pyproject.toml

[build-system]
requires = ["setuptools", "pip", "wheel"]
build-backend = "setuptools.build_meta"

[project]
name = "InvokeAI"
description = "A full-featured AI-assisted image generation environment designed for creatives and enthusiasts."
requires-python = ">=3.12, <3.13"
readme = { content-type = "text/markdown", file = "README.md" }
keywords = ["stable-diffusion", "AI"]
dynamic = ["version"]
license = { file = "LICENSE" }
authors = [{ name = "Invoke", email = "support@invoke.ai" }]
classifiers = [
  'Development Status :: 5 - Production/Stable',
  'Environment :: GPU',
  'Environment :: GPU :: NVIDIA CUDA',
  'Environment :: MacOS X',
  'Intended Audience :: End Users/Desktop',
  'Intended Audience :: Developers',
  'License :: OSI Approved :: Apache Software License',
  'Operating System :: POSIX :: Linux',
  'Operating System :: MacOS',
  'Operating System :: Microsoft :: Windows',
  'Programming Language :: Python :: 3 :: Only',
  'Programming Language :: Python :: 3.12',
  'Topic :: Artistic Software',
  'Topic :: Internet :: WWW/HTTP :: WSGI :: Application',
  'Topic :: Internet :: WWW/HTTP :: WSGI :: Server',
  'Topic :: Multimedia :: Graphics',
  'Topic :: Scientific/Engineering :: Artificial Intelligence',
  'Topic :: Scientific/Engineering :: Image Processing',
]
dependencies = [
  # Core generation dependencies, pinned for reproducible builds.
  "accelerate",
  "bitsandbytes; sys_platform!='darwin'",
  "compel>=2.4.0,<3",
  "diffusers[torch]==0.40.0",
  "gguf",
  "mediapipe==0.10.14",                   # needed for "mediapipeface" controlnet model
  "mistral-common>=1.5.4,<2",             # canonical Tekken tokenizer for FLUX.2 [dev] Mistral encoder; the
                                          # loader depends on private surface (Tekkenizer internals) that
                                          # moves across majors, so cap below 2.x (validated against 1.11.6)
  "numpy<2.0.0",
  "onnx==1.16.1",
  "onnxruntime==1.19.2",
  "opencv-contrib-python",
  "imageio[ffmpeg]>=2.37",                 # video encode (for Wan 2.2 T2V/I2V output); encode behavior (macro_block_size) is version-sensitive
  "psutil>=6",                             # video decode worker process-tree termination
  "safetensors",
  "sentencepiece==0.2.0",                 # 0.2.1 coredumps windows when loading t5 tokenizer
  "spandrel",
  # Loosely pinned, will respect requirement of `diffusers[torch]`. Split by platform: linux/win allow
  # the 2.13 line every backend extra pins (the lock itself is held there by `constraint-dependencies` in
  # [tool.uv]), while macOS stays on 2.7.x — newer macOS torch wheels exercise MPS on CI runners (no
  # usable Metal GPU) and fail with MPS OOM.
  # Deliberately NOT capped below 2.12 here. torch 2.12.x+rocm7.1 was reported to break generation (#9410),
  # but this range is what every backend-agnostic install resolves against — manual installs use
  # `--torch-backend=<x>` with no extra (see docs/start-here/manual), so a blanket cap would also
  # reject torch>=2.12 on Windows/Linux CUDA, CPU and ARM64, where 2.12 has no known problem.
  # The ROCm path is constrained where it can actually be targeted: the `rocm` extra below pins an
  # exact version — 2.13.0+rocm7.2, validated end-to-end on gfx1100 (W7900), where its bundled
  # AOTriton 0.12b also triples flash-attention throughput vs the 2.10.0+rocm7.1 pin.
  "torch>=2.7.0,<3.0; sys_platform != 'darwin'",
  "torch>=2.7.0,<2.8.0; sys_platform == 'darwin'",
  "torchsde",                             # diffusers needs this for SDE solvers, but it is not an explicit dep of diffusers
  "torchvision",
  "transformers>=5.5,<5.6",

