File size: 26,693 Bytes
d0ab6a4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d85bcb0
d0ab6a4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
405bf72
 
d0ab6a4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d85bcb0
 
d0ab6a4
 
 
 
 
 
 
 
 
d85bcb0
d0ab6a4
 
 
 
 
 
 
 
 
d85bcb0
d0ab6a4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import spaces
import argparse
from pathlib import Path
import os
import torch
from diffusers import (DiffusionPipeline, AutoencoderKL, FlowMatchEulerDiscreteScheduler, StableDiffusionXLPipeline, StableDiffusionPipeline,
                       FluxPipeline, FluxTransformer2DModel, SD3Transformer2DModel, StableDiffusion3Pipeline)
from transformers import CLIPTokenizer, CLIPTextModel, CLIPTextModelWithProjection, CLIPFeatureExtractor, AutoTokenizer, T5EncoderModel, BitsAndBytesConfig as TFBitsAndBytesConfig
from huggingface_hub import save_torch_state_dict, snapshot_download
from diffusers.loaders.single_file_utils import (convert_flux_transformer_checkpoint_to_diffusers, convert_sd3_transformer_checkpoint_to_diffusers,
                                                 convert_sd3_t5_checkpoint_to_diffusers)
from diffusers.pipelines.stable_diffusion import StableDiffusionSafetyChecker
import safetensors.torch
import gradio as gr
import shutil
import gc
import tempfile
# also requires aria, gdown, peft, huggingface_hub, safetensors, transformers, accelerate, pytorch_lightning
from utils import (get_token, set_token, is_repo_exists, is_repo_name, get_download_file, upload_repo, gate_repo)
from sdutils import (SCHEDULER_CONFIG_MAP, get_scheduler_config, fuse_loras, DTYPE_DEFAULT, get_dtype, get_dtypes, get_model_type_from_key, get_process_dtype)


@spaces.GPU
def fake_gpu():
    pass


try:
    from diffusers import BitsAndBytesConfig
    is_nf4 = True
except Exception:
    is_nf4 = False


FLUX_BASE_REPOS = ["camenduru/FLUX.1-dev-diffusers", "black-forest-labs/FLUX.1-schnell", "John6666/flux1-dev-fp8-flux", "John6666/flux1-schnell-fp8-flux"]
FLUX_T5_URL = "https://huggingface.co/camenduru/FLUX.1-dev/blob/main/t5xxl_fp8_e4m3fn.safetensors"
SD35_BASE_REPOS = ["adamo1139/stable-diffusion-3.5-large-ungated", "adamo1139/stable-diffusion-3.5-large-turbo-ungated"]
SD35_T5_URL = "https://huggingface.co/adamo1139/stable-diffusion-3.5-large-turbo-ungated/blob/main/text_encoders/t5xxl_fp8_e4m3fn.safetensors"
TEMP_DIR = tempfile.mkdtemp()
IS_ZERO = os.environ.get("SPACES_ZERO_GPU") is not None
IS_CUDA = torch.cuda.is_available()


def safe_clean(path: str):
    try:
        if Path(path).exists():
            if Path(path).is_dir(): shutil.rmtree(str(Path(path)))
            else: Path(path).unlink()
            print(f"Deleted: {path}")
        else: print(f"File not found: {path}")
    except Exception as e:
        print(f"Failed to delete: {path} {e}")


def save_readme_md(dir, url):
    orig_url = ""
    orig_name = ""
    if is_repo_name(url): 
        orig_name = url
        orig_url = f"https://huggingface.co/{url}/"
    elif "http" in url:
        orig_name = url
        orig_url = url
    if orig_name and orig_url:
       md = f"""---

license: other

language:

- en

library_name: diffusers

pipeline_tag: text-to-image

tags:

- text-to-image

---

Converted from [{orig_name}]({orig_url}).

