make ip_adapters work with stable-fast

only compile model the first time :-)

probe for availability of stable-fast compiler and triton at startup time

simplify config logic

.github/workflows/lint-frontend.yml

@@ -21,16 +21,16 @@ jobs:
     if: github.event.pull_request.draft == false
     runs-on: ubuntu-22.04
     steps:
-      - name: Setup Node 20
+      - name: Setup Node 18
         uses: actions/setup-node@v4
         with:
-          node-version: '20'
+          node-version: '18'
       - name: Checkout
         uses: actions/checkout@v4
       - name: Setup pnpm
         uses: pnpm/action-setup@v2
         with:
-          version: 8
+          version: '8.12.1'
       - name: Install dependencies
         run: 'pnpm install --prefer-frozen-lockfile'
       - name: Typescript

.github/workflows/pypi-release.yml

@@ -1,13 +1,15 @@
 name: PyPI Release
 
 on:
-  push:
-    paths:
-      - 'invokeai/version/invokeai_version.py'
   workflow_dispatch:
+    inputs:
+      publish_package:
+        description: 'Publish build on PyPi? [true/false]'
+        required: true
+        default: 'false'
 
 jobs:
-  release:
+  build-and-release:
     if: github.repository == 'invoke-ai/InvokeAI'
     runs-on: ubuntu-22.04
     env:
@@ -15,33 +17,39 @@ jobs:
       TWINE_PASSWORD: ${{ secrets.PYPI_API_TOKEN }}
       TWINE_NON_INTERACTIVE: 1
     steps:
-      - name: Checkout sources
+      - name: Checkout
         uses: actions/checkout@v4
 
-      - name: Setup Node 20
+      - name: Setup Node 18
         uses: actions/setup-node@v4
         with:
-          node-version: '20'
+          node-version: '18'
 
       - name: Setup pnpm
         uses: pnpm/action-setup@v2
         with:
-          version: 8
+          version: '8.12.1'
 
-      - name: Install pnpm dependencies
+      - name: Install frontend dependencies
+        run: pnpm install --prefer-frozen-lockfile
         working-directory: invokeai/frontend/web
-        run: 'pnpm install --prefer-frozen-lockfile'
 
       - name: Build frontend
+        run: pnpm run build
         working-directory: invokeai/frontend/web
-        run: 'pnpm build'
 
-      - name: Install python deps
+      - name: Install python dependencies
         run: pip install --upgrade build twine
 
-      - name: Build wheel package
+      - name: Build python package
         run: python3 -m build
 
+      - name: Upload build as workflow artifact
+        uses: actions/upload-artifact@v4
+        with:
+          name: dist
+          path: dist
+
       - name: Check distribution
         run: twine check dist/*
 
@@ -54,6 +62,6 @@ jobs:
           EXISTS=scripts.pypi_helper.local_on_pypi(); \
           print(f'PACKAGE_EXISTS={EXISTS}')" >> $GITHUB_ENV
 
-      - name: Upload package
-        if: env.PACKAGE_EXISTS == 'False' && env.TWINE_PASSWORD != ''
+      - name: Publish build on PyPi
+        if: env.PACKAGE_EXISTS == 'False' && env.TWINE_PASSWORD != '' && github.event.inputs.publish_package == 'true'
         run: twine upload dist/*

docker/Dockerfile

@@ -59,14 +59,16 @@ RUN --mount=type=cache,target=/root/.cache/pip \
 
 # #### Build the Web UI ------------------------------------
 
-FROM node:18 AS web-builder
+FROM node:18-slim AS web-builder
+ENV PNPM_HOME="/pnpm"
+ENV PATH="$PNPM_HOME:$PATH"
+RUN corepack enable
+
 WORKDIR /build
 COPY invokeai/frontend/web/ ./
-RUN --mount=type=cache,target=/usr/lib/node_modules \
-    npm install --include dev
-RUN --mount=type=cache,target=/usr/lib/node_modules \
-    yarn vite build
-
+RUN --mount=type=cache,target=/pnpm/store \
+    pnpm install --frozen-lockfile
+RUN pnpm run build
 
 #### Runtime stage ---------------------------------------
 

docker/README.md

@@ -23,7 +23,7 @@ This is done via Docker Desktop preferences
 1. Make a copy of `env.sample` and name it `.env` (`cp env.sample .env` (Mac/Linux) or `copy example.env .env` (Windows)). Make changes as necessary. Set `INVOKEAI_ROOT` to an absolute path to:
     a. the desired location of the InvokeAI runtime directory, or
     b. an existing, v3.0.0 compatible runtime directory.
-1. `docker compose up`
+1. Execute `run.sh`
 
 The image will be built automatically if needed.
 
@@ -39,7 +39,7 @@ The Docker daemon on the system must be already set up to use the GPU. In case o
 
 ## Customize
 
-Check the `.env.sample` file. It contains some environment variables for running in Docker. Copy it, name it `.env`, and fill it in with your own values. Next time you run `docker compose up`, your custom values will be used.
+Check the `.env.sample` file. It contains some environment variables for running in Docker. Copy it, name it `.env`, and fill it in with your own values. Next time you run `run.sh`, your custom values will be used.
 
 You can also set these values in `docker-compose.yml` directly, but `.env` will help avoid conflicts when code is updated.
 

docker/build.sh

@@ -1,11 +0,0 @@
-#!/usr/bin/env bash
-set -e
-
-build_args=""
-
-[[ -f ".env" ]] && build_args=$(awk '$1 ~ /\=[^$]/ {print "--build-arg " $0 " "}' .env)
-
-echo "docker compose build args:"
-echo $build_args
-
-docker compose build $build_args

docker/docker-compose.yml

@@ -2,23 +2,8 @@
 
 version: '3.8'
 
-services:
-  invokeai:
+x-invokeai: &invokeai
     image: "local/invokeai:latest"
-    # edit below to run on a container runtime other than nvidia-container-runtime.
-    # not yet tested with rocm/AMD GPUs
-    # Comment out the "deploy" section to run on CPU only
-    deploy:
-      resources:
-        reservations:
-          devices:
-            - driver: nvidia
-              count: 1
-              capabilities: [gpu]
-    # For AMD support, comment out the deploy section above and uncomment the devices section below:
-    #devices:
-    #  - /dev/kfd:/dev/kfd
-    #  - /dev/dri:/dev/dri
     build:
       context: ..
       dockerfile: docker/Dockerfile
@@ -50,3 +35,27 @@ services:
     #   - |
     #     invokeai-model-install --yes --default-only --config_file ${INVOKEAI_ROOT}/config_custom.yaml
     #     invokeai-nodes-web --host 0.0.0.0
+
+services:
+  invokeai-nvidia:
+    <<: *invokeai
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              count: 1
+              capabilities: [gpu]
+
+  invokeai-cpu:
+    <<: *invokeai
+    profiles:
+      - cpu
+
+  invokeai-rocm:
+    <<: *invokeai
+    devices:
+      - /dev/kfd:/dev/kfd
+      - /dev/dri:/dev/dri
+    profiles:
+      - rocm

docker/run.sh

@@ -1,11 +1,28 @@
 #!/usr/bin/env bash
 set -e
 
-# This script is provided for backwards compatibility with the old docker setup.
-# it doesn't do much aside from wrapping the usual docker compose CLI.
+run() {
+  local scriptdir=$(dirname "${BASH_SOURCE[0]}")
+  cd "$scriptdir" || exit 1
 
-SCRIPTDIR=$(dirname "${BASH_SOURCE[0]}")
-cd "$SCRIPTDIR" || exit 1
+  local build_args=""
+  local profile=""
 