  # Core application dependencies, pinned for reproducible builds.
  "fastapi-events",
  # The old pin sat at 0.118.3 because 0.119.0 crashed generating our OpenAPI schema. That was a FastAPI bug, not
  # ours (`KeyError: '$ref'` in fastapi/_compat/v2.py, which assumed every field mapping carries a `$ref`), and it
  # is fixed as of 0.124.0 — no change to AnyInvocation was needed.
  #
  # Two later changes did need adapting to, both handled: 0.130 emits `contentMediaType` instead of
  # `format: binary` for file uploads (see the Blob mapping in frontend/api/scripts/typegen.js), and 0.141 keeps
  # included routers as a single node in `app.routes` rather than copying their routes into it (see
  # `_iter_route_contexts` in tests/app/routers/test_model_manager_authorization.py). Keep the minor-version bound:
  # both of those were silent breakages that only surfaced because something happened to assert on them.
  "fastapi>=0.141.1,<0.142",
  "huggingface-hub",
  "networkx",
  "pydantic-settings",
  "pydantic",
  "python-socketio",
  "uvicorn[standard]",

  # Auxiliary dependencies, pinned only if necessary.
  "blake3",
  "bcrypt<4.0.0",
  "Deprecated",
  "dnspython",
  "dynamicprompts",
  "einops",
  "email-validator>=2.0.0",
  "fonttools[woff]",
  "passlib[bcrypt]>=1.7.4",
  "picklescan",
  "pillow",
  "prompt-toolkit",
  "pypatchmatch",
  "python-jose[cryptography]>=3.3.0",
  "python-multipart",
  "requests",
  "semver~=3.0.1",
  "PyWavelets",

  # Semantic image index (image map / semantic search).
  # numba is a umap-learn transitive, pinned to the range this stack has been
  # exercised against rather than for numpy compatibility: every release from
  # 0.59 to at least 0.63 accepts the numpy 1.26.x resolved under numpy<2.0
  # above (0.62 widened its floor back to >=1.22). Raise the ceiling once a
  # newer numba has been run against the image map, not just resolved.
  "umap-learn~=0.5.7",
  "scikit-learn~=1.5",
  "numba>=0.59,<0.62",
]

[project.optional-dependencies]
"xformers" = [
  # Core generation dependencies, pinned for reproducible builds.
  "xformers>=0.0.28.post1; sys_platform!='darwin'",
  # torch 2.4+cu carries its own triton dependency
]