"""
    else:
        md = f"""---

license: other

language:

- en

library_name: diffusers

pipeline_tag: text-to-image

tags:

- text-to-image

---

"""
    path = str(Path(dir, "README.md"))
    with open(path, mode='w', encoding="utf-8") as f:
        f.write(md)


def save_module(model, name: str, dir: str, dtype: str="fp8", progress=gr.Progress(track_tqdm=True)): # doesn't work
    if name in ["vae", "transformer", "unet"]: pattern = "diffusion_pytorch_model{suffix}.safetensors"
    else: pattern = "model{suffix}.safetensors"
    if name in ["transformer", "unet"]: size = "10GB"
    else: size = "5GB"
    path = str(Path(f"{dir.removesuffix('/')}/{name}"))
    os.makedirs(path, exist_ok=True)
    progress(0, desc=f"Saving {name} to {dir}...")
    print(f"Saving {name} to {dir}...")
    model.to("cpu")
    sd = dict(model.state_dict())
    new_sd = {}
    for key in list(sd.keys()):
        q = sd.pop(key)
        if dtype == "fp8": new_sd[key] = q if q.dtype == torch.float8_e4m3fn else q.to(torch.float8_e4m3fn)
        else: new_sd[key] = q
    del sd
    gc.collect()
    save_torch_state_dict(state_dict=new_sd, save_directory=path, filename_pattern=pattern, max_shard_size=size)
    del new_sd
    gc.collect()


def save_module_sd(sd: dict, name: str, dir: str, dtype: str="fp8", progress=gr.Progress(track_tqdm=True)):
    if name in ["vae", "transformer", "unet"]: pattern = "diffusion_pytorch_model{suffix}.safetensors"
    else: pattern = "model{suffix}.safetensors"
    if name in ["transformer", "unet"]: size = "10GB"
    else: size = "5GB"
    path = str(Path(f"{dir.removesuffix('/')}/{name}"))
    os.makedirs(path, exist_ok=True)
    progress(0, desc=f"Saving state_dict of {name} to {dir}...")
    print(f"Saving state_dict of {name} to {dir}...")
    new_sd = {}
    for key in list(sd.keys()):
        q = sd.pop(key).to("cpu")
        if dtype == "fp8": new_sd[key] = q if q.dtype == torch.float8_e4m3fn else q.to(torch.float8_e4m3fn)
        else: new_sd[key] = q
    save_torch_state_dict(state_dict=new_sd, save_directory=path, filename_pattern=pattern, max_shard_size=size)
    del new_sd
    gc.collect()


def convert_flux_fp8_cpu(new_file: str, new_dir: str, dtype: str, base_repo: str, civitai_key: str, kwargs: dict, progress=gr.Progress(track_tqdm=True)):
    temp_dir = TEMP_DIR
    down_dir = str(Path(f"{TEMP_DIR}/down"))
    os.makedirs(down_dir, exist_ok=True)
    hf_token = get_token()
    progress(0.25, desc=f"Loading {new_file}...")
    orig_sd = safetensors.torch.load_file(new_file)
    progress(0.3, desc=f"Converting {new_file}...")
    conv_sd = convert_flux_transformer_checkpoint_to_diffusers(orig_sd)
    del orig_sd
    gc.collect()
    progress(0.35, desc=f"Saving {new_file}...")
    save_module_sd(conv_sd, "transformer", new_dir, dtype)
    del conv_sd
    gc.collect()
    progress(0.5, desc=f"Loading text_encoder_2 from {FLUX_T5_URL}...")
    t5_file = get_download_file(temp_dir, FLUX_T5_URL, civitai_key)
    if not t5_file: raise Exception(f"Safetensors file not found: {FLUX_T5_URL}")
    t5_sd = safetensors.torch.load_file(t5_file)
    safe_clean(t5_file)
    save_module_sd(t5_sd, "text_encoder_2", new_dir, dtype)
    del t5_sd
    gc.collect()
    progress(0.6, desc=f"Loading other components from {base_repo}...")
    pipe = FluxPipeline.from_pretrained(base_repo, transformer=None, text_encoder_2=None, use_safetensors=True, **kwargs,
                                        torch_dtype=torch.bfloat16, token=hf_token)
    pipe.save_pretrained(new_dir)
    progress(0.75, desc=f"Loading nontensor files from {base_repo}...")
    snapshot_download(repo_id=base_repo, local_dir=down_dir, token=hf_token, force_download=True,
                      ignore_patterns=["*.safetensors", "*.sft", ".*", "README*", "*.md", "*.index", "*.jpg", "*.jpeg", "*.png", "*.webp"])
    shutil.copytree(down_dir, new_dir, ignore=shutil.ignore_patterns(".*", "README*", "*.md", "*.jpg", "*.jpeg", "*.png", "*.webp"), dirs_exist_ok=True)
    safe_clean(down_dir)


