Merge branch 'main' into release/invokeai-3-0-1

Author: Lincoln Stein

README.md

@@ -123,7 +123,7 @@ and go to http://localhost:9090.
 
 ### Command-Line Installation (for developers and users familiar with Terminals)
 
-You must have Python 3.9 or 3.10 installed on your machine. Earlier or
+You must have Python 3.9 through 3.11 installed on your machine. Earlier or
 later versions are not supported.
 Node.js also needs to be installed along with yarn (can be installed with
 the command `npm install -g yarn` if needed)

docs/installation/010_INSTALL_AUTOMATED.md

@@ -40,10 +40,8 @@ experimental versions later.
     this, open up a command-line window ("Terminal" on Linux and
     Macintosh, "Command" or "Powershell" on Windows) and type `python
     --version`. If Python is installed, it will print out the version
-    number. If it is version `3.9.*` or `3.10.*`, you meet
-    requirements. We do not recommend using Python 3.11 or higher,
-    as not all the libraries that InvokeAI depends on work properly
-    with this version.
+    number. If it is version `3.9.*`, `3.10.*` or `3.11.*` you meet
+    requirements.
 
     !!! warning "What to do if you have an unsupported version"
 

docs/installation/020_INSTALL_MANUAL.md

@@ -32,7 +32,7 @@ gaming):
 
 * **Python**
 
-    version 3.9 or 3.10 (3.11 is not recommended).
+    version 3.9 through 3.11
 
 * **CUDA Tools**
 
@@ -65,7 +65,7 @@ gaming):
 To install InvokeAI with virtual environments and the PIP package
 manager, please follow these steps:
 
-1.  Please make sure you are using Python 3.9 or 3.10. The rest of the install
+1.  Please make sure you are using Python 3.9 through 3.11. The rest of the install
     procedure depends on this and will not work with other versions:
 
     ```bash

installer/install.sh.in

@@ -9,16 +9,20 @@ cd $scriptdir
 function version { echo "$@" | awk -F. '{ printf("%d%03d%03d%03d\n", $1,$2,$3,$4); }'; }
 
 MINIMUM_PYTHON_VERSION=3.9.0
-MAXIMUM_PYTHON_VERSION=3.11.0
+MAXIMUM_PYTHON_VERSION=3.11.100
 PYTHON=""
-for candidate in python3.10 python3.9 python3 python ; do
+for candidate in python3.11 python3.10 python3.9 python3 python ; do
     if ppath=`which $candidate`; then
+        # when using `pyenv`, the executable for an inactive Python version will exist but will not be operational
+        # we check that this found executable can actually run
+        if [ $($candidate --version &>/dev/null; echo ${PIPESTATUS}) -gt 0 ]; then continue; fi
+
         python_version=$($ppath -V | awk '{ print $2 }')
         if [ $(version $python_version) -ge $(version "$MINIMUM_PYTHON_VERSION") ]; then
-	    if [ $(version $python_version) -lt $(version "$MAXIMUM_PYTHON_VERSION") ]; then
-		PYTHON=$ppath
-		break
-	    fi
+            if [ $(version $python_version) -le $(version "$MAXIMUM_PYTHON_VERSION") ]; then
+                PYTHON=$ppath
+                break
+            fi
         fi
     fi
 done

invokeai/app/api/routers/models.py

@@ -90,7 +90,7 @@ async def update_model(
                 new_name=info.model_name,
                 new_base=info.base_model,
             )
-            logger.info(f"Successfully renamed {base_model}/{model_name}=>{info.base_model}/{info.model_name}")
+            logger.info(f"Successfully renamed {base_model.value}/{model_name}=>{info.base_model}/{info.model_name}")
             # update information to support an update of attributes
             model_name = info.model_name
             base_model = info.base_model

invokeai/app/invocations/latent.py

@@ -12,7 +12,7 @@
 
 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
@@ -311,70 +311,71 @@ def prep_control_data(
 
     @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,
-                )
-                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)
+            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
 
