Instructions to use arlaz/modular-flux2-multidiffusion with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use arlaz/modular-flux2-multidiffusion with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("arlaz/modular-flux2-multidiffusion", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
File size: 18,256 Bytes
d8bf41d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 | from types import SimpleNamespace
import pytest
import torch
from PIL import Image
from examples import example as runner
from examples import example_remote as remote_runner
class FakeImage:
def __init__(self, size):
self.size = size
self.saved = []
def save(self, path, **kwargs):
self.saved.append((path, kwargs))
def _args(*extra):
return runner.apply_window_stride_defaults(runner.parse_args(["--base-model", "model", *extra]))
def test_quantization_parser_defaults_to_none():
args = _args()
assert args.quantization == "none"
assert args.transformer_quantization is None
assert args.text_encoder_quantization is None
assert args.vae_quantization is None
assert args.allow_tf32 is False
def test_allow_tf32_parser_flag():
args = _args("--allow-tf32")
assert args.allow_tf32 is True
def test_configure_torch_skips_tf32_when_not_allowed(monkeypatch):
calls = []
monkeypatch.setattr(runner.torch.cuda, "is_available", lambda: True)
monkeypatch.setattr(runner.torch, "set_float32_matmul_precision", calls.append)
runner.configure_torch(allow_tf32=False)
assert calls == []
def test_configure_torch_enables_tf32_when_allowed_on_cuda(monkeypatch):
calls = []
original_fp32_precision = getattr(runner.torch.backends, "fp32_precision", None)
monkeypatch.setattr(runner.torch.cuda, "is_available", lambda: True)
monkeypatch.setattr(runner.torch, "set_float32_matmul_precision", calls.append)
try:
runner.configure_torch(allow_tf32=True)
assert runner.torch.backends.fp32_precision == "tf32"
assert calls == ["high"]
finally:
if original_fp32_precision is not None:
runner.torch.backends.fp32_precision = original_fp32_precision
def test_configure_torch_skips_tf32_without_cuda(monkeypatch):
calls = []
monkeypatch.setattr(runner.torch.cuda, "is_available", lambda: False)
monkeypatch.setattr(runner.torch, "set_float32_matmul_precision", calls.append)
runner.configure_torch(allow_tf32=True)
assert calls == []
def test_save_images_defaults_to_png_for_small_images(tmp_path):
image = FakeImage((4096, 4096))
runner.save_images([image], tmp_path / "output", num_images_per_prompt=1)
assert image.saved == [(tmp_path / "output.png", {})]
def test_save_images_defaults_to_bigtiff_for_large_images(tmp_path):
image = FakeImage((4097, 4096))
runner.save_images([image], tmp_path / "output", num_images_per_prompt=1)
assert image.saved == [(tmp_path / "output.tif", {"format": "TIFF", "big_tiff": True})]
def test_save_images_switches_large_png_output_to_bigtiff(tmp_path):
image = FakeImage((4097, 4096))
with pytest.warns(UserWarning, match="saving BigTIFF"):
runner.save_images([image], tmp_path / "output.png", num_images_per_prompt=1)
assert image.saved == [(tmp_path / "output.tif", {"format": "TIFF", "big_tiff": True})]
def test_save_images_preserves_numbered_large_bigtiff_stems(tmp_path):
images = [FakeImage((4097, 4096)), FakeImage((4097, 4096))]
runner.save_images(images, tmp_path / "output", num_images_per_prompt=2)
assert images[0].saved == [(tmp_path / "output_0.tif", {"format": "TIFF", "big_tiff": True})]
assert images[1].saved == [(tmp_path / "output_1.tif", {"format": "TIFF", "big_tiff": True})]
def test_quantization_flag_without_value_uses_float8_weight_only_default():
args = _args("--quantization")
assert args.quantization == "float8wo"
def test_quantization_parser_accepts_component_overrides():
args = _args(
"--quantization",
"float8wo",
"--transformer-quantization",
"int4wo",
"--text-encoder-quantization",
"none",
"--vae-quantization",
"float8dyn",
)
assert args.quantization == "float8wo"
assert args.transformer_quantization == "int4wo"
assert args.text_encoder_quantization == "none"
assert args.vae_quantization == "float8dyn"