def convert_sd35_fp8_cpu(new_file: str, new_dir: str, dtype: str, base_repo: str, civitai_key: str, kwargs: dict, progress=gr.Progress(track_tqdm=True)):
    temp_dir = TEMP_DIR
    down_dir = str(Path(f"{TEMP_DIR}/down"))
    os.makedirs(down_dir, exist_ok=True)
    hf_token = get_token()
    progress(0.25, desc=f"Loading {new_file}...")
    orig_sd = safetensors.torch.load_file(new_file)
    progress(0.3, desc=f"Converting {new_file}...")
    conv_sd = convert_sd3_transformer_checkpoint_to_diffusers(orig_sd)
    del orig_sd
    gc.collect()
    progress(0.35, desc=f"Saving {new_file}...")
    save_module_sd(conv_sd, "transformer", new_dir, dtype)
    del conv_sd
    gc.collect()
    progress(0.5, desc=f"Loading text_encoder_3 from {SD35_T5_URL}...")
    t5_file = get_download_file(temp_dir, SD35_T5_URL, civitai_key)
    if not t5_file: raise Exception(f"Safetensors file not found: {SD35_T5_URL}")
    t5_sd = safetensors.torch.load_file(t5_file)
    safe_clean(t5_file)
    conv_t5_sd = convert_sd3_t5_checkpoint_to_diffusers(t5_sd)
    del t5_sd
    gc.collect()
    save_module_sd(conv_t5_sd, "text_encoder_3", new_dir, dtype)
    del conv_t5_sd
    gc.collect()
    progress(0.6, desc=f"Loading other components from {base_repo}...")
    pipe = StableDiffusion3Pipeline.from_pretrained(base_repo, transformer=None, text_encoder_3=None, use_safetensors=True, **kwargs,
                                                    torch_dtype=torch.bfloat16, token=hf_token)
    pipe.save_pretrained(new_dir)
    progress(0.75, desc=f"Loading nontensor files from {base_repo}...")
    snapshot_download(repo_id=base_repo, local_dir=down_dir, token=hf_token, force_download=True,
                      ignore_patterns=["*.safetensors", "*.sft", ".*", "README*", "*.md", "*.index", "*.jpg", "*.jpeg", "*.png", "*.webp"])
    shutil.copytree(down_dir, new_dir, ignore=shutil.ignore_patterns(".*", "README*", "*.md", "*.jpg", "*.jpeg", "*.png", "*.webp"), dirs_exist_ok=True)
    safe_clean(down_dir)


#@spaces.GPU(duration=60)
def load_and_save_pipeline(pipe, model_type: str, url: str, new_file: str, new_dir: str, dtype: str,

                           scheduler: str, ema: bool, image_size: str, is_safety_checker: bool, base_repo: str, civitai_key: str, lora_dict: dict,

                           my_vae, my_clip_tokenizer, my_clip_encoder, my_t5_tokenizer, my_t5_encoder,

                           kwargs: dict, dkwargs: dict, progress=gr.Progress(track_tqdm=True)):
    try:
        hf_token = get_token()
        temp_dir = TEMP_DIR
        qkwargs = {}
        tfqkwargs = {}
        if is_nf4:
            nf4_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                                            bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16)
            nf4_config_tf = TFBitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4",
                                                 bnb_4bit_use_double_quant=True, bnb_4bit_compute_dtype=torch.bfloat16)
        else:
            nf4_config = None
            nf4_config_tf = None
        if dtype == "NF4" and nf4_config is not None and nf4_config_tf is not None:
            qkwargs["quantization_config"] = nf4_config
            tfqkwargs["quantization_config"] = nf4_config_tf