-docker compose up -d
-docker compose logs -f
+  [[ -f ".env" ]] &&
+    build_args=$(awk '$1 ~ /=[^$]/ && $0 !~ /^#/ {print "--build-arg " $0 " "}' .env) &&
+    profile="$(awk -F '=' '/GPU_DRIVER/ {print $2}' .env)"
+
+  local service_name="invokeai-$profile"
+
+  printf "%s\n" "docker compose build args:"
+  printf "%s\n" "$build_args"
+
+  docker compose build $build_args
+  unset build_args
+
+  printf "%s\n" "starting service $service_name"
+  docker compose --profile "$profile" up -d "$service_name"
+  docker compose logs -f
+}
+
+run

installer/create_installer.sh

@@ -91,9 +91,11 @@ rm -rf InvokeAI-Installer
 
 # copy content
 mkdir InvokeAI-Installer
-for f in templates lib *.txt *.reg; do
+for f in templates *.txt *.reg; do
     cp -r ${f} InvokeAI-Installer/
 done
+mkdir InvokeAI-Installer/lib
+cp lib/*.py InvokeAI-Installer/lib
 
 # Move the wheel
 mv dist/*.whl InvokeAI-Installer/lib/
@@ -111,6 +113,6 @@ cp WinLongPathsEnabled.reg InvokeAI-Installer/
 zip -r InvokeAI-installer-$VERSION.zip InvokeAI-Installer
 
 # clean up
-rm -rf InvokeAI-Installer tmp dist
+rm -rf InvokeAI-Installer tmp dist ../invokeai/frontend/web/dist/
 
 exit 0

invokeai/app/api/routers/model_records.py

@@ -26,7 +26,7 @@ from invokeai.backend.model_manager.config import (
 
 from ..dependencies import ApiDependencies
 
-model_records_router = APIRouter(prefix="/v1/model/record", tags=["model_manager_v2"])
+model_records_router = APIRouter(prefix="/v1/model/record", tags=["model_manager_v2_unstable"])
 
 
 class ModelsList(BaseModel):

invokeai/app/invocations/tiles.py

@@ -77,7 +77,7 @@ class CalculateImageTilesInvocation(BaseInvocation):
     title="Calculate Image Tiles Even Split",
     tags=["tiles"],
     category="tiles",
-    version="1.0.0",
+    version="1.1.0",
     classification=Classification.Beta,
 )
 class CalculateImageTilesEvenSplitInvocation(BaseInvocation):
@@ -97,11 +97,11 @@ class CalculateImageTilesEvenSplitInvocation(BaseInvocation):
         ge=1,
         description="Number of tiles to divide image into on the y axis",
     )
-    overlap_fraction: float = InputField(
-        default=0.25,
+    overlap: int = InputField(
+        default=128,
         ge=0,
-        lt=1,
-        description="Overlap between adjacent tiles as a fraction of the tile's dimensions (0-1)",
+        multiple_of=8,
+        description="The overlap, in pixels, between adjacent tiles.",
     )
 
     def invoke(self, context: InvocationContext) -> CalculateImageTilesOutput:
@@ -110,7 +110,7 @@ class CalculateImageTilesEvenSplitInvocation(BaseInvocation):
             image_width=self.image_width,
             num_tiles_x=self.num_tiles_x,
             num_tiles_y=self.num_tiles_y,
-            overlap_fraction=self.overlap_fraction,
+            overlap=self.overlap,
         )
         return CalculateImageTilesOutput(tiles=tiles)
 

invokeai/app/services/config/config_default.py

@@ -271,6 +271,7 @@ class InvokeAIAppConfig(InvokeAISettings):
     attention_slice_size: Literal["auto", "balanced", "max", 1, 2, 3, 4, 5, 6, 7, 8] = Field(default="auto", description='Slice size, valid when attention_type=="sliced"', json_schema_extra=Categories.Generation)
     force_tiled_decode  : bool = Field(default=False, description="Whether to enable tiled VAE decode (reduces memory consumption with some performance penalty)", json_schema_extra=Categories.Generation)
     png_compress_level  : int = Field(default=6, description="The compress_level setting of PIL.Image.save(), used for PNG encoding. All settings are lossless. 0 = fastest, largest filesize, 9 = slowest, smallest filesize", json_schema_extra=Categories.Generation)
+    stable_fast         : bool = Field(default=True, description="Enable stable-fast performance optimizations, if the library is installed and functional", json_schema_extra=Categories.Generation)
 
     # QUEUE
     max_queue_size      : int = Field(default=10000, gt=0, description="Maximum number of items in the session queue", json_schema_extra=Categories.Queue)

invokeai/app/services/shared/sqlite/sqlite_util.py

@@ -5,6 +5,7 @@ from invokeai.app.services.image_files.image_files_base import ImageFileStorageB
 from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_1 import build_migration_1
 from invokeai.app.services.shared.sqlite_migrator.migrations.migration_2 import build_migration_2
+from invokeai.app.services.shared.sqlite_migrator.migrations.migration_3 import build_migration_3
 from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_impl import SqliteMigrator
 
 
@@ -27,6 +28,7 @@ def init_db(config: InvokeAIAppConfig, logger: Logger, image_files: ImageFileSto
     migrator = SqliteMigrator(db=db)
     migrator.register_migration(build_migration_1())
     migrator.register_migration(build_migration_2(image_files=image_files, logger=logger))
+    migrator.register_migration(build_migration_3())
     migrator.run_migrations()
 
     return db

invokeai/app/services/shared/sqlite_migrator/migrations/migration_2.py

@@ -11,6 +11,8 @@ from invokeai.app.services.workflow_records.workflow_records_common import (
     UnsafeWorkflowWithVersionValidator,
 )
 
+from .util.migrate_yaml_config_1 import MigrateModelYamlToDb1
+
 
 class Migration2Callback:
     def __init__(self, image_files: ImageFileStorageBase, logger: Logger):
@@ -24,6 +26,7 @@ class Migration2Callback:
         self._add_workflow_library(cursor)
         self._drop_model_manager_metadata(cursor)
         self._recreate_model_config(cursor)
+        self._migrate_model_config_records(cursor)
         self._migrate_embedded_workflows(cursor)
 
     def _add_images_has_workflow(self, cursor: sqlite3.Cursor) -> None:
@@ -131,6 +134,11 @@ class Migration2Callback:
             """
         )
 
+    def _migrate_model_config_records(self, cursor: sqlite3.Cursor) -> None:
+        """After updating the model config table, we repopulate it."""
+        model_record_migrator = MigrateModelYamlToDb1(cursor)
+        model_record_migrator.migrate()
+
     def _migrate_embedded_workflows(self, cursor: sqlite3.Cursor) -> None:
         """
         In the v3.5.0 release, InvokeAI changed how it handles embedded workflows. The `images` table in

invokeai/app/services/shared/sqlite_migrator/migrations/migration_3.py

@@ -0,0 +1,75 @@
+import sqlite3
+
+from invokeai.app.services.shared.sqlite_migrator.sqlite_migrator_common import Migration
+
+from .util.migrate_yaml_config_1 import MigrateModelYamlToDb1
+
+
+class Migration3Callback:
+    def __init__(self) -> None:
+        pass
+
+    def __call__(self, cursor: sqlite3.Cursor) -> None:
+        self._drop_model_manager_metadata(cursor)
+        self._recreate_model_config(cursor)
+        self._migrate_model_config_records(cursor)
+
+    def _drop_model_manager_metadata(self, cursor: sqlite3.Cursor) -> None:
+        """Drops the `model_manager_metadata` table."""
+        cursor.execute("DROP TABLE IF EXISTS model_manager_metadata;")
+
+    def _recreate_model_config(self, cursor: sqlite3.Cursor) -> None:
+        """
+        Drops the `model_config` table, recreating it.
+
+        In 3.4.0, this table used explicit columns but was changed to use json_extract 3.5.0.
+
+        Because this table is not used in production, we are able to simply drop it and recreate it.
+        """
+
+        cursor.execute("DROP TABLE IF EXISTS model_config;")
+
+        cursor.execute(
+            """--sql
+            CREATE TABLE IF NOT EXISTS model_config (
+                id TEXT NOT NULL PRIMARY KEY,
+                -- The next 3 fields are enums in python, unrestricted string here
+                base TEXT GENERATED ALWAYS as (json_extract(config, '$.base')) VIRTUAL NOT NULL,
+                type TEXT GENERATED ALWAYS as (json_extract(config, '$.type')) VIRTUAL NOT NULL,
+                name TEXT GENERATED ALWAYS as (json_extract(config, '$.name')) VIRTUAL NOT NULL,
+                path TEXT GENERATED ALWAYS as (json_extract(config, '$.path')) VIRTUAL NOT NULL,
+                format TEXT GENERATED ALWAYS as (json_extract(config, '$.format')) VIRTUAL NOT NULL,
+                original_hash TEXT, -- could be null
+                -- Serialized JSON representation of the whole config object,
+                -- which will contain additional fields from subclasses
+                config TEXT NOT NULL,
+                created_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
+                -- Updated via trigger
+                updated_at DATETIME NOT NULL DEFAULT(STRFTIME('%Y-%m-%d %H:%M:%f', 'NOW')),
+                -- unique constraint on combo of name, base and type
+                UNIQUE(name, base, type)
+            );
+            """
+        )
+
+    def _migrate_model_config_records(self, cursor: sqlite3.Cursor) -> None:
+        """After updating the model config table, we repopulate it."""
+        model_record_migrator = MigrateModelYamlToDb1(cursor)
+        model_record_migrator.migrate()
+
+
+def build_migration_3() -> Migration:
+    """
+    Build the migration from database version 2 to 3.
+
+    This migration does the following:
+    - Drops the `model_config` table, recreating it
+    - Migrates data from `models.yaml` into the `model_config` table
+    """
+    migration_3 = Migration(
+        from_version=2,
+        to_version=3,
+        callback=Migration3Callback(),
+    )
+
+    return migration_3