def test_resolve_quantization_mapping_applies_base_strategy_to_all_quantizable_components():
mapping = runner.resolve_quantization_mapping("float8wo")
assert mapping == {
"transformer": "float8wo",
"text_encoder": "float8wo",
"vae": "float8wo",
}
def test_resolve_quantization_mapping_uses_component_overrides():
mapping = runner.resolve_quantization_mapping(
"float8wo",
transformer_quantization="int4wo",
text_encoder_quantization="none",
vae_quantization="float8dyn",
)
assert mapping == {"transformer": "int4wo", "vae": "float8dyn"}
def test_build_quantization_config_returns_none_when_disabled():
assert runner.build_quantization_config("none") is None
def test_build_quantization_config_uses_component_specific_torchao_configs():
pytest.importorskip("torchao")
config = runner.build_quantization_config(
"float8wo",
transformer_quantization="int4wo",
text_encoder_quantization="int8wo",
vae_quantization="float8dyn",
)
assert set(config.quant_mapping) == {"transformer", "text_encoder", "vae"}
assert type(config.quant_mapping["transformer"].quant_type).__name__ == "Int4WeightOnlyConfig"
assert type(config.quant_mapping["text_encoder"].quant_type).__name__ == "Int8WeightOnlyConfig"
assert type(config.quant_mapping["vae"].quant_type).__name__ == "Float8DynamicActivationFloat8WeightConfig"
@pytest.mark.parametrize("weighting_type", ["none", "linear", "cosine"])
def test_call_kwargs_forward_all_weighting_types(weighting_type):
args = _args("--weighting-type", weighting_type)
generator = object()
call_kwargs = runner.build_call_kwargs(args, SimpleNamespace(), generator)
assert call_kwargs["weighting_type"] == weighting_type
assert "num_inference_steps" not in call_kwargs
assert call_kwargs["generator"] is generator
def test_call_kwargs_forward_offset_and_panorama_options():
call_kwargs = runner.build_call_kwargs(
_args(
"--window-stride-height-offset",
"128",
"--window-stride-width-offset",
"256",
"--panorama-width",
"--panorama-height",
),
SimpleNamespace(),
object(),
)
assert call_kwargs["window_stride_height_offset"] == 128
assert call_kwargs["window_stride_width_offset"] == 256
assert call_kwargs["panorama_width"] is True
assert call_kwargs["panorama_height"] is True
def test_call_kwargs_forward_window_batch_size():
call_kwargs = runner.build_call_kwargs(
_args("--batch-size", "6"),
SimpleNamespace(),
object(),
)
assert call_kwargs["window_batch_size"] == 6
def test_call_kwargs_rejects_invalid_window_batch_size():
with pytest.raises(ValueError, match="batch-size"):
runner.build_call_kwargs(
_args("--batch-size", "0"),
SimpleNamespace(),
object(),
)
def test_csv_prompt_loads_regional_masks_in_csv_order(tmp_path):
masks_dir = tmp_path / "masks"
masks_dir.mkdir()
Image.new("L", (32, 32), 255).save(masks_dir / "00.png")
Image.new("L", (32, 32), 255).save(masks_dir / "08.png")
csv_path = tmp_path / "prompts.csv"
csv_path.write_text('mask,prompt\n08.png,"a forest"\n00.png,"a mountain"\n', encoding="utf-8")
call_kwargs = runner.build_call_kwargs(
_args("--prompt", str(csv_path), "--masks", str(masks_dir), "--height", "32", "--width", "32"),
SimpleNamespace(vae_scale_factor=8),
object(),
)
assert call_kwargs["prompt"] == ["", "a forest", "a mountain"]
assert torch.equal(call_kwargs["regional_masks"], torch.ones((2, 2, 2)))
def test_csv_prompt_requires_masks_folder(tmp_path):
csv_path = tmp_path / "prompts.csv"
csv_path.write_text('mask,prompt\n00.png,"a mountain"\n', encoding="utf-8")
with pytest.raises(ValueError, match="--masks"):
runner.build_call_kwargs(
_args("--prompt", str(csv_path), "--height", "32", "--width", "32"),
SimpleNamespace(vae_scale_factor=8),
object(),
)
def test_regional_mask_loader_warns_for_non_binary_masks(tmp_path):
masks_dir = tmp_path / "masks"
masks_dir.mkdir()
Image.new("L", (32, 32), 128).save(masks_dir / "00.png")
csv_path = tmp_path / "prompts.csv"
csv_path.write_text('mask,prompt\n00.png,"a foggy valley"\n', encoding="utf-8")
with pytest.warns(UserWarning, match="not binary"):
call_kwargs = runner.build_call_kwargs(