# The +cpu/+cu130/+rocm7.2 pins live on PyTorch's WHL indexes and are gated to the
# platforms the project resolves and tests them on: Windows and Linux x86_64 (rocm: Linux
# only). macOS is capped below torch 2.8 (see the base requirement), and linux_aarch64 stays
# on the CPU-only PyPI 2.7.1 it was set up with (an aarch64 install requesting `cuda` gets
# CPU torch), so both fall back to the base `torch` / `torchvision` from PyPI. The index has
# carried aarch64 torchvision since 0.28, so moving aarch64 onto the pinned pairs -- and real
# CUDA-on-aarch64 (e.g. Grace/GH200) -- is open, just untested.
#
# cu130 builds cover compute capability 7.5 and newer (Turing through Blackwell) and need
# an NVIDIA driver from the R580 series or newer. CUDA 13 cannot build for Maxwell, Pascal
# or Volta; the cu126 builds that still can have no Blackwell kernels, and cu128 stops at
# torch 2.11.
"cpu" = [
  "torch==2.13.0+cpu; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  "torchvision==0.28.0+cpu; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  # macOS / linux_aarch64 fallback, resolved from PyPI. Without these explicit entries
  # the extra's conflict-universe would contain no torch at all there (uv partitions
  # the base declarations into the no-extra universe).
  "torch==2.7.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
  "torchvision==0.22.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
]
"cuda" = [
  "torch==2.13.0+cu130; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  "torchvision==0.28.0+cu130; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  # macOS / linux_aarch64 fallback, see note under "cpu"
  "torch==2.7.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
  "torchvision==0.22.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
]
"rocm" = [
  # ROCm wheels are x86_64-linux-only; gate by platform so macOS/win/linux-aarch64 resolution is unaffected.
  # 2.13.0+rocm7.2 bundles AOTriton 0.12b, whose regenerated gfx1100 tuning database triples
  # flash-attention throughput on RDNA3 vs the previous 2.10.0+rocm7.1 pin (measured 17.9 ->
  # 57.3 TFLOPS effective at the MiniMax H3 attention shape on a W7900).
  "torch==2.13.0+rocm7.2; sys_platform == 'linux' and platform_machine != 'aarch64'",
  "torchvision==0.28.0+rocm7.2; sys_platform == 'linux' and platform_machine != 'aarch64'",
  "triton-rocm==3.7.1; sys_platform == 'linux' and platform_machine != 'aarch64'",
  # linux_aarch64 fallback, see note under "cpu"
  "torch==2.7.1; sys_platform == 'linux' and platform_machine == 'aarch64'",
  "torchvision==0.22.1; sys_platform == 'linux' and platform_machine == 'aarch64'",
]
# Intel XPU (Arc / Battlemage) wheels are published only for linux-x86_64 and
# windows-amd64 on PyTorch's WHL index. Gate the +xpu pins to those platforms; the
# Intel oneAPI runtime libs (intel-sycl-rt, etc.) come in automatically as deps of
# torch+xpu.
# Intel's XPU backend matured considerably after 2.7.1: torch.xpu.mem_get_info() works on
# driver/kernel combinations where it previously raised, and the bundled oneAPI runtime ships
# with the wheel (intel-sycl-rt), so upgrading torch upgrades the user-space runtime too.
"xpu" = [
  "torch==2.13.0+xpu; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  "torchvision==0.28.0+xpu; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  # XPU triton (for torch.compile); the WHL index ships linux-x86_64 and win_amd64 wheels.
  # Renamed from pytorch-triton-xpu to triton-xpu upstream.
  "triton-xpu==3.7.2; (sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'",
  # macOS / linux_aarch64 have no +xpu wheels: fall back to base torch so the extra's
  # conflict-universe always contains a torch. See note under "cpu".
  "torch==2.7.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
  "torchvision==0.22.1; sys_platform == 'darwin' or (sys_platform == 'linux' and platform_machine == 'aarch64')",
]

"onnx" = ["onnxruntime"]
"onnx-cuda" = ["onnxruntime-gpu"]
"onnx-directml" = ["onnxruntime-directml"]
"dist" = ["pip-tools", "pipdeptree", "twine"]
"dev" = ["jurigged", "pudb", "snakeviz", "gprof2dot"]
"test" = [
  "ruff~=0.11.2",
  "ruff-lsp~=0.0.62",
  "mypy",
  "pre-commit",
  "pytest>6.0.0",
  "pytest-cov",
  "pytest-timeout",
  "pytest-xdist",
  "pytest-datadir",
  "requests_testadapter",
  "httpx",
  "polyfactory==2.19.0",
  "humanize==4.12.1",
]