invokeai/app/services/shared/sqlite_migrator/migrations/util/migrate_yaml_config_1.py

@@ -1,8 +1,12 @@
 # Copyright (c) 2023 Lincoln D. Stein
 """Migrate from the InvokeAI v2 models.yaml format to the v3 sqlite format."""
 
+import json
+import sqlite3
 from hashlib import sha1
 from logging import Logger
+from pathlib import Path
+from typing import Optional
 
 from omegaconf import DictConfig, OmegaConf
 from pydantic import TypeAdapter
@@ -10,13 +14,12 @@ from pydantic import TypeAdapter
 from invokeai.app.services.config import InvokeAIAppConfig
 from invokeai.app.services.model_records import (
     DuplicateModelException,
-    ModelRecordServiceSQL,
     UnknownModelException,
 )
-from invokeai.app.services.shared.sqlite.sqlite_database import SqliteDatabase
 from invokeai.backend.model_manager.config import (
     AnyModelConfig,
     BaseModelType,
+    ModelConfigFactory,
     ModelType,
 )
 from invokeai.backend.model_manager.hash import FastModelHash
@@ -25,9 +28,9 @@ from invokeai.backend.util.logging import InvokeAILogger
 ModelsValidator = TypeAdapter(AnyModelConfig)
 
 
-class MigrateModelYamlToDb:
+class MigrateModelYamlToDb1:
     """
-    Migrate the InvokeAI models.yaml format (VERSION 3.0.0) to SQL3 database format (VERSION 3.2.0)
+    Migrate the InvokeAI models.yaml format (VERSION 3.0.0) to SQL3 database format (VERSION 3.5.0).
 
     The class has one externally useful method, migrate(), which scans the
     currently models.yaml file and imports all its entries into invokeai.db.
@@ -41,17 +44,13 @@ class MigrateModelYamlToDb:
 
     config: InvokeAIAppConfig
     logger: Logger
+    cursor: sqlite3.Cursor
 
-    def __init__(self) -> None:
+    def __init__(self, cursor: sqlite3.Cursor = None) -> None:
         self.config = InvokeAIAppConfig.get_config()
         self.config.parse_args()
         self.logger = InvokeAILogger.get_logger()
-
-    def get_db(self) -> ModelRecordServiceSQL:
-        """Fetch the sqlite3 database for this installation."""
-        db_path = None if self.config.use_memory_db else self.config.db_path
-        db = SqliteDatabase(db_path=db_path, logger=self.logger, verbose=self.config.log_sql)
-        return ModelRecordServiceSQL(db)
+        self.cursor = cursor
 
     def get_yaml(self) -> DictConfig:
         """Fetch the models.yaml DictConfig for this installation."""
@@ -62,8 +61,10 @@ class MigrateModelYamlToDb:
 
     def migrate(self) -> None:
         """Do the migration from models.yaml to invokeai.db."""
-        db = self.get_db()
-        yaml = self.get_yaml()
+        try:
+            yaml = self.get_yaml()
+        except OSError:
+            return
 
         for model_key, stanza in yaml.items():
             if model_key == "__metadata__":
@@ -86,22 +87,62 @@ class MigrateModelYamlToDb:
             new_config: AnyModelConfig = ModelsValidator.validate_python(stanza)  # type: ignore # see https://github.com/pydantic/pydantic/discussions/7094
 
             try:
-                if original_record := db.search_by_path(stanza.path):
-                    key = original_record[0].key
+                if original_record := self._search_by_path(stanza.path):
+                    key = original_record.key
                     self.logger.info(f"Updating model {model_name} with information from models.yaml using key {key}")
-                    db.update_model(key, new_config)
+                    self._update_model(key, new_config)
                 else:
                     self.logger.info(f"Adding model {model_name} with key {model_key}")
-                    db.add_model(new_key, new_config)
+                    self._add_model(new_key, new_config)
             except DuplicateModelException:
                 self.logger.warning(f"Model {model_name} is already in the database")
             except UnknownModelException:
                 self.logger.warning(f"Model at {stanza.path} could not be found in database")
 
+    def _search_by_path(self, path: Path) -> Optional[AnyModelConfig]:
+        self.cursor.execute(
+            """--sql
+            SELECT config FROM model_config
+            WHERE path=?;
+            """,
+            (str(path),),
+        )
+        results = [ModelConfigFactory.make_config(json.loads(x[0])) for x in self.cursor.fetchall()]
+        return results[0] if results else None
 
-def main():
-    MigrateModelYamlToDb().migrate()
+    def _update_model(self, key: str, config: AnyModelConfig) -> None:
+        record = ModelConfigFactory.make_config(config, key=key)  # ensure it is a valid config obect
+        json_serialized = record.model_dump_json()  # and turn it into a json string.
+        self.cursor.execute(
+            """--sql
+            UPDATE model_config
+            SET
+                config=?
+            WHERE id=?;
+            """,
+            (json_serialized, key),
+        )
+        if self.cursor.rowcount == 0:
+            raise UnknownModelException("model not found")
 
-
-if __name__ == "__main__":
-    main()
+    def _add_model(self, key: str, config: AnyModelConfig) -> None:
+        record = ModelConfigFactory.make_config(config, key=key)  # ensure it is a valid config obect.
+        json_serialized = record.model_dump_json()  # and turn it into a json string.
+        try:
+            self.cursor.execute(
+                """--sql
+                INSERT INTO model_config (
+                   id,
+                   original_hash,
+                   config
+                  )
+                VALUES (?,?,?);
+                """,
+                (
+                    key,
+                    record.original_hash,
+                    json_serialized,
+                ),
+            )
+        except sqlite3.IntegrityError as exc:
+            raise DuplicateModelException(f"{record.name}: model is already in database") from exc

invokeai/backend/ip_adapter/attention_processor.py

@@ -141,7 +141,6 @@ class IPAttnProcessor2_0(torch.nn.Module):
                 ip_hidden_states = ipa_embed
 
                 # Expected ip_hidden_state shape: (batch_size, num_ip_images, ip_seq_len, ip_image_embedding)
-
                 ip_key = ipa_weights.to_k_ip(ip_hidden_states)
                 ip_value = ipa_weights.to_v_ip(ip_hidden_states)
 

invokeai/backend/ip_adapter/ip_attention_weights.py

@@ -12,6 +12,8 @@ class IPAttentionProcessorWeights(torch.nn.Module):
         super().__init__()
         self.to_k_ip = torch.nn.Linear(in_dim, out_dim, bias=False)
         self.to_v_ip = torch.nn.Linear(in_dim, out_dim, bias=False)
+        for param in self.parameters():
+            param.requires_grad = False
 
 
 class IPAttentionWeights(torch.nn.Module):

invokeai/backend/model_management/model_cache.py

@@ -24,12 +24,14 @@ import sys
 import time
 from contextlib import suppress
 from dataclasses import dataclass, field
+from importlib.util import find_spec
 from pathlib import Path
 from typing import Any, Dict, Optional, Type, Union, types
 
 import torch
 
 import invokeai.backend.util.logging as logger
+from invokeai.app.services.config import InvokeAIAppConfig
 from invokeai.backend.model_management.memory_snapshot import MemorySnapshot, get_pretty_snapshot_diff
 from invokeai.backend.model_management.model_load_optimizations import skip_torch_weight_init
 
