Add DPMPP Single, Euler Karras and DPMPP2 Multi Karras Schedulers

invokeai/app/invocations/latent.py

@@ -60,8 +60,13 @@ SAMPLER_NAME_VALUES = Literal[
 
 def get_scheduler(scheduler_name:str, model: StableDiffusionGeneratorPipeline)->Scheduler:
     scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP['ddim'])
-    scheduler_config = {**model.scheduler.config, **scheduler_extra_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/args.py

@@ -108,18 +108,18 @@ APP_VERSION = invokeai.version.__version__
 
 SAMPLER_CHOICES = [
     "ddim",
-    "k_dpm_2_a",
-    "k_dpm_2",
-    "k_dpmpp_2_a",
-    "k_dpmpp_2",
-    "k_euler_a",
-    "k_euler",
-    "k_heun",
     "k_lms",
     "plms",
-    # diffusers:
+    "k_heun",
-    "pndm",
+    "k_euler",
-    "unipc"
+    "euler_karras",
+    "k_euler_a",
+    "k_dpm_2",
+    "k_dpm_2_a",
+    "dpmpp_2s",
+    "k_dpmpp_2",
+    "k_dpmpp_2_karras",
+    "unipc",
 ]
 
 PRECISION_CHOICES = [

invokeai/backend/generator/base.py

@@ -170,8 +170,13 @@ class InvokeAIGenerator(metaclass=ABCMeta):
 
     def get_scheduler(self, scheduler_name:str, model: StableDiffusionGeneratorPipeline)->Scheduler:
         scheduler_class, scheduler_extra_config = SCHEDULER_MAP.get(scheduler_name, SCHEDULER_MAP['ddim'])
-        scheduler_config = {**model.scheduler.config, **scheduler_extra_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/stable_diffusion/schedulers/schedulers.py

@@ -1,17 +1,20 @@
 from diffusers import DDIMScheduler, DPMSolverMultistepScheduler, KDPM2DiscreteScheduler, \
     KDPM2AncestralDiscreteScheduler, EulerDiscreteScheduler, EulerAncestralDiscreteScheduler, \
-    HeunDiscreteScheduler, LMSDiscreteScheduler, PNDMScheduler, UniPCMultistepScheduler
+    HeunDiscreteScheduler, LMSDiscreteScheduler, PNDMScheduler, UniPCMultistepScheduler, \
+    DPMSolverSinglestepScheduler
 
 SCHEDULER_MAP = dict(
-    ddim=(DDIMScheduler, dict(cpu_only=False)),
+    ddim=(DDIMScheduler, dict()),
-    dpmpp_2=(DPMSolverMultistepScheduler, dict(cpu_only=False)),
+    k_lms=(LMSDiscreteScheduler, dict()),
-    k_dpm_2=(KDPM2DiscreteScheduler, dict(cpu_only=False)),
+    plms=(PNDMScheduler, dict()),
-    k_dpm_2_a=(KDPM2AncestralDiscreteScheduler, dict(cpu_only=False)),
+    k_euler=(EulerDiscreteScheduler, dict(use_karras_sigmas=False)),
-    k_dpmpp_2=(DPMSolverMultistepScheduler, dict(cpu_only=False)),
+    euler_karras=(EulerDiscreteScheduler, dict(use_karras_sigmas=True)),
-    k_euler=(EulerDiscreteScheduler, dict(cpu_only=False)),
+    k_euler_a=(EulerAncestralDiscreteScheduler, dict()),
-    k_euler_a=(EulerAncestralDiscreteScheduler, dict(cpu_only=False)),
+    k_dpm_2=(KDPM2DiscreteScheduler, dict()),
-    k_heun=(HeunDiscreteScheduler, dict(cpu_only=False)),
+    k_dpm_2_a=(KDPM2AncestralDiscreteScheduler, dict()),
-    k_lms=(LMSDiscreteScheduler, dict(cpu_only=False)),
+    dpmpp_2s=(DPMSolverSinglestepScheduler, dict()),
-    plms=(PNDMScheduler, dict(cpu_only=False)),
+    k_dpmpp_2=(DPMSolverMultistepScheduler, dict(use_karras_sigmas=False)),
+    k_dpmpp_2_karras=(DPMSolverMultistepScheduler, dict(use_karras_sigmas=True)),
+    k_heun=(HeunDiscreteScheduler, dict()),
     unipc=(UniPCMultistepScheduler, dict(cpu_only=True))
 )

invokeai/backend/web/modules/parameters.py

@@ -4,18 +4,18 @@ from .parse_seed_weights import parse_seed_weights
 
 SAMPLER_CHOICES = [
     "ddim",
-    "k_dpm_2_a",
-    "k_dpm_2",
-    "k_dpmpp_2_a",
-    "k_dpmpp_2",
-    "k_euler_a",
-    "k_euler",
-    "k_heun",
     "k_lms",
     "plms",
-    # diffusers:
+    "k_heun",
-    "pndm",
+    "k_euler",
-    "unipc"
+    "euler_karras",
+    "k_euler_a",
+    "k_dpm_2",
+    "k_dpm_2_a",
+    "dpmpp_2s",
+    "k_dpmpp_2",
+    "k_dpmpp_2_karras",
+    "unipc",
 ]
 
 

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

@@ -2,15 +2,17 @@
 
 export const DIFFUSERS_SCHEDULERS: Array<string> = [
   'ddim',
-  'plms',
   'k_lms',
-  'dpmpp_2',
+  'plms',
+  'k_heun',
+  'k_euler',
+  'euler_karras',
+  'k_euler_a',
   'k_dpm_2',
   'k_dpm_2_a',
+  'dpmpp_2s',
   'k_dpmpp_2',
-  'k_euler',
+  'k_dpmpp_2_karras',
-  'k_euler_a',
-  'k_heun',
   'unipc',
 ];
 

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

@@ -47,15 +47,18 @@ export type CommonGeneratedImageMetadata = {
   postprocessing: null | Array<ESRGANMetadata | FacetoolMetadata>;
   sampler:
     | 'ddim'
-    | 'k_dpm_2_a'
-    | 'k_dpm_2'
-    | 'k_dpmpp_2_a'
-    | 'k_dpmpp_2'
-    | 'k_euler_a'
-    | 'k_euler'
-    | 'k_heun'
     | 'k_lms'
-    | 'plms';
+    | 'plms'
+    | 'k_heun'
+    | 'k_euler'
+    | 'euler_karras'
+    | 'k_euler_a'
+    | 'k_dpm_2'
+    | 'k_dpm_2_a'
+    | 'dpmpp_2s'
+    | 'k_dpmpp_2'
+    | 'k_dpmpp_2_karras'
+    | 'unipc';
   prompt: Prompt;
   seed: number;
   variations: SeedWeights;