        #print(f"model_type:{model_type}, dtype:{dtype}, scheduler:{scheduler}, ema:{ema}, base_repo:{base_repo}")
        #print("lora_dict:", lora_dict, "kwargs:", kwargs, "dkwargs:", dkwargs)
        
        #t5 = None
        
        if model_type == "SDXL":
            if is_repo_name(url): pipe = StableDiffusionXLPipeline.from_pretrained(url, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
            else: pipe = StableDiffusionXLPipeline.from_single_file(new_file, use_safetensors=True, **kwargs, **dkwargs)
            pipe = fuse_loras(pipe, lora_dict, temp_dir, civitai_key, dkwargs)
            sconf = get_scheduler_config(scheduler)
            pipe.scheduler = sconf[0].from_config(pipe.scheduler.config, **sconf[1])
            pipe.save_pretrained(new_dir)
        elif model_type == "SD 1.5":
            if is_safety_checker:
                safety_checker = StableDiffusionSafetyChecker.from_pretrained("CompVis/stable-diffusion-safety-checker")
                feature_extractor = CLIPFeatureExtractor.from_pretrained("openai/clip-vit-base-patch32")
                kwargs["requires_safety_checker"] = True
                kwargs["safety_checker"] = safety_checker
                kwargs["feature_extractor"] = feature_extractor
            else: kwargs["requires_safety_checker"] = False
            if is_repo_name(url): pipe = StableDiffusionPipeline.from_pretrained(url, extract_ema=ema, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
            else: pipe = StableDiffusionPipeline.from_single_file(new_file, extract_ema=ema, use_safetensors=True, **kwargs, **dkwargs)
            pipe = fuse_loras(pipe, lora_dict, temp_dir, civitai_key, dkwargs)
            sconf = get_scheduler_config(scheduler)
            pipe.scheduler = sconf[0].from_config(pipe.scheduler.config, **sconf[1])
            if image_size != "512": pipe.vae = AutoencoderKL.from_config(pipe.vae.config, sample_size=int(image_size))
            pipe.save_pretrained(new_dir)
        elif model_type == "FLUX":
            if dtype != "fp8":
                if is_repo_name(url):
                    transformer = FluxTransformer2DModel.from_pretrained(url, subfolder="transformer", config=base_repo, **dkwargs, **qkwargs)
                    #if my_t5_encoder is None:
                    #    t5 = T5EncoderModel.from_pretrained(url, subfolder="text_encoder_2", config=base_repo, **dkwargs, **tfqkwargs)
                    #    kwargs["text_encoder_2"] = t5
                    pipe = FluxPipeline.from_pretrained(url, transformer=transformer, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
                else:
                    transformer = FluxTransformer2DModel.from_single_file(new_file, subfolder="transformer", config=base_repo, **dkwargs, **qkwargs)
                    #if my_t5_encoder is None:
                    #    t5 = T5EncoderModel.from_pretrained(base_repo, subfolder="text_encoder_2", config=base_repo, **dkwargs, **tfqkwargs)
                    #    kwargs["text_encoder_2"] = t5
                    pipe = FluxPipeline.from_pretrained(base_repo, transformer=transformer, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
                pipe = fuse_loras(pipe, lora_dict, temp_dir, civitai_key, dkwargs)
                pipe.save_pretrained(new_dir)
            elif not is_repo_name(url): convert_flux_fp8_cpu(new_file, new_dir, dtype, base_repo, civitai_key, kwargs)
        elif model_type == "SD 3.5":
            if dtype != "fp8":
                if is_repo_name(url):
                    transformer = SD3Transformer2DModel.from_pretrained(url, subfolder="transformer", config=base_repo, **dkwargs, **qkwargs)
                    #if my_t5_encoder is None:
                    #    t5 = T5EncoderModel.from_pretrained(url, subfolder="text_encoder_3", config=base_repo, **dkwargs, **tfqkwargs)
                    #    kwargs["text_encoder_3"] = t5
                    pipe = StableDiffusion3Pipeline.from_pretrained(url, transformer=transformer, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
                else:
                    transformer = SD3Transformer2DModel.from_single_file(new_file, subfolder="transformer", config=base_repo, **dkwargs, **qkwargs)
                    #if my_t5_encoder is None:
                    #    t5 = T5EncoderModel.from_pretrained(base_repo, subfolder="text_encoder_3", config=base_repo, **dkwargs, **tfqkwargs)
                    #    kwargs["text_encoder_3"] = t5
                    pipe = StableDiffusion3Pipeline.from_pretrained(base_repo, transformer=transformer, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
                pipe = fuse_loras(pipe, lora_dict, temp_dir, civitai_key, dkwargs)
                pipe.save_pretrained(new_dir)
            elif not is_repo_name(url): convert_sd35_fp8_cpu(new_file, new_dir, dtype, base_repo, civitai_key, kwargs)
        else: # unknown model type
            if is_repo_name(url): pipe = DiffusionPipeline.from_pretrained(url, use_safetensors=True, **kwargs, **dkwargs, token=hf_token)
            else: pipe = DiffusionPipeline.from_single_file(new_file, use_safetensors=True, **kwargs, **dkwargs)
            pipe = fuse_loras(pipe, lora_dict, temp_dir, civitai_key, dkwargs)
            pipe.save_pretrained(new_dir)
    except Exception as e:
        print(f"Failed to load pipeline. {e}")
        raise Exception("Failed to load pipeline.") from e
    finally:
        return pipe