@@ -39,6 +41,26 @@ from .models import BaseModelType, ModelBase, ModelType, SubModelType
 if choose_torch_device() == torch.device("mps"):
     from torch import mps
 
+SFAST_AVAILABLE = False
+TRITON_AVAILABLE = False
+XFORMERS_AVAILABLE = False
+SFAST_CONFIG = None
+
+TRITON_AVAILABLE = find_spec("triton") is not None
+XFORMERS_AVAILABLE = find_spec("xformers") is not None
+
+try:
+    from sfast.compilers.diffusion_pipeline_compiler import CompilationConfig, compile_unet, compile_vae
+
+    SFAST_CONFIG = CompilationConfig.Default()
+    SFAST_CONFIG.enable_cuda_graph = True
+    SFAST_CONFIG.enable_xformers = XFORMERS_AVAILABLE
+    SFAST_CONFIG.enable_triton = TRITON_AVAILABLE
+    SFAST_AVAILABLE = True
+except ImportError:
+    pass
+
+
 # Maximum size of the cache, in gigs
 # Default is roughly enough to hold three fp16 diffusers models in RAM simultaneously
 DEFAULT_MAX_CACHE_SIZE = 6.0
@@ -110,6 +132,7 @@ class _CacheRecord:
 class ModelCache(object):
     def __init__(
         self,
+        app_config: InvokeAIAppConfig,
         max_cache_size: float = DEFAULT_MAX_CACHE_SIZE,
         max_vram_cache_size: float = DEFAULT_MAX_VRAM_CACHE_SIZE,
         execution_device: torch.device = torch.device("cuda"),
@@ -122,6 +145,7 @@ class ModelCache(object):
         log_memory_usage: bool = False,
     ):
         """
+        :param app_config: InvokeAIAppConfig for application
         :param max_cache_size: Maximum size of the RAM cache [6.0 GB]
         :param execution_device: Torch device to load active model into [torch.device('cuda')]
         :param storage_device: Torch device to save inactive model in [torch.device('cpu')]
@@ -135,6 +159,7 @@ class ModelCache(object):
             behaviour.
         """
         self.model_infos: Dict[str, ModelBase] = {}
+        self.app_config = app_config
         # allow lazy offloading only when vram cache enabled
         self.lazy_offloading = lazy_offloading and max_vram_cache_size > 0
         self.precision: torch.dtype = precision
@@ -239,6 +264,9 @@ class ModelCache(object):
             snapshot_before = self._capture_memory_snapshot()
             with skip_torch_weight_init():
                 model = model_info.get_model(child_type=submodel, torch_dtype=self.precision)
+            if SFAST_AVAILABLE and self.app_config.stable_fast and submodel:
+                model = self._compile_model(model, submodel)
+
             snapshot_after = self._capture_memory_snapshot()
             end_load_time = time.time()
 
@@ -322,6 +350,16 @@ class ModelCache(object):
                     f"{get_pretty_snapshot_diff(snapshot_before, snapshot_after)}"
                 )
 
+    def _compile_model(self, model: Any, model_type: SubModelType) -> Any:
+        if model_type == SubModelType("unet"):
+            self.logger.info("SFast-compiling unet model")
+            return compile_unet(model, SFAST_CONFIG)
+        elif model_type == SubModelType("vae"):
+            self.logger.info("SFast-compiling vae model")
+            return compile_vae(model, SFAST_CONFIG)
+        else:
+            return model
+
     class ModelLocker(object):
         def __init__(self, cache, key, model, gpu_load, size_needed):
             """

invokeai/backend/model_management/model_manager.py

@@ -344,6 +344,7 @@ class ModelManager(object):
         self.app_config = InvokeAIAppConfig.get_config()
         self.logger = logger
         self.cache = ModelCache(
+            app_config=self.app_config,
             max_cache_size=max_cache_size,
             max_vram_cache_size=self.app_config.vram_cache_size,
             lazy_offloading=self.app_config.lazy_offload,

invokeai/backend/model_management/model_probe.py

@@ -389,7 +389,7 @@ class TextualInversionCheckpointProbe(CheckpointProbeBase):
         elif "clip_g" in checkpoint:
             token_dim = checkpoint["clip_g"].shape[-1]
         else:
-            token_dim = list(checkpoint.values())[0].shape[0]
+            token_dim = list(checkpoint.values())[0].shape[-1]
         if token_dim == 768:
             return BaseModelType.StableDiffusion1
         elif token_dim == 1024:

invokeai/backend/model_management/util.py

@@ -9,7 +9,7 @@ def lora_token_vector_length(checkpoint: dict) -> int:
     :param checkpoint: The checkpoint
     """
 
-    def _get_shape_1(key, tensor, checkpoint):
+    def _get_shape_1(key: str, tensor, checkpoint) -> int:
         lora_token_vector_length = None
 
         if "." not in key:
@@ -57,6 +57,10 @@ def lora_token_vector_length(checkpoint: dict) -> int:
     for key, tensor in checkpoint.items():
         if key.startswith("lora_unet_") and ("_attn2_to_k." in key or "_attn2_to_v." in key):
             lora_token_vector_length = _get_shape_1(key, tensor, checkpoint)
+        elif key.startswith("lora_unet_") and (
+            "time_emb_proj.lora_down" in key
+        ):  # recognizes format at https://civitai.com/models/224641
+            lora_token_vector_length = _get_shape_1(key, tensor, checkpoint)
         elif key.startswith("lora_te") and "_self_attn_" in key:
             tmp_length = _get_shape_1(key, tensor, checkpoint)
             if key.startswith("lora_te_"):

invokeai/backend/model_manager/probe.py

@@ -400,6 +400,8 @@ class LoRACheckpointProbe(CheckpointProbeBase):
             return BaseModelType.StableDiffusion1
         elif token_vector_length == 1024:
             return BaseModelType.StableDiffusion2
+        elif token_vector_length == 1280:
+            return BaseModelType.StableDiffusionXL  # recognizes format at https://civitai.com/models/224641
         elif token_vector_length == 2048:
             return BaseModelType.StableDiffusionXL
         else:

invokeai/backend/tiles/tiles.py

@@ -102,7 +102,7 @@ def calc_tiles_with_overlap(
 
 
 def calc_tiles_even_split(
-    image_height: int, image_width: int, num_tiles_x: int, num_tiles_y: int, overlap_fraction: float = 0
+    image_height: int, image_width: int, num_tiles_x: int, num_tiles_y: int, overlap: int = 0
 ) -> list[Tile]:
     """Calculate the tile coordinates for a given image shape with the number of tiles requested.
 
@@ -111,31 +111,35 @@ def calc_tiles_even_split(
         image_width (int): The image width in px.
         num_x_tiles (int): The number of tile to split the image into on the X-axis.
         num_y_tiles (int): The number of tile to split the image into on the Y-axis.
-        overlap_fraction (float, optional): The target overlap as fraction of the tiles size. Defaults to 0.
+        overlap (int, optional): The overlap between adjacent tiles in pixels. Defaults to 0.
 