_args("--prompt", str(csv_path), "--masks", str(masks_dir), "--height", "32", "--width", "32"),
SimpleNamespace(vae_scale_factor=8),
object(),
)
assert torch.equal(call_kwargs["regional_masks"], torch.full((1, 2, 2), 128 / 255))
def test_regional_mask_loader_rejects_wrong_size(tmp_path):
masks_dir = tmp_path / "masks"
masks_dir.mkdir()
Image.new("L", (16, 32), 255).save(masks_dir / "00.png")
csv_path = tmp_path / "prompts.csv"
csv_path.write_text('mask,prompt\n00.png,"a mountain"\n', encoding="utf-8")
with pytest.raises(ValueError, match="must have size"):
runner.build_call_kwargs(
_args("--prompt", str(csv_path), "--masks", str(masks_dir), "--height", "32", "--width", "32"),
SimpleNamespace(vae_scale_factor=8),
object(),
)
@pytest.mark.parametrize(
("config", "expected_class_name"),
[
({"_class_name": "Flux2Pipeline"}, "Flux2MultiDiffusionAutoBlocks"),
({"_class_name": "Flux2KleinPipeline", "is_distilled": True}, "Flux2KleinMultiDiffusionAutoBlocks"),
({"_class_name": "Flux2KleinPipeline", "is_distilled": False}, "Flux2KleinBaseMultiDiffusionAutoBlocks"),
({"_class_name": "Flux2KleinPipeline"}, "Flux2KleinBaseMultiDiffusionAutoBlocks"),
],
)
def test_multidiffusion_blocks_are_selected_from_loaded_pipeline_config(config, expected_class_name):
assert runner.build_multidiffusion_blocks(config).__class__.__name__ == expected_class_name
def test_explicit_num_inference_steps_are_forwarded():
call_kwargs = runner.build_call_kwargs(
_args("--num-inference-steps", "7"),
SimpleNamespace(),
object(),
{"_class_name": "Flux2KleinPipeline", "is_distilled": True},
)
assert call_kwargs["num_inference_steps"] == 7
def test_distilled_klein_default_num_inference_steps_is_forwarded():
call_kwargs = runner.build_call_kwargs(
_args(),
SimpleNamespace(),
object(),
{"_class_name": "Flux2KleinPipeline", "is_distilled": True},
)
assert call_kwargs["num_inference_steps"] == 4
@pytest.mark.parametrize(
"config",
[
{"_class_name": "Flux2Pipeline"},
{"_class_name": "Flux2KleinPipeline", "is_distilled": False},
{"_class_name": "Flux2KleinModularPipeline", "is_distilled": False},
{"_class_name": "Flux2KleinPipeline"},
{},
],
)
def test_non_distilled_models_keep_pipeline_num_inference_steps_default(config):
call_kwargs = runner.build_call_kwargs(
_args(),
SimpleNamespace(),
object(),
config,
)
assert "num_inference_steps" not in call_kwargs
def test_enable_tiling_and_slicing_call_vae_methods():
class FakeVae:
def __init__(self):
self.tiling_enabled = False
self.slicing_enabled = False
def enable_tiling(self):
self.tiling_enabled = True
def enable_slicing(self):
self.slicing_enabled = True
pipe = SimpleNamespace(vae=FakeVae())
runner.apply_vae_memory_options(pipe, enable_tiling=True, enable_slicing=True)
assert pipe.vae.tiling_enabled
assert pipe.vae.slicing_enabled
def test_init_selects_blocks_from_model_config_and_keeps_model_guidance_default(monkeypatch):
calls = []
class FakeBlocks:
def init_pipeline(self, base_model, components_manager):
calls.append((base_model, components_manager))
return FakePipe()
class FakePipe:
def __init__(self):
self.guider = object()
self.updated_components = None
def load_components(self, **kwargs):
self.load_kwargs = kwargs
def update_components(self, **components):
self.updated_components = components
monkeypatch.setattr(
runner, "build_multidiffusion_blocks", lambda model_config: calls.append(model_config) or FakeBlocks()
)
pipe = runner.init_modular_pipeline(
base_model="black-forest-labs/FLUX.2-klein-base-4B",
model_config={"_class_name": "Flux2KleinPipeline"},
guidance_scale=None,
dtype="float32",
device="cpu",
local_files_only=True,
compile=False,
)
assert calls[0] == {"_class_name": "Flux2KleinPipeline"}
assert calls[1][0] == "black-forest-labs/FLUX.2-klein-base-4B"
assert pipe.updated_components is None
def test_init_applies_explicit_guidance_scale(monkeypatch):
class FakeGuiderSpec:
@staticmethod
def create(guidance_scale):
return {"guidance_scale": guidance_scale}
class FakePipe:
def __init__(self):
self.guider = object()
self.updated_components = None
def load_components(self, **kwargs):