[tool.uv]
# Prevent opencv-python from ever being chosen during dependency resolution.
# This prevents conflicts with opencv-contrib-python, which Invoke requires.
override-dependencies = ["opencv-python; sys_platform=='never'"]
conflicts = [[{ extra = "cpu" }, { extra = "cuda" }, { extra = "rocm" }, { extra = "xpu" }]]
index-strategy = "unsafe-best-match"
# Restrict resolution to the platforms we support: x86_64/aarch64 Linux, Windows,
# and macOS. (ROCm and the PyTorch WHL-index pins are x86_64-only; on aarch64
# Linux torch/torchvision fall back to PyPI via the markers below.)
environments = ["sys_platform == 'win32' or sys_platform == 'darwin' or (sys_platform == 'linux' and (platform_machine == 'x86_64' or platform_machine == 'aarch64'))"]
# Hold every resolution on Windows and Linux x86_64 to the torch the backend extras pin -- including the
# no-extra one, which is what CI installs and tests. Without this the lock is free to keep an older torch
# there, or move it to the newest PyPI release, independently of the extras. `tool.uv` never reaches the
# published metadata, so installing the released package with `--torch-backend` keeps the open range of the base
# requirement; only `uv pip install` inside a checkout is held to it.
# `==2.13.0` also matches the local builds (+cpu, +cu130, +rocm7.2, +xpu).
constraint-dependencies = [
  "torch==2.13.0; sys_platform == 'win32' or (sys_platform == 'linux' and platform_machine == 'x86_64')",
  "torchvision==0.28.0; sys_platform == 'win32' or (sys_platform == 'linux' and platform_machine == 'x86_64')",
]

# compel, could you please not pull in all of Jupyter. (compel declares `notebook>=6.5.7` but does not
# reference IPython/Jupyter anywhere in its code.) Dropping it takes ~65 packages out of the environment.
#
# Deliberately the *global* form rather than `{ package = ..., dependencies = ... }`: the per-package form
# needs uv >= 0.11.25 and is a hard parse error on 0.10.0 - 0.11.24, while this form has been understood
# since uv 0.9.8 and resolves to exactly the same set. Older uv only warns and ignores it, so no install
# path breaks -- and the launcher syncs `--frozen` from the lockfile below, so users get the trimmed
# environment whichever uv they have. Keep it that way: a `required-version` here would hard-fail the uv
# 0.6.12 bundled in launcher <= 1.8.1 for no gain, since CI's `uv lock --locked` already catches a
# lockfile regenerated by a uv too old to honor this.
exclude-dependencies = ["notebook"]

[tool.uv.sources]
# Keep linux_aarch64 (and, for cpu/cuda, macOS) off the PyTorch WHL indexes so that uv
# resolves their 2.7.1 fallback pins from PyPI; see the note above the extras.
torch = [
  { index = "torch-cpu", extra = "cpu", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "torch-cuda", extra = "cuda", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "torch-rocm", extra = "rocm", marker = "sys_platform != 'linux' or platform_machine != 'aarch64'" },
  { index = "torch-xpu", extra = "xpu", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "pypi", marker = "sys_platform == 'linux' and platform_machine == 'aarch64'" },
]
torchvision = [
  { index = "torch-cpu", extra = "cpu", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "torch-cuda", extra = "cuda", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "torch-rocm", extra = "rocm", marker = "sys_platform != 'linux' or platform_machine != 'aarch64'" },
  { index = "torch-xpu", extra = "xpu", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
  { index = "pypi", marker = "sys_platform == 'linux' and platform_machine == 'aarch64'" },
]
triton-rocm = [
  { index = "torch-rocm", marker = "sys_platform == 'linux' and platform_machine != 'aarch64'" },
]
# PyPI's xformers wheels are built against CUDA 12.8 and cannot load their extension next to a CUDA 13 torch (every
# memory-efficient attention call then fails); PyTorch's cu130 index carries the CUDA 13 build of the same release.
# Not scoped to the `cuda` extra: uv only allows that for packages the extra itself lists, and the CUDA 13 torch that
# PyPI serves on Linux needs the same build. The cpu/rocm/xpu builds cannot use xformers' CUDA kernels either way.
xformers = [
  { index = "torch-cuda", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
]
triton-xpu = [
  { index = "torch-xpu", marker = "(sys_platform == 'linux' and platform_machine == 'x86_64') or sys_platform == 'win32'" },
]

[[tool.uv.index]]
name = "pypi"
url = "https://pypi.org/simple"