     Returns:
         list[Tile]: A list of tiles that cover the image shape. Ordered from left-to-right, top-to-bottom.
     """
-
-    # Ensure tile size is divisible by 8
+    # Ensure the image is divisible by LATENT_SCALE_FACTOR
     if image_width % LATENT_SCALE_FACTOR != 0 or image_height % LATENT_SCALE_FACTOR != 0:
         raise ValueError(f"image size (({image_width}, {image_height})) must be divisible by {LATENT_SCALE_FACTOR}")
 
-    # Calculate the overlap size based on the percentage and adjust it to be divisible by 8 (rounding up)
-    overlap_x = LATENT_SCALE_FACTOR * math.ceil(
-        int((image_width / num_tiles_x) * overlap_fraction) / LATENT_SCALE_FACTOR
-    )
-    overlap_y = LATENT_SCALE_FACTOR * math.ceil(
-        int((image_height / num_tiles_y) * overlap_fraction) / LATENT_SCALE_FACTOR
-    )
-
     # Calculate the tile size based on the number of tiles and overlap, and ensure it's divisible by 8 (rounding down)
-    tile_size_x = LATENT_SCALE_FACTOR * math.floor(
-        ((image_width + overlap_x * (num_tiles_x - 1)) // num_tiles_x) / LATENT_SCALE_FACTOR
-    )
-    tile_size_y = LATENT_SCALE_FACTOR * math.floor(
-        ((image_height + overlap_y * (num_tiles_y - 1)) // num_tiles_y) / LATENT_SCALE_FACTOR
-    )
+    if num_tiles_x > 1:
+        # ensure the overlap is not more than the maximum overlap if we only have 1 tile then we dont care about overlap
+        assert overlap <= image_width - (LATENT_SCALE_FACTOR * (num_tiles_x - 1))
+        tile_size_x = LATENT_SCALE_FACTOR * math.floor(
+            ((image_width + overlap * (num_tiles_x - 1)) // num_tiles_x) / LATENT_SCALE_FACTOR
+        )
+        assert overlap < tile_size_x
+    else:
+        tile_size_x = image_width
+
+    if num_tiles_y > 1:
+        # ensure the overlap is not more than the maximum overlap if we only have 1 tile then we dont care about overlap
+        assert overlap <= image_height - (LATENT_SCALE_FACTOR * (num_tiles_y - 1))
+        tile_size_y = LATENT_SCALE_FACTOR * math.floor(
+            ((image_height + overlap * (num_tiles_y - 1)) // num_tiles_y) / LATENT_SCALE_FACTOR
+        )
+        assert overlap < tile_size_y
+    else:
+        tile_size_y = image_height
 
     # tiles[y * num_tiles_x + x] is the tile for the y'th row, x'th column.
     tiles: list[Tile] = []
@@ -143,7 +147,7 @@ def calc_tiles_even_split(
     # Calculate tile coordinates. (Ignore overlap values for now.)
     for tile_idx_y in range(num_tiles_y):
         # Calculate the top and bottom of the row
-        top = tile_idx_y * (tile_size_y - overlap_y)
+        top = tile_idx_y * (tile_size_y - overlap)
         bottom = min(top + tile_size_y, image_height)
         # For the last row adjust bottom to be the height of the image
         if tile_idx_y == num_tiles_y - 1:
@@ -151,7 +155,7 @@ def calc_tiles_even_split(
 
         for tile_idx_x in range(num_tiles_x):
             # Calculate the left & right coordinate of each tile
-            left = tile_idx_x * (tile_size_x - overlap_x)
+            left = tile_idx_x * (tile_size_x - overlap)
             right = min(left + tile_size_x, image_width)
             # For the last tile in the row adjust right to be the width of the image
             if tile_idx_x == num_tiles_x - 1:

invokeai/frontend/install/invokeai_update.py

@@ -4,6 +4,7 @@ pip install <path_to_git_source>.
 """
 import os
 import platform
+from distutils.version import LooseVersion
 
 import pkg_resources
 import psutil
@@ -31,10 +32,6 @@ else:
     console = Console(style=Style(color="grey74", bgcolor="grey19"))
 
 
-def get_versions() -> dict:
-    return requests.get(url=INVOKE_AI_REL).json()
-
-
 def invokeai_is_running() -> bool:
     for p in psutil.process_iter():
         try:
@@ -50,6 +47,20 @@ def invokeai_is_running() -> bool:
     return False
 
 
+def get_pypi_versions():
+    url = "https://pypi.org/pypi/invokeai/json"
+    try:
+        data = requests.get(url).json()
+    except Exception:
+        raise Exception("Unable to fetch version information from PyPi")
+
+    versions = list(data["releases"].keys())
+    versions.sort(key=LooseVersion, reverse=True)
+    latest_version = [v for v in versions if "rc" not in v][0]
+    latest_release_candidate = [v for v in versions if "rc" in v][0]
+    return latest_version, latest_release_candidate, versions
+
+
 def welcome(latest_release: str, latest_prerelease: str):
     @group()
     def text():
@@ -63,8 +74,7 @@ def welcome(latest_release: str, latest_prerelease: str):
         yield "[bold yellow]Options:"
         yield f"""[1] Update to the latest [bold]official release[/bold] ([italic]{latest_release}[/italic])
 [2] Update to the latest [bold]pre-release[/bold] (may be buggy; caveat emptor!) ([italic]{latest_prerelease}[/italic])
-[3] Manually enter the [bold]tag name[/bold] for the version you wish to update to
-[4] Manually enter the [bold]branch name[/bold] for the version you wish to update to"""
+[3] Manually enter the [bold]version[/bold] you wish to update to"""
 
     console.rule()
     print(
@@ -92,44 +102,35 @@ def get_extras():
 
 
 def main():
-    versions = get_versions()
-    released_versions = [x for x in versions if not (x["draft"] or x["prerelease"])]
-    prerelease_versions = [x for x in versions if not x["draft"] and x["prerelease"]]
-    latest_release = released_versions[0]["tag_name"] if len(released_versions) else None
-    latest_prerelease = prerelease_versions[0]["tag_name"] if len(prerelease_versions) else None
-
     if invokeai_is_running():
         print(":exclamation: [bold red]Please terminate all running instances of InvokeAI before updating.[/red bold]")
         input("Press any key to continue...")
         return
 
+    latest_release, latest_prerelease, versions = get_pypi_versions()
+
     welcome(latest_release, latest_prerelease)
 
-    tag = None
-    branch = None
-    release = None
-    choice = Prompt.ask("Choice:", choices=["1", "2", "3", "4"], default="1")
+    release = latest_release
+    choice = Prompt.ask("Choice:", choices=["1", "2", "3"], default="1")
 
     if choice == "1":
         release = latest_release
     elif choice == "2":
         release = latest_prerelease
     elif choice == "3":
-        while not tag:
-            tag = Prompt.ask("Enter an InvokeAI tag name")
-    elif choice == "4":
-        while not branch:
-            branch = Prompt.ask("Enter an InvokeAI branch name")
+        while True:
+            release = Prompt.ask("Enter an InvokeAI version")
+            release.strip()
+            if release in versions:
+                break
+            print(f":exclamation: [bold red]'{release}' is not a recognized InvokeAI release.[/red bold]")
 
     extras = get_extras()
 
-    print(f":crossed_fingers: Upgrading to [yellow]{tag or release or branch}[/yellow]")
-    if release:
-        cmd = f'pip install "invokeai{extras} @ {INVOKE_AI_SRC}/{release}.zip" --use-pep517 --upgrade'
-    elif tag:
-        cmd = f'pip install "invokeai{extras} @ {INVOKE_AI_TAG}/{tag}.zip" --use-pep517 --upgrade'
-    else:
-        cmd = f'pip install "invokeai{extras} @ {INVOKE_AI_BRANCH}/{branch}.zip" --use-pep517 --upgrade'
+    print(f":crossed_fingers: Upgrading to [yellow]{release}[/yellow]")
+    cmd = f'pip install "invokeai{extras}=={release}" --use-pep517 --upgrade'
+
     print("")
     print("")
     if os.system(cmd) == 0:

invokeai/frontend/web/public/locales/es.json

@@ -727,9 +727,6 @@
         "showMinimapnodes": "Mostrar el minimapa",
         "reloadNodeTemplates": "Recargar las plantillas de nodos",
         "loadWorkflow": "Cargar el flujo de trabajo",
-        "resetWorkflow": "Reiniciar e flujo de trabajo",
-        "resetWorkflowDesc": "¿Está seguro de que deseas restablecer este flujo de trabajo?",
-        "resetWorkflowDesc2": "Al reiniciar el flujo de trabajo se borrarán todos los nodos, aristas y detalles del flujo de trabajo.",
         "downloadWorkflow": "Descargar el flujo de trabajo en un archivo JSON"
     }
 }