def convert_url_to_diffusers(url: str, civitai_key: str="", is_upload_sf: bool=False, dtype: str="fp16", vae: str="", clip: str="", t5: str="",

                             scheduler: str="Euler a", ema: bool=True, image_size: str="768", safety_checker: bool=False,

                             base_repo: str="", mtype: str="", lora_dict: dict={}, is_local: bool=True, progress=gr.Progress(track_tqdm=True)):
    try:
        hf_token = get_token()
        progress(0, desc="Start converting...")
        temp_dir = TEMP_DIR

        if is_repo_name(url) and is_repo_exists(url):
            new_file = url
            model_type = mtype
        else:
            new_file = get_download_file(temp_dir, url, civitai_key)
            if not new_file: raise Exception(f"Safetensors file not found: {url}")
            model_type = get_model_type_from_key(new_file)
        new_dir = Path(new_file).stem.replace(" ", "_").replace(",", "_").replace(".", "_") #
        
        kwargs = {}
        dkwargs = {}
        if dtype != DTYPE_DEFAULT: dkwargs["torch_dtype"] = get_process_dtype(dtype, model_type)
        pipe = None

        print(f"Model type: {model_type} / VAE: {vae} / CLIP: {clip} / T5: {t5} / Scheduler: {scheduler} / dtype: {dtype} / EMA: {ema} / Base repo: {base_repo} / LoRAs: {lora_dict}")

        my_vae = None
        if vae:
            progress(0, desc=f"Loading VAE: {vae}...")
            if is_repo_name(vae): my_vae = AutoencoderKL.from_pretrained(vae, **dkwargs, token=hf_token)
            else:
                new_vae_file = get_download_file(temp_dir, vae, civitai_key)
                my_vae = AutoencoderKL.from_single_file(new_vae_file, **dkwargs) if new_vae_file else None
                safe_clean(new_vae_file)
            if my_vae: kwargs["vae"] = my_vae

        my_clip_tokenizer = None
        my_clip_encoder = None
        if clip:
            progress(0, desc=f"Loading CLIP: {clip}...")
            if is_repo_name(clip):
                my_clip_tokenizer = CLIPTokenizer.from_pretrained(clip, token=hf_token)
                if model_type == "SD 3.5": my_clip_encoder = CLIPTextModelWithProjection.from_pretrained(clip, **dkwargs, token=hf_token)
                else: my_clip_encoder = CLIPTextModel.from_pretrained(clip, **dkwargs, token=hf_token)
            else:
                new_clip_file = get_download_file(temp_dir, clip, civitai_key)
                if model_type == "SD 3.5": my_clip_encoder = CLIPTextModelWithProjection.from_single_file(new_clip_file, **dkwargs) if new_clip_file else None
                else: my_clip_encoder = CLIPTextModel.from_single_file(new_clip_file, **dkwargs) if new_clip_file else None
                safe_clean(new_clip_file)
            if model_type == "SD 3.5":
                if my_clip_tokenizer:
                    kwargs["tokenizer"] = my_clip_tokenizer
                    kwargs["tokenizer_2"] = my_clip_tokenizer
                if my_clip_encoder:
                    kwargs["text_encoder"] = my_clip_encoder
                    kwargs["text_encoder_2"] = my_clip_encoder
            else:
                if my_clip_tokenizer: kwargs["tokenizer"] = my_clip_tokenizer
                if my_clip_encoder: kwargs["text_encoder"] = my_clip_encoder

        my_t5_tokenizer = None
        my_t5_encoder = None
        if t5:
            progress(0, desc=f"Loading T5: {t5}...")
            if is_repo_name(t5):
                my_t5_tokenizer = AutoTokenizer.from_pretrained(t5, token=hf_token)
                my_t5_encoder = T5EncoderModel.from_pretrained(t5, **dkwargs, token=hf_token)
            else:
                new_t5_file = get_download_file(temp_dir, t5, civitai_key)
                my_t5_encoder = T5EncoderModel.from_single_file(new_t5_file, **dkwargs) if new_t5_file else None
                safe_clean(new_t5_file)
            if model_type == "SD 3.5":
                if my_t5_tokenizer: kwargs["tokenizer_3"] = my_t5_tokenizer
                if my_t5_encoder: kwargs["text_encoder_3"] = my_t5_encoder
            else:
                if my_t5_tokenizer: kwargs["tokenizer_2"] = my_t5_tokenizer
                if my_t5_encoder: kwargs["text_encoder_2"] = my_t5_encoder

        pipe = load_and_save_pipeline(pipe, model_type, url, new_file, new_dir, dtype, scheduler, ema, image_size, safety_checker, base_repo, civitai_key, lora_dict,
                                      my_vae, my_clip_tokenizer, my_clip_encoder, my_t5_tokenizer, my_t5_encoder, kwargs, dkwargs)
        
        if Path(new_dir).exists(): save_readme_md(new_dir, url)

        if not is_local:
            if not is_repo_name(new_file) and is_upload_sf: shutil.move(str(Path(new_file).resolve()), str(Path(new_dir, Path(new_file).name).resolve()))
            else: safe_clean(new_file)

        progress(1, desc="Converted.")
        return new_dir
    except Exception as e:
        print(f"Failed to convert. {e}")
        raise Exception("Failed to convert.") from e
    finally:
        del pipe
        torch.cuda.empty_cache()
        gc.collect()


def convert_url_to_diffusers_repo(dl_url: str, hf_user: str, hf_repo: str, hf_token: str, civitai_key="", is_private: bool=True,

                                  gated: str="False", is_overwrite: bool=False, is_pr: bool=False,

                                  is_upload_sf: bool=False, urls: list=[], dtype: str="fp16", vae: str="", clip: str="", t5: str="", scheduler: str="Euler a",

                                  ema: bool=True, image_size: str="768", safety_checker: bool=False,

                                  base_repo: str="", mtype: str="", lora1: str="", lora1s=1.0, lora2: str="", lora2s=1.0, lora3: str="", lora3s=1.0,