[[tool.uv.index]]
name = "torch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true

[[tool.uv.index]]
name = "torch-cuda"
url = "https://download.pytorch.org/whl/cu130"
explicit = true

[[tool.uv.index]]
name = "torch-rocm"
url = "https://download.pytorch.org/whl/rocm7.2"
explicit = true

[[tool.uv.index]]
name = "torch-xpu"
url = "https://download.pytorch.org/whl/xpu"
explicit = true

[project.scripts]
"invokeai-web" = "invokeai.app.run_app:run_app"
"invoke-useradd" = "invokeai.app.util.user_management:useradd"
"invoke-userdel" = "invokeai.app.util.user_management:userdel"
"invoke-userlist" = "invokeai.app.util.user_management:userlist"
"invoke-usermod" = "invokeai.app.util.user_management:usermod"

[project.urls]
"Homepage" = "https://invoke.ai/"
"Documentation" = "https://invoke.ai/"
"Source" = "https://github.com/invoke-ai/InvokeAI/"
"Bug Reports" = "https://github.com/invoke-ai/InvokeAI/issues"
"Discord" = "https://discord.gg/ZmtBAhwWhy"

[tool.setuptools.dynamic]
version = { attr = "invokeai.version.__version__" }

[tool.setuptools.packages.find]
"where" = ["."]
"include" = [
  "invokeai.assets.fonts*",
  "invokeai.version*",
  "invokeai.generator*",
  "invokeai.backend*",
  "invokeai.frontend*",
  "invokeai.frontend.webv1.dist*",
  "invokeai.frontend.webv1.static*",
  "invokeai.configs*",
  "invokeai.app*",
  "invokeai.invocation_api*",
]

[tool.setuptools.package-data]
"invokeai.app.assets" = ["**/*.png"]
"invokeai.backend.qwen3" = ["tokenizer/*.json", "tokenizer/*.json.gz"]
"invokeai.backend.qwen3_vl" = ["tokenizer/*.json", "tokenizer/*.json.gz", "config/*.json"]
"invokeai.backend.qwen3_5" = ["*.json", "tokenizer/*.json", "tokenizer/*.json.gz"]
"invokeai.backend.qwen2_5_vl" = ["*.json", "tokenizer/*.json", "tokenizer/*.json.gz"]
# Read at load time by the MiniMax H3 single-file encoder loader. Without this the file is
# absent from a wheel install and that loader raises FileNotFoundError.
"invokeai.backend.minimax_h3" = ["*.json"]
"invokeai.backend.t5" = ["tokenizer/*.json"]
"invokeai.app.services.image_index" = ["*.txt"]
"invokeai.app.services.workflow_records.default_workflows" = ["*.json"]
"invokeai.backend.hidiffusion" = ["sd_module_key/*.txt"]
"invokeai.app.services.style_preset_records" = ["*.json"]
"invokeai.app.services.style_preset_images.default_style_preset_images" = [
  "*.png",
]
"invokeai.assets.fonts" = ["**/*.ttf"]
"invokeai.backend" = ["**.png", "**/*.icc"]
"invokeai.configs" = ["*.example", "**/*.yaml", "*.txt"]
"invokeai.frontend.webv1.dist" = ["**"]
"invokeai.frontend.webv2.dist" = ["**"]
"invokeai.frontend.webv1.static" = ["**"]
"invokeai.app.invocations" = ["**"]