invokeai/frontend/web/public/locales/it.json

@@ -898,11 +898,8 @@
         "zoomInNodes": "Ingrandire",
         "fitViewportNodes": "Adatta vista",
         "showGraphNodes": "Mostra sovrapposizione grafico",
-        "resetWorkflowDesc2": "Il ripristino dell'editor del flusso di lavoro cancellerà tutti i nodi, le connessioni e i dettagli del flusso di lavoro. I flussi di lavoro salvati non saranno interessati.",
         "reloadNodeTemplates": "Ricarica i modelli di nodo",
         "loadWorkflow": "Importa flusso di lavoro JSON",
-        "resetWorkflow": "Reimposta l'editor del flusso di lavoro",
-        "resetWorkflowDesc": "Sei sicuro di voler reimpostare l'editor del flusso di lavoro?",
         "downloadWorkflow": "Esporta flusso di lavoro JSON",
         "scheduler": "Campionatore",
         "addNode": "Aggiungi nodo",
@@ -1112,7 +1109,10 @@
         "deletedInvalidEdge": "Eliminata connessione non valida {{source}} -> {{target}}",
         "unknownInput": "Input sconosciuto: {{name}}",
         "prototypeDesc": "Questa invocazione è un prototipo. Potrebbe subire modifiche sostanziali durante gli aggiornamenti dell'app e potrebbe essere rimossa in qualsiasi momento.",
-        "betaDesc": "Questa invocazione è in versione beta. Fino a quando non sarà stabile, potrebbe subire modifiche importanti durante gli aggiornamenti dell'app. Abbiamo intenzione di supportare questa invocazione a lungo termine."
+        "betaDesc": "Questa invocazione è in versione beta. Fino a quando non sarà stabile, potrebbe subire modifiche importanti durante gli aggiornamenti dell'app. Abbiamo intenzione di supportare questa invocazione a lungo termine.",
+        "newWorkflow": "Nuovo flusso di lavoro",
+        "newWorkflowDesc": "Creare un nuovo flusso di lavoro?",
+        "newWorkflowDesc2": "Il flusso di lavoro attuale presenta modifiche non salvate."
     },
     "boards": {
         "autoAddBoard": "Aggiungi automaticamente bacheca",
@@ -1619,7 +1619,6 @@
         "saveWorkflow": "Salva flusso di lavoro",
         "openWorkflow": "Apri flusso di lavoro",
         "clearWorkflowSearchFilter": "Cancella il filtro di ricerca del flusso di lavoro",
-        "workflowEditorReset": "Reimpostazione dell'editor del flusso di lavoro",
         "workflowLibrary": "Libreria",
         "noRecentWorkflows": "Nessun flusso di lavoro recente",
         "workflowSaved": "Flusso di lavoro salvato",
@@ -1633,7 +1632,10 @@
         "deleteWorkflow": "Elimina flusso di lavoro",
         "workflows": "Flussi di lavoro",
         "noDescription": "Nessuna descrizione",
-        "userWorkflows": "I miei flussi di lavoro"
+        "userWorkflows": "I miei flussi di lavoro",
+        "newWorkflowCreated": "Nuovo flusso di lavoro creato",
+        "downloadWorkflow": "Salva su file",
+        "uploadWorkflow": "Carica da file"
     },
     "app": {
         "storeNotInitialized": "Il negozio non è inizializzato"

invokeai/frontend/web/public/locales/nl.json

@@ -844,9 +844,6 @@
         "hideLegendNodes": "Typelegende veld verbergen",
         "reloadNodeTemplates": "Herlaad knooppuntsjablonen",
         "loadWorkflow": "Laad werkstroom",
-        "resetWorkflow": "Herstel werkstroom",
-        "resetWorkflowDesc": "Weet je zeker dat je deze werkstroom wilt herstellen?",
-        "resetWorkflowDesc2": "Herstel van een werkstroom haalt alle knooppunten, randen en werkstroomdetails weg.",
         "downloadWorkflow": "Download JSON van werkstroom",
         "booleanPolymorphicDescription": "Een verzameling Booleanse waarden.",
         "scheduler": "Planner",

invokeai/frontend/web/public/locales/ru.json

@@ -909,9 +909,6 @@
         "hideLegendNodes": "Скрыть тип поля",
         "showMinimapnodes": "Показать миникарту",
         "loadWorkflow": "Загрузить рабочий процесс",
-        "resetWorkflowDesc2": "Сброс рабочего процесса очистит все узлы, ребра и детали рабочего процесса.",
-        "resetWorkflow": "Сбросить рабочий процесс",
-        "resetWorkflowDesc": "Вы уверены, что хотите сбросить этот рабочий процесс?",
         "reloadNodeTemplates": "Перезагрузить шаблоны узлов",
         "downloadWorkflow": "Скачать JSON рабочего процесса",
         "booleanPolymorphicDescription": "Коллекция логических значений.",
@@ -1599,7 +1596,6 @@
         "saveWorkflow": "Сохранить рабочий процесс",
         "openWorkflow": "Открытый рабочий процесс",
         "clearWorkflowSearchFilter": "Очистить фильтр поиска рабочих процессов",
-        "workflowEditorReset": "Сброс редактора рабочих процессов",
         "workflowLibrary": "Библиотека",
         "downloadWorkflow": "Скачать рабочий процесс",
         "noRecentWorkflows": "Нет недавних рабочих процессов",

invokeai/frontend/web/public/locales/zh_CN.json

@@ -892,11 +892,8 @@
     },
     "nodes": {
         "zoomInNodes": "放大",
-        "resetWorkflowDesc": "是否确定要重置工作流编辑器?",
-        "resetWorkflow": "重置工作流编辑器",
         "loadWorkflow": "加载工作流",
         "zoomOutNodes": "缩小",
-        "resetWorkflowDesc2": "重置工作流编辑器将清除所有节点、边际和节点图详情。不影响已保存的工作流。",
         "reloadNodeTemplates": "重载节点模板",
         "hideGraphNodes": "隐藏节点图信息",
         "fitViewportNodes": "自适应视图",
@@ -1122,7 +1119,10 @@
         "deletedInvalidEdge": "已删除无效的边缘 {{source}} -> {{target}}",
         "unknownInput": "未知输入：{{name}}",
         "prototypeDesc": "此调用是一个原型 (prototype)。它可能会在本项目更新期间发生破坏性更改，并且随时可能被删除。",
-        "betaDesc": "此调用尚处于测试阶段。在稳定之前，它可能会在项目更新期间发生破坏性更改。本项目计划长期支持这种调用。"
+        "betaDesc": "此调用尚处于测试阶段。在稳定之前，它可能会在项目更新期间发生破坏性更改。本项目计划长期支持这种调用。",
+        "newWorkflow": "新建工作流",
+        "newWorkflowDesc": "是否创建一个新的工作流？",
+        "newWorkflowDesc2": "当前工作流有未保存的更改。"
     },
     "controlnet": {
         "resize": "直接缩放",
@@ -1637,9 +1637,8 @@
         "saveWorkflow": "保存工作流",
         "openWorkflow": "打开工作流",
         "clearWorkflowSearchFilter": "清除工作流检索过滤器",
-        "workflowEditorReset": "工作流编辑器重置",
         "workflowLibrary": "工作流库",
-        "downloadWorkflow": "下载工作流",
+        "downloadWorkflow": "保存到文件",
         "noRecentWorkflows": "无最近工作流",
         "workflowSaved": "已保存工作流",
         "workflowIsOpen": "工作流已打开",
@@ -1652,8 +1651,9 @@
         "deleteWorkflow": "删除工作流",
         "workflows": "工作流",
         "noDescription": "无描述",
-        "uploadWorkflow": "上传工作流",
-        "userWorkflows": "我的工作流"
+        "uploadWorkflow": "从文件中加载",
+        "userWorkflows": "我的工作流",
+        "newWorkflowCreated": "已创建新的工作流"
     },
     "app": {
         "storeNotInitialized": "商店尚未初始化"

