Spaces:
ginipick
/
Running on Zero

File size: 12,392 Bytes
4450790
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from nodes import MAX_RESOLUTION, ConditioningZeroOut, ConditioningSetTimestepRange, ConditioningCombine
import re

class CLIPTextEncodeSDXLSimplified:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "width": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
            "height": ("INT", {"default": 1024.0, "min": 0, "max": MAX_RESOLUTION}),
            "size_cond_factor": ("INT", {"default": 4, "min": 1, "max": 16 }),
            "text": ("STRING", {"multiline": True, "dynamicPrompts": True, "default": ""}),
            "clip": ("CLIP", ),
            }}
    RETURN_TYPES = ("CONDITIONING",)
    FUNCTION = "execute"
    CATEGORY = "essentials/conditioning"

    def execute(self, clip, width, height, size_cond_factor, text):
        crop_w = 0
        crop_h = 0
        width = width*size_cond_factor
        height = height*size_cond_factor
        target_width = width
        target_height = height
        text_g = text_l = text

        tokens = clip.tokenize(text_g)
        tokens["l"] = clip.tokenize(text_l)["l"]
        if len(tokens["l"]) != len(tokens["g"]):
            empty = clip.tokenize("")
            while len(tokens["l"]) < len(tokens["g"]):
                tokens["l"] += empty["l"]
            while len(tokens["l"]) > len(tokens["g"]):
                tokens["g"] += empty["g"]
        cond, pooled = clip.encode_from_tokens(tokens, return_pooled=True)
        return ([[cond, {"pooled_output": pooled, "width": width, "height": height, "crop_w": crop_w, "crop_h": crop_h, "target_width": target_width, "target_height": target_height}]], )

class ConditioningCombineMultiple:
    @classmethod
    def INPUT_TYPES(s):
        return {
            "required": {
                "conditioning_1": ("CONDITIONING",),
                "conditioning_2": ("CONDITIONING",),
            }, "optional": {
                "conditioning_3": ("CONDITIONING",),
                "conditioning_4": ("CONDITIONING",),
                "conditioning_5": ("CONDITIONING",),
            },
        }
    RETURN_TYPES = ("CONDITIONING",)
    FUNCTION = "execute"
    CATEGORY = "essentials/conditioning"

    def execute(self, conditioning_1, conditioning_2, conditioning_3=None, conditioning_4=None, conditioning_5=None):
        c = conditioning_1 + conditioning_2

        if conditioning_3 is not None:
            c += conditioning_3
        if conditioning_4 is not None:
            c += conditioning_4
        if conditioning_5 is not None:
            c += conditioning_5

        return (c,)

class SD3NegativeConditioning:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "conditioning": ("CONDITIONING",),
            "end": ("FLOAT", {"default": 0.1, "min": 0.0, "max": 1.0, "step": 0.001 }),
        }}
    RETURN_TYPES = ("CONDITIONING",)
    FUNCTION = "execute"
    CATEGORY = "essentials/conditioning"

    def execute(self, conditioning, end):
        zero_c = ConditioningZeroOut().zero_out(conditioning)[0]

        if end == 0:
            return (zero_c, )

        c = ConditioningSetTimestepRange().set_range(conditioning, 0, end)[0]
        zero_c = ConditioningSetTimestepRange().set_range(zero_c, end, 1.0)[0]
        c = ConditioningCombine().combine(zero_c, c)[0]

        return (c, )

class FluxAttentionSeeker:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "clip": ("CLIP",),
            "apply_to_query": ("BOOLEAN", { "default": True }),
            "apply_to_key": ("BOOLEAN", { "default": True }),
            "apply_to_value": ("BOOLEAN", { "default": True }),
            "apply_to_out": ("BOOLEAN", { "default": True }),
            **{f"clip_l_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(12)},
            **{f"t5xxl_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(24)},
        }}

    RETURN_TYPES = ("CLIP",)
    FUNCTION = "execute"