                                  lora4: str="", lora4s=1.0, lora5: str="", lora5s=1.0, args: str="", progress=gr.Progress(track_tqdm=True)):
    try:
        is_local = False
        if not civitai_key and os.environ.get("CIVITAI_API_KEY"): civitai_key = os.environ.get("CIVITAI_API_KEY") # default Civitai API key
        if not hf_token and os.environ.get("HF_TOKEN"): hf_token = os.environ.get("HF_TOKEN") # default HF write token
        if not hf_user: raise gr.Error(f"Invalid user name: {hf_user}")
        if gated != "False" and is_private: raise gr.Error(f"Gated repo must be public")
        set_token(hf_token)
        lora_dict = {lora1: lora1s, lora2: lora2s, lora3: lora3s, lora4: lora4s, lora5: lora5s}
        new_path = convert_url_to_diffusers(dl_url, civitai_key, is_upload_sf, dtype, vae, clip, t5, scheduler, ema, image_size, safety_checker, base_repo, mtype, lora_dict, is_local)
        if not new_path: return ""
        new_repo_id = f"{hf_user}/{Path(new_path).stem}"
        if hf_repo != "": new_repo_id = f"{hf_user}/{hf_repo}"
        if not is_repo_name(new_repo_id): raise gr.Error(f"Invalid repo name: {new_repo_id}")
        if not is_overwrite and is_repo_exists(new_repo_id) and not is_pr: raise gr.Error(f"Repo already exists: {new_repo_id}")
        repo_url = upload_repo(new_repo_id, new_path, is_private, is_pr)
        gate_repo(new_repo_id, gated)
        safe_clean(new_path)
        if not urls: urls = []
        urls.append(repo_url)
        md = "### Your new repo:\n"
        for u in urls:
            md += f"[{str(u).split('/')[-2]}/{str(u).split('/')[-1]}]({str(u)})<br>"
        return gr.update(value=urls, choices=urls), gr.update(value=md)
    except Exception as e:
        print(f"Error occured. {e}")
        raise gr.Error(f"Error occured. {e}")


if __name__ == "__main__":
    parser = argparse.ArgumentParser()

    parser.add_argument("--url", type=str, required=True, help="URL of the model to convert.")
    parser.add_argument("--dtype", default="fp16", type=str, choices=get_dtypes(), help='Output data type. (Default: "fp16")')
    parser.add_argument("--scheduler", default="Euler a", type=str, choices=list(SCHEDULER_CONFIG_MAP.keys()), required=False, help="Scheduler name to use.")
    parser.add_argument("--vae", default="", type=str, required=False, help="URL or Repo ID of the VAE to use.")
    parser.add_argument("--clip", default="", type=str, required=False, help="URL or Repo ID of the CLIP to use.")
    parser.add_argument("--t5", default="", type=str, required=False, help="URL or Repo ID of the T5 to use.")
    parser.add_argument("--base", default="", type=str, required=False, help="Repo ID of the base repo.")
    parser.add_argument("--nonema", action="store_true", default=False, help="Don't extract EMA (for SD 1.5).")
    parser.add_argument("--civitai_key", default="", type=str, required=False, help="Civitai API Key (If you want to download file from Civitai).")
    parser.add_argument("--lora1", default="", type=str, required=False, help="URL of the LoRA to use.")
    parser.add_argument("--lora1s", default=1.0, type=float, required=False, help="LoRA weight scale of --lora1.")
    parser.add_argument("--lora2", default="", type=str, required=False, help="URL of the LoRA to use.")
    parser.add_argument("--lora2s", default=1.0, type=float, required=False, help="LoRA weight scale of --lora2.")
    parser.add_argument("--lora3", default="", type=str, required=False, help="URL of the LoRA to use.")
    parser.add_argument("--lora3s", default=1.0, type=float, required=False, help="LoRA weight scale of --lora3.")
    parser.add_argument("--lora4", default="", type=str, required=False, help="URL of the LoRA to use.")
    parser.add_argument("--lora4s", default=1.0, type=float, required=False, help="LoRA weight scale of --lora4.")
    parser.add_argument("--lora5", default="", type=str, required=False, help="URL of the LoRA to use.")
    parser.add_argument("--lora5s", default=1.0, type=float, required=False, help="LoRA weight scale of --lora5.")
    parser.add_argument("--loras", default="", type=str, required=False, help="Folder of the LoRA to use.")

    args = parser.parse_args()
    assert args.url is not None, "Must provide a URL!"

    is_local = True
    lora_dict = {args.lora1: args.lora1s, args.lora2: args.lora2s, args.lora3: args.lora3s, args.lora4: args.lora4s, args.lora5: args.lora5s}
    if args.loras and Path(args.loras).exists():
        for p in Path(args.loras).glob('**/*.safetensors'):
            lora_dict[str(p)] = 1.0
    ema = not args.nonema
    mtype = "SDXL"

    convert_url_to_diffusers(args.url, args.civitai_key, args.dtype, args.vae, args.clip, args.t5, args.scheduler, ema, args.base, mtype, lora_dict, is_local)