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

@@ -34,6 +34,7 @@ import { actionSanitizer } from './middleware/devtools/actionSanitizer';
 import { actionsDenylist } from './middleware/devtools/actionsDenylist';
 import { stateSanitizer } from './middleware/devtools/stateSanitizer';
 import { listenerMiddleware } from './middleware/listenerMiddleware';
+import { authToastMiddleware } from 'services/api/authToastMiddleware';
 
 const allReducers = {
   canvas: canvasReducer,
@@ -96,6 +97,7 @@ export const createStore = (uniqueStoreKey?: string, persist = true) =>
       })
         .concat(api.middleware)
         .concat(dynamicMiddlewares)
+        .concat(authToastMiddleware)
         .prepend(listenerMiddleware.middleware),
     enhancers: (getDefaultEnhancers) => {
       const _enhancers = getDefaultEnhancers().concat(autoBatchEnhancer());

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasImageToImageGraph.ts

@@ -144,6 +144,7 @@ export const buildCanvasImageToImageGraph = (
         type: 'l2i',
         id: CANVAS_OUTPUT,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [
@@ -255,6 +256,7 @@ export const buildCanvasImageToImageGraph = (
       is_intermediate,
       width: width,
       height: height,
+      use_cache: false,
     };
 
     graph.edges.push(
@@ -295,6 +297,7 @@ export const buildCanvasImageToImageGraph = (
       id: CANVAS_OUTPUT,
       is_intermediate,
       fp32,
+      use_cache: false,
     };
 
     (graph.nodes[IMAGE_TO_LATENTS] as ImageToLatentsInvocation).image =

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasInpaintGraph.ts

@@ -191,6 +191,7 @@ export const buildCanvasInpaintGraph = (
         id: CANVAS_OUTPUT,
         is_intermediate,
         reference: canvasInitImage,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasOutpaintGraph.ts

@@ -199,6 +199,7 @@ export const buildCanvasOutpaintGraph = (
         type: 'color_correct',
         id: CANVAS_OUTPUT,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasSDXLImageToImageGraph.ts

@@ -266,6 +266,7 @@ export const buildCanvasSDXLImageToImageGraph = (
       is_intermediate,
       width: width,
       height: height,
+      use_cache: false,
     };
 
     graph.edges.push(
@@ -306,6 +307,7 @@ export const buildCanvasSDXLImageToImageGraph = (
       id: CANVAS_OUTPUT,
       is_intermediate,
       fp32,
+      use_cache: false,
     };
 
     (graph.nodes[IMAGE_TO_LATENTS] as ImageToLatentsInvocation).image =

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasSDXLInpaintGraph.ts

@@ -196,6 +196,7 @@ export const buildCanvasSDXLInpaintGraph = (
         id: CANVAS_OUTPUT,
         is_intermediate,
         reference: canvasInitImage,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasSDXLOutpaintGraph.ts

@@ -204,6 +204,7 @@ export const buildCanvasSDXLOutpaintGraph = (
         type: 'color_correct',
         id: CANVAS_OUTPUT,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasSDXLTextToImageGraph.ts

@@ -258,6 +258,7 @@ export const buildCanvasSDXLTextToImageGraph = (
       is_intermediate,
       width: width,
       height: height,
+      use_cache: false,
     };
 
     graph.edges.push(
@@ -288,6 +289,7 @@ export const buildCanvasSDXLTextToImageGraph = (
       id: CANVAS_OUTPUT,
       is_intermediate,
       fp32,
+      use_cache: false,
     };
 
     graph.edges.push({

invokeai/frontend/web/src/features/nodes/util/graph/buildCanvasTextToImageGraph.ts

@@ -246,6 +246,7 @@ export const buildCanvasTextToImageGraph = (
       is_intermediate,
       width: width,
       height: height,
+      use_cache: false,
     };
 
     graph.edges.push(
@@ -276,6 +277,7 @@ export const buildCanvasTextToImageGraph = (
       id: CANVAS_OUTPUT,
       is_intermediate,
       fp32,
+      use_cache: false,
     };
 
     graph.edges.push({

invokeai/frontend/web/src/features/nodes/util/graph/buildLinearImageToImageGraph.ts

@@ -143,6 +143,7 @@ export const buildLinearImageToImageGraph = (
         // },
         fp32,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildLinearSDXLImageToImageGraph.ts

@@ -154,6 +154,7 @@ export const buildLinearSDXLImageToImageGraph = (
         // },
         fp32,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildLinearSDXLTextToImageGraph.ts

@@ -127,6 +127,7 @@ export const buildLinearSDXLTextToImageGraph = (
         id: LATENTS_TO_IMAGE,
         fp32,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/features/nodes/util/graph/buildLinearTextToImageGraph.ts

@@ -146,6 +146,7 @@ export const buildLinearTextToImageGraph = (
         id: LATENTS_TO_IMAGE,
         fp32,
         is_intermediate,
+        use_cache: false,
       },
     },
     edges: [

invokeai/frontend/web/src/services/api/authToastMiddleware.ts

@@ -5,12 +5,10 @@ import { t } from 'i18next';
 import { z } from 'zod';
 
 const zRejectedForbiddenAction = z.object({
-  action: z.object({
-    payload: z.object({
-      status: z.literal(403),
-      data: z.object({
-        detail: z.string(),
-      }),
+  payload: z.object({
+    status: z.literal(403),
+    data: z.object({
+      detail: z.string(),
     }),
   }),
 });
@@ -22,8 +20,8 @@ export const authToastMiddleware: Middleware =
         const parsed = zRejectedForbiddenAction.parse(action);
         const { dispatch } = api;
         const customMessage =
-          parsed.action.payload.data.detail !== 'Forbidden'
-            ? parsed.action.payload.data.detail
+          parsed.payload.data.detail !== 'Forbidden'
+            ? parsed.payload.data.detail
             : undefined;
         dispatch(
           addToast({
@@ -32,7 +30,7 @@ export const authToastMiddleware: Middleware =
             description: customMessage,
           })
         );
-      } catch {
+      } catch (error) {
         // no-op
       }
     }

pyproject.toml

@@ -113,6 +113,7 @@ dependencies = [
 "onnx" = ["onnxruntime"]
 "onnx-cuda" = ["onnxruntime-gpu"]
 "onnx-directml" = ["onnxruntime-directml"]
+"stable-fast" = ["stable-fast"]
 
 [project.scripts]
 
@@ -138,7 +139,6 @@ dependencies = [
 "invokeai-node-web" = "invokeai.app.api_app:invoke_api"
 "invokeai-import-images" = "invokeai.frontend.install.import_images:main"
 "invokeai-db-maintenance" = "invokeai.backend.util.db_maintenance:main"
-"invokeai-migrate-models-to-db" = "invokeai.backend.model_manager.migrate_to_db:main"
 
 [project.urls]
 "Homepage" = "https://invoke-ai.github.io/InvokeAI/"

tests/backend/tiles/test_tiles.py

@@ -305,9 +305,7 @@ def test_calc_tiles_min_overlap_input_validation(
 
 def test_calc_tiles_even_split_single_tile():
     """Test calc_tiles_even_split() behavior when a single tile covers the image."""
-    tiles = calc_tiles_even_split(
-        image_height=512, image_width=1024, num_tiles_x=1, num_tiles_y=1, overlap_fraction=0.25
-    )
+    tiles = calc_tiles_even_split(image_height=512, image_width=1024, num_tiles_x=1, num_tiles_y=1, overlap=64)
 
     expected_tiles = [
         Tile(
@@ -322,36 +320,34 @@ def test_calc_tiles_even_split_single_tile():
 def test_calc_tiles_even_split_evenly_divisible():
     """Test calc_tiles_even_split() behavior when the image is evenly covered by multiple tiles."""
     # Parameters mimic roughly the same output as the original tile generations of the same test name
-    tiles = calc_tiles_even_split(
-        image_height=576, image_width=1600, num_tiles_x=3, num_tiles_y=2, overlap_fraction=0.25
-    )
+    tiles = calc_tiles_even_split(image_height=576, image_width=1600, num_tiles_x=3, num_tiles_y=2, overlap=64)
 