#=== Begin: PyTest and Coverage
[tool.pytest.ini_options]
# `--dist loadfile` keeps a file's tests together on one xdist worker, so module-scoped
# fixtures are set up once per file instead of in every worker that received part of it.
# It is not isolation: a worker process runs many files, in an order that depends on the
# worker count and on which file finished first, so a file must still set up its own
# process-global state. It costs nothing -- per-test distribution measured the same wall
# clock on the full suite -- and it lives here rather than in the CI command so that it
# also holds when `-n` is passed by hand. Coverage is opt-in via `--cov`.
addopts = "--strict-markers --dist loadfile -m \"not slow\""
markers = [
  "slow: Marks tests as slow. Disabled by default. To run all tests, use -m \"\". To run only slow tests, use -m \"slow\".",
  "timeout: Marks the timeout override.",
]
[tool.coverage.run]
branch = true
source = ["invokeai"]
omit = ["*tests*", "*migrations*", ".venv/*", "*.env"]
[tool.coverage.report]
show_missing = true
fail_under = 85     # let's set something sensible on Day 1 ...
[tool.coverage.json]
output = "coverage/coverage.json"
pretty_print = true
[tool.coverage.html]
directory = "coverage/html"
[tool.coverage.xml]
output = "coverage/index.xml"
#=== End: PyTest and Coverage

#=== Begin: Ruff
[tool.ruff]
line-length = 120
exclude = [
  ".git",
  "__pycache__",
  "build",
  "dist",
  "invokeai/frontend/api/node_modules/",
  "invokeai/frontend/webv1/node_modules/",
  "invokeai/frontend/webv2/node_modules/",
  ".venv*",
  "*.ipynb",
  "invokeai/backend/image_util/mediapipe_face/", # External code
  "invokeai/backend/image_util/mlsd/",           # External code
  "invokeai/backend/image_util/normal_bae/",     # External code
  "invokeai/backend/image_util/pidi/",           # External code
  "invokeai/backend/image_util/imwatermark/",    # External code
]

[tool.ruff.lint]
ignore = [
  "E501", # https://docs.astral.sh/ruff/rules/line-too-long/
  "C901", # https://docs.astral.sh/ruff/rules/complex-structure/
  "B008", # https://docs.astral.sh/ruff/rules/function-call-in-default-argument/
  "B904", # https://docs.astral.sh/ruff/rules/raise-without-from-inside-except/
]
select = ["B", "C", "E", "F", "W", "I", "TID"]

[tool.ruff.lint.per-file-ignores]
# Vendored from the diffusers MiniMax-H3 branch (see invokeai/backend/minimax_h3/__init__.py);
# kept as close to upstream as possible, so upstream's zip() style is tolerated.
"invokeai/backend/minimax_h3/transformer_minimax_h3.py" = ["B905"]
"invokeai/backend/minimax_h3/autoencoder_kl_minimax_h3.py" = ["B905"]
"invokeai/backend/minimax_h3/autoencoder_kl_minimax_h3_audio.py" = ["B905"]
"invokeai/backend/minimax_h3/scheduling_minimax_h3.py" = ["B905"]

[tool.ruff.format]
# The vendored MiniMax-H3 files are kept byte-identical to upstream diffusers (modulo the
# absolute-import rewrite) so that re-syncing when a tagged diffusers ships them is a clean copy
# rather than a manual re-application of formatting. Upstream wraps at a narrower line length, so
# `ruff format` would rewrap lines for no semantic gain. `ruff check` still runs on them, so the
# import sorting this project does enforce stays enforced.
#
# Listed file by file rather than as a directory glob: the package also holds first-party modules
# (presets.py, sampling.py, denoise.py, int8_convrot.py, ...) which must stay formatted.
exclude = [
  "invokeai/backend/minimax_h3/packing.py",
  "invokeai/backend/minimax_h3/transformer_minimax_h3.py",
  "invokeai/backend/minimax_h3/autoencoder_kl_minimax_h3.py",
  "invokeai/backend/minimax_h3/autoencoder_kl_minimax_h3_audio.py",
  "invokeai/backend/minimax_h3/scheduling_minimax_h3.py",
]

[tool.ruff.lint.flake8-tidy-imports]
# Disallow all relative imports.
ban-relative-imports = "all"

