File size: 6,439 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
import os
import torch
from shared.utils.hf import build_hf_url


class family_handler:
    @staticmethod
    def query_supported_types():
        return ["longcat_video", "longcat_avatar"]

    @staticmethod
    def query_family_maps():
        return {}, {}

    @staticmethod
    def query_model_family():
        return "longcat"

    @staticmethod
    def query_family_infos():
        return {"longcat": (60, "LongCat")}

    @staticmethod
    def register_lora_cli_args(parser):
        parser.add_argument(
            "--lora-dir-longcat",
            type=str,
            default=os.path.join("loras", "longcat"),
            help="Path to a directory that contains LongCat Video LoRAs",
        )
        parser.add_argument(
            "--lora-dir-longcat-avatar",
            type=str,
            default=os.path.join("loras", "longcat_avatar"),
            help="Path to a directory that contains LongCat Avatar LoRAs",
        )

    @staticmethod
    def get_lora_dir(base_model_type, args):
        if base_model_type == "longcat_avatar":
            return args.lora_dir_longcat_avatar
        return args.lora_dir_longcat

    @staticmethod
    def query_model_def(base_model_type, model_def):
        extra_model_def = {
            "frames_minimum": 5,
            "frames_steps": 4,
            "sliding_window": True,
            "guidance_max_phases": 1,
            "image_prompt_types_allowed": "TSVL",
            "video_continuation": True,
            "sample_solvers": [
                ("Auto (Continuation = Enhanced HF)", "auto"),
                ("Default", ""),
                ("Enhanced HF", "enhance_hf"),
                ("Distill", "distill"),
            ],
        }
        text_encoder_folder = "umt5-xxl"
        extra_model_def["text_encoder_URLs"] = [
            build_hf_url("DeepBeepMeep/Wan2.1", text_encoder_folder, "models_t5_umt5-xxl-enc-bf16.safetensors"),
            build_hf_url("DeepBeepMeep/Wan2.1", text_encoder_folder, "models_t5_umt5-xxl-enc-quanto_int8.safetensors"),
        ]
        extra_model_def["text_encoder_folder"] = text_encoder_folder

        if base_model_type == "longcat_video":
            extra_model_def.update(
                {
                    "fps": 15,
                    "profiles_dir": ["longcat_video"],
                }
            )
        elif base_model_type == "longcat_avatar":
            extra_model_def.update(
                {
                    "fps": 16,
                    "profiles_dir": [base_model_type],
                    "audio_guide_label": "Voice to follow",
                    "audio_guide2_label": "Voice to follow #2",
                    "audio_guidance": True,
                    "any_audio_prompt": True,
                    "audio_prompt_choices": True,                
                    "image_ref_choices": {
                        "choices": [("None", ""), ("Anchor Reference Image", "KI")],
                        "letters_filter": "KI",
                        "visible": True,
                        "label": "Anchor Reference Image",
                    },
                    "reference_image_enabled": True,
                    "no_background_removal": True,
                    "image_prompt_types_allowed": "TSVL",
                }
            )


        return extra_model_def

    @staticmethod
    def get_rgb_factors(base_model_type):
        from shared.RGB_factors import get_rgb_factors

        return get_rgb_factors("wan")

    @staticmethod
    def query_model_files(computeList, base_model_type, model_def=None):
        download_def = [
            {
                "repoId": "DeepBeepMeep/Wan2.1",
                "sourceFolderList": ["umt5-xxl", "chinese-wav2vec2-base"],
                "fileList": [
                    ["special_tokens_map.json", "spiece.model", "tokenizer.json", "tokenizer_config.json"],
                    [
                        "config.json",
                        "preprocessor_config.json",
                        "pytorch_model.bin",
                        "readme.txt",
                    ],
                ],
            }
        ]
        download_def += [
            {
                "repoId": "DeepBeepMeep/Wan2.1",
                "sourceFolderList": [""],
                "fileList": [["Wan2.1_VAE_bf16.safetensors"]],
            }
        ]
        return download_def

    @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,
        **kwargs,
    ):
        from .longcat_main import LongCatModel

        longcat_model = LongCatModel(
            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": longcat_model.transformer,
            "vae": longcat_model.vae,
            "text_encoder": longcat_model.text_encoder.model,
        }
        if longcat_model.audio_encoder is not None:
            pipe["wav2vec"] = longcat_model.audio_encoder

        return longcat_model, pipe

    @staticmethod
    def update_default_settings(base_model_type, model_def, ui_defaults):
        ui_defaults.update(
            {
                "guidance_scale": 4.0,
                "num_inference_steps": 50,
                "audio_guidance_scale": 4.0,
                "sliding_window_overlap": 13,
                "sliding_window_size": 93,                 
            }
        )
        if base_model_type == "longcat_video":
            ui_defaults.update({"video_length": 93})

        if base_model_type in ["longcat_avatar"]:
            ui_defaults.update({"video_length": 93, "video_prompt_type": ""})

        if ui_defaults.get("sample_solver", "") == "":
            ui_defaults["sample_solver"] = "auto"