     expected_tiles = [
         # Row 0
         Tile(
-            coords=TBLR(top=0, bottom=320, left=0, right=624),
-            overlap=TBLR(top=0, bottom=72, left=0, right=136),
+            coords=TBLR(top=0, bottom=320, left=0, right=576),
+            overlap=TBLR(top=0, bottom=64, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=0, bottom=320, left=488, right=1112),
-            overlap=TBLR(top=0, bottom=72, left=136, right=136),
+            coords=TBLR(top=0, bottom=320, left=512, right=1088),
+            overlap=TBLR(top=0, bottom=64, left=64, right=64),
         ),
         Tile(
-            coords=TBLR(top=0, bottom=320, left=976, right=1600),
-            overlap=TBLR(top=0, bottom=72, left=136, right=0),
+            coords=TBLR(top=0, bottom=320, left=1024, right=1600),
+            overlap=TBLR(top=0, bottom=64, left=64, right=0),
         ),
         # Row 1
         Tile(
-            coords=TBLR(top=248, bottom=576, left=0, right=624),
-            overlap=TBLR(top=72, bottom=0, left=0, right=136),
+            coords=TBLR(top=256, bottom=576, left=0, right=576),
+            overlap=TBLR(top=64, bottom=0, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=248, bottom=576, left=488, right=1112),
-            overlap=TBLR(top=72, bottom=0, left=136, right=136),
+            coords=TBLR(top=256, bottom=576, left=512, right=1088),
+            overlap=TBLR(top=64, bottom=0, left=64, right=64),
         ),
         Tile(
-            coords=TBLR(top=248, bottom=576, left=976, right=1600),
-            overlap=TBLR(top=72, bottom=0, left=136, right=0),
+            coords=TBLR(top=256, bottom=576, left=1024, right=1600),
+            overlap=TBLR(top=64, bottom=0, left=64, right=0),
         ),
     ]
     assert tiles == expected_tiles
@@ -360,36 +356,34 @@ def test_calc_tiles_even_split_evenly_divisible():
 def test_calc_tiles_even_split_not_evenly_divisible():
     """Test calc_tiles_even_split() behavior when the image requires 'uneven' overlaps to achieve proper coverage."""
     # Parameters mimic roughly the same output as the original tile generations of the same test name
-    tiles = calc_tiles_even_split(
-        image_height=400, image_width=1200, num_tiles_x=3, num_tiles_y=2, overlap_fraction=0.25
-    )
+    tiles = calc_tiles_even_split(image_height=400, image_width=1200, num_tiles_x=3, num_tiles_y=2, overlap=64)
 
     expected_tiles = [
         # Row 0
         Tile(
-            coords=TBLR(top=0, bottom=224, left=0, right=464),
-            overlap=TBLR(top=0, bottom=56, left=0, right=104),
+            coords=TBLR(top=0, bottom=232, left=0, right=440),
+            overlap=TBLR(top=0, bottom=64, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=0, bottom=224, left=360, right=824),
-            overlap=TBLR(top=0, bottom=56, left=104, right=104),
+            coords=TBLR(top=0, bottom=232, left=376, right=816),
+            overlap=TBLR(top=0, bottom=64, left=64, right=64),
         ),
         Tile(
-            coords=TBLR(top=0, bottom=224, left=720, right=1200),
-            overlap=TBLR(top=0, bottom=56, left=104, right=0),
+            coords=TBLR(top=0, bottom=232, left=752, right=1200),
+            overlap=TBLR(top=0, bottom=64, left=64, right=0),
         ),
         # Row 1
         Tile(
-            coords=TBLR(top=168, bottom=400, left=0, right=464),
-            overlap=TBLR(top=56, bottom=0, left=0, right=104),
+            coords=TBLR(top=168, bottom=400, left=0, right=440),
+            overlap=TBLR(top=64, bottom=0, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=168, bottom=400, left=360, right=824),
-            overlap=TBLR(top=56, bottom=0, left=104, right=104),
+            coords=TBLR(top=168, bottom=400, left=376, right=816),
+            overlap=TBLR(top=64, bottom=0, left=64, right=64),
         ),
         Tile(
-            coords=TBLR(top=168, bottom=400, left=720, right=1200),
-            overlap=TBLR(top=56, bottom=0, left=104, right=0),
+            coords=TBLR(top=168, bottom=400, left=752, right=1200),
+            overlap=TBLR(top=64, bottom=0, left=64, right=0),
         ),
     ]
 
@@ -399,28 +393,26 @@ def test_calc_tiles_even_split_not_evenly_divisible():
 def test_calc_tiles_even_split_difficult_size():
     """Test calc_tiles_even_split() behavior when the image is a difficult size to spilt evenly and keep div8."""
     # Parameters are a difficult size for other tile gen routines to calculate
-    tiles = calc_tiles_even_split(
-        image_height=1000, image_width=1000, num_tiles_x=2, num_tiles_y=2, overlap_fraction=0.25
-    )
+    tiles = calc_tiles_even_split(image_height=1000, image_width=1000, num_tiles_x=2, num_tiles_y=2, overlap=64)
 
     expected_tiles = [
         # Row 0
         Tile(
-            coords=TBLR(top=0, bottom=560, left=0, right=560),
-            overlap=TBLR(top=0, bottom=128, left=0, right=128),
+            coords=TBLR(top=0, bottom=528, left=0, right=528),
+            overlap=TBLR(top=0, bottom=64, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=0, bottom=560, left=432, right=1000),
-            overlap=TBLR(top=0, bottom=128, left=128, right=0),
+            coords=TBLR(top=0, bottom=528, left=464, right=1000),
+            overlap=TBLR(top=0, bottom=64, left=64, right=0),
         ),
         # Row 1
         Tile(
-            coords=TBLR(top=432, bottom=1000, left=0, right=560),
-            overlap=TBLR(top=128, bottom=0, left=0, right=128),
+            coords=TBLR(top=464, bottom=1000, left=0, right=528),
+            overlap=TBLR(top=64, bottom=0, left=0, right=64),
         ),
         Tile(
-            coords=TBLR(top=432, bottom=1000, left=432, right=1000),
-            overlap=TBLR(top=128, bottom=0, left=128, right=0),
+            coords=TBLR(top=464, bottom=1000, left=464, right=1000),
+            overlap=TBLR(top=64, bottom=0, left=64, right=0),
         ),
     ]
 
@@ -428,11 +420,13 @@ def test_calc_tiles_even_split_difficult_size():
 
 
 @pytest.mark.parametrize(
-    ["image_height", "image_width", "num_tiles_x", "num_tiles_y", "overlap_fraction", "raises"],
+    ["image_height", "image_width", "num_tiles_x", "num_tiles_y", "overlap", "raises"],
     [
-        (128, 128, 1, 1, 0.25, False),  # OK
+        (128, 128, 1, 1, 127, False),  # OK
         (128, 128, 1, 1, 0, False),  # OK
-        (128, 128, 2, 1, 0, False),  # OK
+        (128, 128, 2, 2, 0, False),  # OK
+        (128, 128, 2, 1, 120, True),  # overlap equals tile_height.
+        (128, 128, 1, 2, 120, True),  # overlap equals tile_width.
         (127, 127, 1, 1, 0, True),  # image size must be dividable by 8
     ],
 )
@@ -441,15 +435,15 @@ def test_calc_tiles_even_split_input_validation(
     image_width: int,
     num_tiles_x: int,
     num_tiles_y: int,
-    overlap_fraction: float,
+    overlap: int,
     raises: bool,
 ):
     """Test that calc_tiles_even_split() raises an exception if the inputs are invalid."""
     if raises:
-        with pytest.raises(ValueError):
-            calc_tiles_even_split(image_height, image_width, num_tiles_x, num_tiles_y, overlap_fraction)
+        with pytest.raises((AssertionError, ValueError)):
+            calc_tiles_even_split(image_height, image_width, num_tiles_x, num_tiles_y, overlap)
     else:
-        calc_tiles_even_split(image_height, image_width, num_tiles_x, num_tiles_y, overlap_fraction)
+        calc_tiles_even_split(image_height, image_width, num_tiles_x, num_tiles_y, overlap)
 
 
 #############################################