#=== End: Ruff

#=== Begin: MyPy

# global mypy config
[tool.mypy]
ignore_missing_imports = true # ignores missing types in third-party libraries
strict = true
plugins = "pydantic.mypy"
exclude = ["tests/*"]

# overrides for specific modules
[[tool.mypy.overrides]]
follow_imports = "skip" # skips type checking of the modules listed below
module = [
  "invokeai.app.api.routers.models",
  "invokeai.app.invocations.text_encoder.compel",
  "invokeai.app.invocations.sd.denoise_latents",
  "invokeai.app.services.invocation_stats.invocation_stats_default",
  "invokeai.app.services.model_manager.model_manager_base",
  "invokeai.app.services.model_manager.model_manager_default",
  "invokeai.app.services.model_manager.store.model_records_sql",
  "invokeai.app.util.controlnet_utils",
  "invokeai.backend.image_util.txt2mask",
  "invokeai.backend.image_util.safety_checker",
  "invokeai.backend.image_util.patchmatch",
  "invokeai.backend.image_util.invisible_watermark",
  "invokeai.backend.install.model_install_backend",
  "invokeai.backend.ip_adapter.ip_adapter",
  "invokeai.backend.ip_adapter.resampler",
  "invokeai.backend.ip_adapter.unet_patcher",
  "invokeai.backend.model_management.convert_ckpt_to_diffusers",
  "invokeai.backend.model_management.lora",
  "invokeai.backend.model_management.model_cache",
  "invokeai.backend.model_management.model_manager",
  "invokeai.backend.model_management.model_merge",
  "invokeai.backend.model_management.model_probe",
  "invokeai.backend.model_management.model_search",
  "invokeai.backend.model_management.models.*",                            # this is needed to ignore the module's `__init__.py`
  "invokeai.backend.model_management.models.base",
  "invokeai.backend.model_management.models.controlnet",
  "invokeai.backend.model_management.models.ip_adapter",
  "invokeai.backend.model_management.models.lora",
  "invokeai.backend.model_management.models.sdxl",
  "invokeai.backend.model_management.models.stable_diffusion",
  "invokeai.backend.model_management.models.vae",
  "invokeai.backend.model_management.seamless",
  "invokeai.backend.model_management.util",
  "invokeai.backend.stable_diffusion.diffusers_pipeline",
  "invokeai.backend.stable_diffusion.diffusion.shared_invokeai_diffusion",
  "invokeai.backend.util.hotfixes",
  "invokeai.backend.util.mps_fixes",
  "invokeai.backend.util.util",
  "invokeai.frontend.install.model_install",
]
#=== End: MyPy

[tool.pyright]
# Start from strict mode
typeCheckingMode = "strict"
# This errors whenever an import is missing a type stub file - way too noisy
reportMissingTypeStubs = "none"
# These are the rest of the rules enabled by strict mode - enable them @ warning
reportConstantRedefinition = "warning"
reportDeprecated = "warning"
reportDuplicateImport = "warning"
reportIncompleteStub = "warning"
reportInconsistentConstructor = "warning"
reportInvalidStubStatement = "warning"
reportMatchNotExhaustive = "warning"
reportMissingParameterType = "warning"
reportMissingTypeArgument = "warning"
reportPrivateUsage = "warning"
reportTypeCommentUsage = "warning"
reportUnknownArgumentType = "warning"
reportUnknownLambdaType = "warning"
reportUnknownMemberType = "warning"
reportUnknownParameterType = "warning"
reportUnknownVariableType = "warning"
reportUnnecessaryCast = "warning"
reportUnnecessaryComparison = "warning"
reportUnnecessaryContains = "warning"
reportUnnecessaryIsInstance = "warning"
reportUnusedClass = "warning"
reportUnusedImport = "warning"
reportUnusedFunction = "warning"
reportUnusedVariable = "warning"
reportUntypedBaseClass = "warning"
reportUntypedClassDecorator = "warning"
reportUntypedFunctionDecorator = "warning"
reportUntypedNamedTuple = "warning"