self.load_kwargs = kwargs
def get_component_spec(self, name):
assert name == "guider"
return FakeGuiderSpec()
def update_components(self, **components):
self.updated_components = components
monkeypatch.setattr(
runner,
"build_multidiffusion_blocks",
lambda model_config: SimpleNamespace(init_pipeline=lambda base_model, components_manager: FakePipe()),
)
pipe = runner.init_modular_pipeline(
base_model="black-forest-labs/FLUX.2-klein-base-4B",
model_config={"_class_name": "Flux2KleinPipeline"},
guidance_scale=2.5,
dtype="float32",
device="cpu",
local_files_only=True,
compile=False,
)
assert pipe.updated_components == {"guider": {"guidance_scale": 2.5}}
def test_remote_parser_requires_hub_repo_id():
args = remote_runner.parse_args(["--repo-id", "user/multidiff-modular"])
assert args.repo_id == "user/multidiff-modular"
def test_remote_init_uses_modular_from_pretrained_with_remote_code(monkeypatch):
calls = {}
class FakePipe:
def __init__(self):
self.guider = None
def load_components(self, **kwargs):
calls["load_components"] = kwargs
def fake_from_pretrained(repo_id, **kwargs):
calls["repo_id"] = repo_id
calls["from_pretrained"] = kwargs
return FakePipe()
monkeypatch.setattr(remote_runner.ModularPipeline, "from_pretrained", fake_from_pretrained)
pipe = remote_runner.init_remote_modular_pipeline(
repo_id="user/multidiff-modular",
guidance_scale=None,
dtype="float32",
device="cpu",
local_files_only=True,
compile=False,
)
assert pipe.guider is None
assert calls["repo_id"] == "user/multidiff-modular"
assert calls["from_pretrained"]["trust_remote_code"] is True
assert calls["from_pretrained"]["local_files_only"] is True
assert calls["load_components"]["torch_dtype"] is runner.DTYPE_MAP["float32"]
assert calls["load_components"]["local_files_only"] is True
def test_image_conditioning_loads_and_forwards_image(monkeypatch):
loaded_image = object()
seen_paths = []
def fake_load_image(path):
seen_paths.append(path)
return loaded_image
monkeypatch.setattr(runner, "load_image", fake_load_image)
call_kwargs = runner.build_call_kwargs(
_args("--image-conditioning", "conditioning.png"),
SimpleNamespace(config={}),
object(),
)
assert seen_paths == ["conditioning.png"]
assert call_kwargs["image"] is loaded_image
def test_image_img2img_prepares_latents_and_forwards_strength(monkeypatch):
loaded_image = object()
raw_latents = object()
generator = object()
vae = object()
class FakeImageProcessor:
def __init__(self):
self.calls = []
def preprocess(self, image, *, height, width, resize_mode):
self.calls.append((image, height, width, resize_mode))
return "preprocessed-image"
class FakeEncoderPipe:
def __init__(self):
self.updated_components = None
self.calls = []
def update_components(self, **components):
self.updated_components = components
def __call__(self, *, condition_images, generator):
self.calls.append((condition_images, generator))
return SimpleNamespace(image_latents=[raw_latents])
encoder_pipe = FakeEncoderPipe()
encoder_block = SimpleNamespace(init_pipeline=lambda: encoder_pipe)
img_conditioning_block = SimpleNamespace(sub_blocks={"encode": encoder_block})
vae_encoder_block = SimpleNamespace(sub_blocks={"img_conditioning": img_conditioning_block})
image_processor = FakeImageProcessor()
pipe = SimpleNamespace(
config={},
vae=vae,
vae_scale_factor=8,
blocks=SimpleNamespace(sub_blocks={"vae_encoder": vae_encoder_block}),
image_processor=image_processor,
)
monkeypatch.setattr(runner, "load_image", lambda path: loaded_image)
monkeypatch.setattr(
runner.Flux2PrepareImageLatentsStep,
"_pack_latents",
staticmethod(lambda latents: ("packed", latents)),
)
call_kwargs = runner.build_call_kwargs(
_args("--image-img2img", "init.png", "--height", "65", "--width", "81", "--strength", "0.42"),
pipe,
generator,
)
assert image_processor.calls == [(loaded_image, 64, 80, "default")]
assert encoder_pipe.updated_components == {"vae": vae}
assert encoder_pipe.calls == [(["preprocessed-image"], generator)]
assert call_kwargs["image_img2img"] == ("packed", raw_latents)
assert call_kwargs["strength"] == 0.42
|