File size: 18,720 Bytes
31112ad | 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 | import os
import torch
from PIL import Image
from shared.utils import files_locator as fl
from shared.utils.hf import build_hf_url
def test_flux2(base_model_type):
return base_model_type in ["flux2_dev", "pi_flux2", "flux2_klein_4b", "flux2_klein_9b"]
def get_text_encoder_name(base_model_type, text_encoder_quantization):
if base_model_type == "flux2_klein_4b":
if text_encoder_quantization == "int8":
return "qwen3_quanto_bf16_int8.safetensors"
return "qwen3_bf16.safetensors"
if base_model_type == "flux2_klein_9b":
if text_encoder_quantization == "int8":
return "qwen3_8b_quanto_bf16_int8.safetensors"
return "qwen3_8b_bf16.safetensors"
if text_encoder_quantization == "int8":
return "mistral3_small_quanto_bf16_int8.safetensors" if test_flux2(base_model_type) else "T5_xxl_1.1_enc_quanto_bf16_int8.safetensors"
return "mistral3_small_bf16.safetensors" if test_flux2(base_model_type) else "T5_xxl_1.1_enc_bf16.safetensors"
class family_handler():
@staticmethod
def query_supported_types():
return [
"flux",
"flux2_dev",
"pi_flux2",
"flux2_klein_4b",
"flux2_klein_9b",
"flux_chroma",
"flux_chroma_radiance",
"flux_dev_kontext",
"flux_dev_umo",
"flux_dev_uso",
"flux_schnell",
"flux_dev_kontext_dreamomni2",
]
@staticmethod
def query_family_maps():
models_eqv_map = {
"flux_dev_kontext" : "flux",
"flux_dev_umo" : "flux",
"flux_dev_uso" : "flux",
"flux_schnell" : "flux",
"flux_chroma" : "flux",
"flux_chroma_radiance": "flux",
"flux_dev_kontext_dreamomni2": "flux",
"flux2_dev": "flux",
"pi_flux2": "flux",
"flux2_klein_4b": "flux",
"flux2_klein_9b": "flux",
}
models_comp_map = {
"flux": ["flux2_dev", "pi_flux2", "flux2_klein_4b", "flux2_klein_9b", "flux_chroma", "flux_chroma_radiance", "flux_dev_kontext", "flux_dev_umo", "flux_dev_uso", "flux_schnell", "flux_dev_kontext_dreamomni2" ]
}
return models_eqv_map, models_comp_map
@staticmethod
def query_model_def(base_model_type, model_def):
flux_model = "flux-dev" if base_model_type == "flux" else base_model_type.replace("_", "-")
pi_flux2 = flux_model == "pi-flux2"
flux2_klein_4b = base_model_type == "flux2_klein_4b"
flux2_klein_9b = base_model_type == "flux2_klein_9b"
flux2_klein = flux2_klein_4b or flux2_klein_9b
flux2 = flux_model.startswith("flux2") or pi_flux2
flux_schnell = flux_model == "flux-schnell"
flux_chroma = flux_model == "flux-chroma"
flux_chroma_radiance = flux_model == "flux-chroma-radiance"
flux_uso = flux_model == "flux-dev-uso"
flux_umo = flux_model == "flux-dev-umo"
flux_kontext = flux_model == "flux-dev-kontext"
flux_kontext_dreamomni2 = flux_model == "flux-dev-kontext-dreamomni2"
extra_model_def = {
"image_outputs" : True,
"no_negative_prompt" : flux2 or not (flux_chroma or flux_chroma_radiance),
"flux-model": flux_model,
}
if flux2:
if flux2_klein:
if flux2_klein_4b:
text_encoder_folder = "Qwen3"
text_encoder_repo = "DeepBeepMeep/Z-Image"
text_encoder_urls = [
build_hf_url(text_encoder_repo, text_encoder_folder, "qwen3_bf16.safetensors"),
build_hf_url(text_encoder_repo, text_encoder_folder, "qwen3_quanto_bf16_int8.safetensors"),
]
else:
text_encoder_folder = "qwen3_8b"
text_encoder_repo = "DeepBeepMeep/Flux2"
text_encoder_urls = [
build_hf_url(text_encoder_repo, text_encoder_folder, "qwen3_8b_bf16.safetensors"),
build_hf_url(text_encoder_repo, text_encoder_folder, "qwen3_8b_quanto_bf16_int8.safetensors"),
]
extra_model_def["text_encoder_type"] = "qwen3"
extra_model_def["text_encoder_URLs"] = text_encoder_urls
else:
text_encoder_folder = "mistral3small"
extra_model_def["text_encoder_URLs"] = [
build_hf_url("DeepBeepMeep/Flux2", text_encoder_folder, "mistral3_small_bf16.safetensors"),
build_hf_url("DeepBeepMeep/Flux2", text_encoder_folder, "mistral3_small_quanto_bf16_int8.safetensors"),
]
extra_model_def["text_encoder_type"] = "mistral3"
extra_model_def["text_encoder_folder"] = text_encoder_folder
else:
text_encoder_folder = "T5_xxl_1.1"
extra_model_def["text_encoder_URLs"] = [
build_hf_url("DeepBeepMeep/LTX_Video", text_encoder_folder, "T5_xxl_1.1_enc_bf16.safetensors"),
build_hf_url("DeepBeepMeep/LTX_Video", text_encoder_folder, "T5_xxl_1.1_enc_quanto_bf16_int8.safetensors"),
]
extra_model_def["text_encoder_folder"] = text_encoder_folder
if flux2_klein:
extra_model_def["profiles_dir"] = ["flux2_klein_4b"] if flux2_klein_4b else ["flux2_klein_9b"]
else:
extra_model_def["profiles_dir"] = [] if (flux_schnell or flux2) else ["flux"]
if flux_chroma or flux_chroma_radiance:
extra_model_def["guidance_max_phases"] = 1
if flux_chroma_radiance:
extra_model_def["radiance"] = True
elif not flux_schnell and not flux2_klein:
extra_model_def["embedded_guidance"] = True
if flux_uso :
extra_model_def["any_image_refs_relative_size"] = True
extra_model_def["no_background_removal"] = True
extra_model_def["image_ref_choices"] = {
"choices":[("First Image is a Reference Image, and then the next ones (up to two) are Style Images", "KI"),
("Up to two Images are Style Images", "KIJ")],
"default": "KI",
"letters_filter": "KIJ",
"label": "Reference Images / Style Images"
}
if flux_kontext or flux_kontext_dreamomni2 or flux2:
extra_model_def["inpaint_support"] = flux_kontext
extra_model_def["image_ref_choices"] = {
"choices": [
("None", ""),
("Conditional Image is first Main Subject / Landscape and may be followed by People / Objects", "KI"),
("Conditional Images are People / Objects", "I"),
],
"letters_filter": "KI",
}
if flux_kontext_dreamomni2:
extra_model_def["no_background_removal"] = True
else:
extra_model_def["background_removal_label"]= "Remove Backgrounds only behind People / Objects except main Subject / Landscape"
elif flux_umo:
extra_model_def["image_ref_choices"] = {
"choices": [
("Conditional Images are People / Objects", "I"),
],
"letters_filter": "I",
"visible": False
}
if flux2:
extra_model_def["group"] ="flux2"
extra_model_def["no_background_removal"] = True
# extra_model_def["inpaint_support"] = True
extra_model_def["mask_preprocessing"] = {
"selection":[ ""],
"visible": False
}
extra_model_def["mask_strength_always_enabled"] = True
extra_model_def["guide_preprocessing"] = {
"selection": ["", "PV", "MV"],
}
extra_model_def["mask_preprocessing"] = {
"selection": ["", "A", "NA"],
"visible": True,
}
# extra_model_def["guide_inpaint_color"] = 0
# extra_model_def["video_guide_outpainting"] = [1,2]
if pi_flux2:
extra_model_def["piflow"] = True
extra_model_def["inpaint_support"] = True
extra_model_def["fit_into_canvas_image_refs"] = 0
return extra_model_def
@staticmethod
def get_rgb_factors(base_model_type ):
if base_model_type in ["flux_chroma_radiance"]:
return None, None
from shared.RGB_factors import get_rgb_factors
latent_rgb_factors, latent_rgb_factors_bias = get_rgb_factors("flux", sub_family= "flux2" if test_flux2(base_model_type) else "flux")
return latent_rgb_factors, latent_rgb_factors_bias
@staticmethod
def preview_latents(base_model_type, latents, meta):
if base_model_type != "flux_chroma_radiance":
return None
from .sampling import patches_to_image
tensor = latents
if not torch.is_tensor(tensor):
return None
tensor = tensor.detach()
C, T, H, W = tensor.shape
image = tensor.cpu().clamp(-1, 1)
image = image.permute(0, 2, 1, 3) # (C, T, H, W) -> (C, H, T, W)
image = image.reshape(C, H, T * W) # (C, H, T, W) -> (C, H, T*W)
image = image.add(1).mul(127.5).clamp(0, 255).to(torch.uint8)
image = image.permute(1, 2, 0).numpy()
preview = Image.fromarray(image)
if preview.height > 0:
scale = 200 / preview.height
width_px = max(1, int(round(preview.width * scale)))
resampling_module = getattr(Image, "Resampling", Image)
resample_filter = getattr(resampling_module, "BILINEAR", Image.BILINEAR)
preview = preview.resize((width_px, 200), resample=resample_filter)
return preview
@staticmethod
def query_model_family():
return "flux"
@staticmethod
def query_family_infos():
return {"flux":(100, "Flux 1"), "flux2":(101, "Flux 2")}
@staticmethod
def register_lora_cli_args(parser):
parser.add_argument(
"--lora-dir-flux",
type=str,
default=os.path.join("loras", "flux"),
help="Path to a directory that contains flux images Loras"
)
parser.add_argument(
"--lora-dir-flux2",
type=str,
default=os.path.join("loras", "flux2"),
help="Path to a directory that contains flux2 images Loras"
)
parser.add_argument(
"--lora-dir-flux2-klein-4b",
type=str,
default=os.path.join("loras", "flux2_klein_4b"),
help="Path to a directory that contains Flux 2 Klein 4B Loras"
)
parser.add_argument(
"--lora-dir-flux2-klein-9b",
type=str,
default=os.path.join("loras", "flux2_klein_9b"),
help="Path to a directory that contains Flux 2 Klein 9B Loras"
)
@staticmethod
def get_lora_dir(base_model_type, args):
if base_model_type == "flux2_klein_4b":
return args.lora_dir_flux2_klein_4b
if base_model_type == "flux2_klein_9b":
return args.lora_dir_flux2_klein_9b
if test_flux2(base_model_type):
return args.lora_dir_flux2
return args.lora_dir_flux
@staticmethod
def query_model_files(computeList, base_model_type, model_def=None):
if base_model_type in ["flux2_klein_4b", "flux2_klein_9b"]:
if base_model_type == "flux2_klein_4b":
text_encoder_folder = "Qwen3"
text_encoder_repo = "DeepBeepMeep/Z-Image"
else:
text_encoder_folder = "qwen3_8b"
text_encoder_repo = "DeepBeepMeep/Flux2"
tokenizer_files = ["config.json", "generation_config.json", "added_tokens.json", "chat_template.jinja", "merges.txt", "special_tokens_map.json", "tokenizer.json", "tokenizer_config.json", "vocab.json"]
ret = [
{
"repoId": text_encoder_repo,
"sourceFolderList": [text_encoder_folder],
"fileList": [tokenizer_files],
},
{
"repoId": "DeepBeepMeep/Flux2",
"sourceFolderList": [""],
"fileList": [["flux2_vae.safetensors"]],
},
]
elif test_flux2(base_model_type):
ret = [
{
"repoId": "DeepBeepMeep/Flux2",
"sourceFolderList": ["mistral3small", ""],
"fileList": [
[ "tokenizer.json", "tokenizer_config.json", "special_tokens_map.json", "processor_config.json", "config.json", "preprocessor_config.json", "chat_template.jinja", ],
[ "flux2_vae.safetensors", ],
],
}
]
else:
ret = [
{
"repoId" : "DeepBeepMeep/LTX_Video",
"sourceFolderList" : ["T5_xxl_1.1"],
"fileList" : [ ["added_tokens.json", "special_tokens_map.json", "spiece.model", "tokenizer_config.json"] ]
},
{
"repoId" : "DeepBeepMeep/HunyuanVideo",
"sourceFolderList" : [ "clip_vit_large_patch14", ],
"fileList" :[
["config.json", "merges.txt", "model.safetensors", "preprocessor_config.json", "special_tokens_map.json", "tokenizer.json", "tokenizer_config.json", "vocab.json"],
]
},
{
"repoId" : "DeepBeepMeep/Flux",
"sourceFolderList" : ["",],
"fileList" : [ ["flux_vae.safetensors"] ]
}]
if base_model_type in ["flux_dev_uso"]:
ret += [
{
"repoId" : "DeepBeepMeep/Flux",
"sourceFolderList" : ["siglip-so400m-patch14-384"],
"fileList" : [ ["config.json", "preprocessor_config.json", "model.safetensors"] ]
}]
if base_model_type in ["flux_dev_kontext_dreamomni2"]:
ret += [
{
"repoId" : "DeepBeepMeep/Flux",
"sourceFolderList" : ["Qwen2.5-VL-7B-DreamOmni2"],
"fileList" : [ ["Qwen2.5-VL-7B-DreamOmni2_quanto_bf16_int8.safetensors", "merges.txt", "tokenizer_config.json", "config.json", "vocab.json", "video_preprocessor_config.json", "preprocessor_config.json", "chat_template.jinja"] ]
}]
return ret
@staticmethod
def load_model(model_filename, model_type, base_model_type, model_def, quantizeTransformer = False, text_encoder_quantization = None, dtype = torch.bfloat16, VAE_dtype = torch.float32, mixed_precision_transformer = False, save_quantized = False, submodel_no_list = None, text_encoder_filename = None):
from .flux_main import model_factory
flux_model = model_factory(
checkpoint_dir="ckpts",
model_filename=model_filename,
model_type = model_type,
model_def = model_def,
base_model_type=base_model_type,
text_encoder_filename= text_encoder_filename,
quantizeTransformer = quantizeTransformer,
dtype = dtype,
VAE_dtype = VAE_dtype,
mixed_precision_transformer = mixed_precision_transformer,
save_quantized = save_quantized
)
pipe = { "transformer": flux_model.model, "vae" : flux_model.vae}
if getattr(flux_model, "clip", None) is not None:
pipe["text_encoder"] = flux_model.clip
if getattr(flux_model, "t5", None) is not None:
pipe["text_encoder_2"] = flux_model.t5
if getattr(flux_model, "mistral", None) is not None:
pipe["text_encoder"] = flux_model.mistral.model
if flux_model.vision_encoder is not None:
pipe["siglip_model"] = flux_model.vision_encoder
if flux_model.feature_embedder is not None:
pipe["feature_embedder"] = flux_model.feature_embedder
if flux_model.vlm_model is not None:
pipe["vlm_model"] = flux_model.vlm_model
return flux_model, pipe
@staticmethod
def fix_settings(base_model_type, settings_version, model_def, ui_defaults):
flux_model = model_def.get("flux-model", "flux-dev")
flux_uso = flux_model == "flux-dev-uso"
if flux_uso and settings_version < 2.29:
video_prompt_type = ui_defaults.get("video_prompt_type", "")
if "I" in video_prompt_type:
video_prompt_type = video_prompt_type.replace("I", "KI")
ui_defaults["video_prompt_type"] = video_prompt_type
if settings_version < 2.34:
ui_defaults["denoising_strength"] = 1.
if flux_model.startswith("flux2"):
ui_defaults["embedded_guidance_scale"] = ui_defaults.get("embedded_guidance_scale", 4.0)
@staticmethod
def update_default_settings(base_model_type, model_def, ui_defaults):
flux_model = model_def.get("flux-model", "flux-dev")
flux_uso = flux_model == "flux-dev-uso"
flux_umo = flux_model == "flux-dev-umo"
flux_kontext = flux_model == "flux-dev-kontext"
flux_kontext_dreamomni2 = flux_model == "flux-dev-kontext-dreamomni2"
flux2 = flux_model.startswith("flux2")
flux2_klein = base_model_type in ["flux2_klein_4b", "flux2_klein_9b"]
ui_defaults.update({
"embedded_guidance_scale": 2.5,
})
if flux2:
ui_defaults.update({
"embedded_guidance_scale": 1.0 if flux2_klein else 4.0,
"denoising_strength": 1.0,
"masking_strength": 0.25,
"remove_background_images_ref" : 0,
})
if flux2_klein:
ui_defaults["num_inference_steps"] = 4
if flux_kontext or flux_uso or flux_kontext_dreamomni2:
ui_defaults.update({
"video_prompt_type": "KI",
"denoising_strength": 1.,
})
elif flux_umo:
ui_defaults.update({
"video_prompt_type": "I",
"remove_background_images_ref": 0,
})
|