    CATEGORY = "essentials/conditioning"

    def execute(self, clip, apply_to_query, apply_to_key, apply_to_value, apply_to_out, **values):
        if not apply_to_key and not apply_to_query and not apply_to_value and not apply_to_out:
            return (clip, )

        m = clip.clone()
        sd = m.patcher.model_state_dict()
        
        for k in sd:
            if "self_attn" in k:
                layer = re.search(r"\.layers\.(\d+)\.", k)
                layer = int(layer.group(1)) if layer else None

                if layer is not None and values[f"clip_l_{layer}"] != 1.0:
                    if (apply_to_query and "q_proj" in k) or (apply_to_key and "k_proj" in k) or (apply_to_value and "v_proj" in k) or (apply_to_out and "out_proj" in k):
                        m.add_patches({k: (None,)}, 0.0, values[f"clip_l_{layer}"])
            elif "SelfAttention" in k:
                block = re.search(r"\.block\.(\d+)\.", k)
                block = int(block.group(1)) if block else None

                if block is not None and values[f"t5xxl_{block}"] != 1.0:
                    if (apply_to_query and ".q." in k) or (apply_to_key and ".k." in k) or (apply_to_value and ".v." in k) or (apply_to_out and ".o." in k):
                        m.add_patches({k: (None,)}, 0.0, values[f"t5xxl_{block}"])

        return (m, )

class SD3AttentionSeekerLG:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "clip": ("CLIP",),
            "apply_to_query": ("BOOLEAN", { "default": True }),
            "apply_to_key": ("BOOLEAN", { "default": True }),
            "apply_to_value": ("BOOLEAN", { "default": True }),
            "apply_to_out": ("BOOLEAN", { "default": True }),
            **{f"clip_l_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(12)},
            **{f"clip_g_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(32)},
        }}

    RETURN_TYPES = ("CLIP",)
    FUNCTION = "execute"

    CATEGORY = "essentials/conditioning"

    def execute(self, clip, apply_to_query, apply_to_key, apply_to_value, apply_to_out, **values):
        if not apply_to_key and not apply_to_query and not apply_to_value and not apply_to_out:
            return (clip, )

        m = clip.clone()
        sd = m.patcher.model_state_dict()
        
        for k in sd:
            if "self_attn" in k:
                layer = re.search(r"\.layers\.(\d+)\.", k)
                layer = int(layer.group(1)) if layer else None

                if layer is not None:
                    if "clip_l" in k and values[f"clip_l_{layer}"] != 1.0:
                        if (apply_to_query and "q_proj" in k) or (apply_to_key and "k_proj" in k) or (apply_to_value and "v_proj" in k) or (apply_to_out and "out_proj" in k):
                            m.add_patches({k: (None,)}, 0.0, values[f"clip_l_{layer}"])
                    elif "clip_g" in k and values[f"clip_g_{layer}"] != 1.0:
                        if (apply_to_query and "q_proj" in k) or (apply_to_key and "k_proj" in k) or (apply_to_value and "v_proj" in k) or (apply_to_out and "out_proj" in k):
                            m.add_patches({k: (None,)}, 0.0, values[f"clip_g_{layer}"])

        return (m, )

class SD3AttentionSeekerT5:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "clip": ("CLIP",),
            "apply_to_query": ("BOOLEAN", { "default": True }),
            "apply_to_key": ("BOOLEAN", { "default": True }),
            "apply_to_value": ("BOOLEAN", { "default": True }),
            "apply_to_out": ("BOOLEAN", { "default": True }),
            **{f"t5xxl_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(24)},
        }}

    RETURN_TYPES = ("CLIP",)
    FUNCTION = "execute"

    CATEGORY = "essentials/conditioning"

    def execute(self, clip, apply_to_query, apply_to_key, apply_to_value, apply_to_out, **values):
        if not apply_to_key and not apply_to_query and not apply_to_value and not apply_to_out:
            return (clip, )

        m = clip.clone()
        sd = m.patcher.model_state_dict()
        
        for k in sd:
            if "SelfAttention" in k:
                block = re.search(r"\.block\.(\d+)\.", k)
                block = int(block.group(1)) if block else None

                if block is not None and values[f"t5xxl_{block}"] != 1.0:
                    if (apply_to_query and ".q." in k) or (apply_to_key and ".k." in k) or (apply_to_value and ".v." in k) or (apply_to_out and ".o." in k):
                        m.add_patches({k: (None,)}, 0.0, values[f"t5xxl_{block}"])

        return (m, )