-            scheduler = get_scheduler(
+            unet_info = context.services.model_manager.get_model(
+                **self.unet.unet.dict(),
                 context=context,
-                scheduler_info=self.unet.scheduler,
-                scheduler_name=self.scheduler,
             )
+            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)
 
-            pipeline = self.create_pipeline(unet, scheduler)
-            conditioning_data = self.get_conditioning_data(context, scheduler, unet)
+                scheduler = get_scheduler(
+                    context=context,
+                    scheduler_info=self.unet.scheduler,
+                    scheduler_name=self.scheduler,
+                )
 
-            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,
-            )
+                pipeline = self.create_pipeline(unet, scheduler)
+                conditioning_data = self.get_conditioning_data(context, scheduler, unet)
 
-            # 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,
-            )
+                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,
+                )
 
-        # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
-        result_latents = result_latents.to("cpu")
-        torch.cuda.empty_cache()
+                # 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,
+                )
 
-        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)
+            # 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)
+            return build_latents_output(latents_name=name, latents=result_latents)
 
 
 class LatentsToLatentsInvocation(TextToLatentsInvocation):
@@ -402,82 +403,83 @@ class Config(InvocationConfig):
 
     @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,
-                )
-                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)
+            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
 
-            scheduler = get_scheduler(
+            unet_info = context.services.model_manager.get_model(
+                **self.unet.unet.dict(),
                 context=context,
-                scheduler_info=self.unet.scheduler,
-                scheduler_name=self.scheduler,
             )
+            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)
 
-            pipeline = self.create_pipeline(unet, scheduler)
-            conditioning_data = self.get_conditioning_data(context, scheduler, unet)
+                scheduler = get_scheduler(
+                    context=context,
+                    scheduler_info=self.unet.scheduler,
+                    scheduler_name=self.scheduler,
+                )
 
-            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,
-            )
+                pipeline = self.create_pipeline(unet, scheduler)
+                conditioning_data = self.get_conditioning_data(context, scheduler, unet)
 
-            # 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)
-            )
+                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,
+                )
 
-            timesteps, _ = pipeline.get_img2img_timesteps(
-                self.steps,
-                self.strength,
-                device=unet.device,
-            )
+                # 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)
+                )
 
-            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,
-            )
+                timesteps, _ = pipeline.get_img2img_timesteps(
+                    self.steps,
+                    self.strength,
+                    device=unet.device,
+                )
 
-        # https://discuss.huggingface.co/t/memory-usage-by-later-pipeline-stages/23699
-        result_latents = result_latents.to("cpu")
-        torch.cuda.empty_cache()
+                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,
+                )
 
-        name = f"{context.graph_execution_state_id}__{self.id}"
-        context.services.latents.save(name, result_latents)
+            # 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)
         return build_latents_output(latents_name=name, latents=result_latents)
 
 
@@ -490,7 +492,7 @@ class LatentsToImageInvocation(BaseInvocation):
     # Inputs
     latents: Optional[LatentsField] = Field(description="The latents to generate an image from")
     vae: VaeField = Field(default=None, description="Vae submodel")
-    tiled: bool = Field(default=False, description="Decode latents by overlapping tiles(less memory consumption)")
+    tiled: bool = Field(default=False, description="Decode latents by overlaping tiles (less memory consumption)")
     fp32: bool = Field(DEFAULT_PRECISION == "float32", description="Decode in full precision")
     metadata: Optional[CoreMetadata] = Field(
         default=None, description="Optional core metadata to be written to the image"

invokeai/backend/model_management/model_manager.py

@@ -401,7 +401,11 @@ def create_key(
         base_model: BaseModelType,
         model_type: ModelType,
     ) -> str:
-        return f"{base_model}/{model_type}/{model_name}"
+        # In 3.11, the behavior of (str,enum) when interpolated into a
+        # string has changed. The next two lines are defensive.
+        base_model = BaseModelType(base_model)
+        model_type = ModelType(model_type)
+        return f"{base_model.value}/{model_type.value}/{model_name}"
 
     @classmethod
     def parse_key(cls, model_key: str) -> Tuple[str, BaseModelType, ModelType]: