_lms",
+        "scheduler": "lms",
         "model": "stable-diffusion-1.5",
     },
 }