class FluxBlocksBuster:
    @classmethod
    def INPUT_TYPES(s):
        return {"required": {
            "model": ("MODEL",),
            "blocks": ("STRING", {"default": "## 0 = 1.0\n## 1 = 1.0\n## 2 = 1.0\n## 3 = 1.0\n## 4 = 1.0\n## 5 = 1.0\n## 6 = 1.0\n## 7 = 1.0\n## 8 = 1.0\n## 9 = 1.0\n## 10 = 1.0\n## 11 = 1.0\n## 12 = 1.0\n## 13 = 1.0\n## 14 = 1.0\n## 15 = 1.0\n## 16 = 1.0\n## 17 = 1.0\n## 18 = 1.0\n# 0 = 1.0\n# 1 = 1.0\n# 2 = 1.0\n# 3 = 1.0\n# 4 = 1.0\n# 5 = 1.0\n# 6 = 1.0\n# 7 = 1.0\n# 8 = 1.0\n# 9 = 1.0\n# 10 = 1.0\n# 11 = 1.0\n# 12 = 1.0\n# 13 = 1.0\n# 14 = 1.0\n# 15 = 1.0\n# 16 = 1.0\n# 17 = 1.0\n# 18 = 1.0\n# 19 = 1.0\n# 20 = 1.0\n# 21 = 1.0\n# 22 = 1.0\n# 23 = 1.0\n# 24 = 1.0\n# 25 = 1.0\n# 26 = 1.0\n# 27 = 1.0\n# 28 = 1.0\n# 29 = 1.0\n# 30 = 1.0\n# 31 = 1.0\n# 32 = 1.0\n# 33 = 1.0\n# 34 = 1.0\n# 35 = 1.0\n# 36 = 1.0\n# 37 = 1.0", "multiline": True, "dynamicPrompts": True}),
            #**{f"double_block_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(19)},
            #**{f"single_block_{s}": ("FLOAT", { "display": "slider", "default": 1.0, "min": 0, "max": 5, "step": 0.05 }) for s in range(38)},
        }}
    RETURN_TYPES = ("MODEL", "STRING")
    RETURN_NAMES = ("MODEL", "patched_blocks")
    FUNCTION = "patch"

    CATEGORY = "essentials/conditioning"

    def patch(self, model, blocks):
        if blocks == "":
            return (model, )

        m = model.clone()
        sd = model.model_state_dict()
        patched_blocks = []

        """
        Also compatible with the following format:

        double_blocks\.0\.(img|txt)_(mod|attn|mlp)\.(lin|qkv|proj|0|2)\.(weight|bias)=1.1
        single_blocks\.0\.(linear[12]|modulation\.lin)\.(weight|bias)=1.1

        The regex is used to match the block names
        """

        blocks = blocks.split("\n")
        blocks = [b.strip() for b in blocks if b.strip()]

        for k in sd:
            for block in blocks:
                block = block.split("=")
                value = float(block[1].strip()) if len(block) > 1 else 1.0
                block = block[0].strip()
                if block.startswith("##"):
                    block = r"double_blocks\." + block[2:].strip() + r"\.(img|txt)_(mod|attn|mlp)\.(lin|qkv|proj|0|2)\.(weight|bias)"
                elif block.startswith("#"):
                    block = r"single_blocks\." + block[1:].strip() + r"\.(linear[12]|modulation\.lin)\.(weight|bias)"

                if value != 1.0 and re.search(block, k):
                    m.add_patches({k: (None,)}, 0.0, value)
                    patched_blocks.append(f"{k}: {value}")

        patched_blocks = "\n".join(patched_blocks)

        return (m, patched_blocks,)


COND_CLASS_MAPPINGS = {
    "CLIPTextEncodeSDXL+": CLIPTextEncodeSDXLSimplified,
    "ConditioningCombineMultiple+": ConditioningCombineMultiple,
    "SD3NegativeConditioning+": SD3NegativeConditioning,
    "FluxAttentionSeeker+": FluxAttentionSeeker,
    "SD3AttentionSeekerLG+": SD3AttentionSeekerLG,
    "SD3AttentionSeekerT5+": SD3AttentionSeekerT5,
    "FluxBlocksBuster+": FluxBlocksBuster,
}

COND_NAME_MAPPINGS = {
    "CLIPTextEncodeSDXL+": "πŸ”§ SDXL CLIPTextEncode",
    "ConditioningCombineMultiple+": "πŸ”§ Cond Combine Multiple",
    "SD3NegativeConditioning+": "πŸ”§ SD3 Negative Conditioning",
    "FluxAttentionSeeker+": "πŸ”§ Flux Attention Seeker",
    "SD3AttentionSeekerLG+": "πŸ”§ SD3 Attention Seeker L/G",
    "SD3AttentionSeekerT5+": "πŸ”§ SD3 Attention Seeker T5",
    "FluxBlocksBuster+": "πŸ”§ Flux Model Blocks Buster",
}