Fix ADV_CLIP_emb support

__init__.py

@@ -2,12 +2,15 @@
 from .random_gen import RandomPromptGenerator
 from .cascade_utils import StableCascade_AutoCompLatent
 from .clip_misc import CLIPTextEncodeBREAK, CLIPMicroConditioning, CLIPTokenCounter
-from .clip_negpip import CLIPNegPip, hijack_adv_encode
+from .clip_negpip import CLIPNegPip
 from .attention_couple_ppm import AttentionCouplePPM
 from .guidance_limiter import GuidanceLimiter
 from .samplers import CFGPPSamplerSelect, inject_samplers
 from .schedulers import hijack_schedulers
 
+from .compat.advanced_encode import hijack_adv_encode
+
+
 WEB_DIRECTORY = "./js"
 
 NODE_CLASS_MAPPINGS = {
@@ -43,6 +46,8 @@
     "CFGPPSamplerSelect": "CFG++SamplerSelect",
 }
 
+
 inject_samplers()
 hijack_schedulers()
+
 hijack_adv_encode()

clip_negpip.py

@@ -3,16 +3,13 @@
 # https://github.com/hako-mikan/sd-webui-negpip
 from functools import partial
 import torch
-from math import copysign
 
 from comfy import model_management
 from comfy.model_patcher import ModelPatcher
 from comfy.sd import CLIP
 from comfy.sd1_clip import SD1ClipModel, gen_empty_tokens, ClipTokenWeightEncoder
 from comfy.sdxl_clip import SDXLClipModel, SDXLRefinerClipModel
 
-INITIALIZED = False
-
 
 def has_negpip(model_options: dict):
     try:
@@ -76,65 +73,6 @@ def _encode_token_weights_negpip(_self: ClipTokenWeightEncoder, token_weight_pai
     return torch.cat(output, dim=-2).to(model_management.intermediate_device()), first_pooled
 
 
-def _advanced_encode_from_tokens_negpip_wrapper(advanced_encode_from_tokens, from_zero):
-
-    def advanced_encode_from_tokens_negpip(
-        tokenized,
-        token_normalization,
-        weight_interpretation,
-        encode_func,
-        m_token=266,
-        length=77,
-        w_max=1.0,
-        return_pooled=False,
-        apply_to_pooled=False,
-    ):
-        tokenized_abs = [[(t, abs(w), p) for t, w, p in x] for x in tokenized]
-        weights_sign = [[copysign(1, w) for _, w, _ in x] for x in tokenized]
-
-        def _encoded_with_negpip(encode_func, m_token=266, length=77):
-            tokens = [[(m_token, 1.0) for _ in range(length)]]
-            emb, _ = encode_func(tokens)
-            if emb.shape[1] == length:
-                return False
-            elif emb.shape[1] == length * 2:
-                return True
-            raise ValueError("Unknown tensor shape - perhaps you've applied NegPip node more than once")
-
-        encoded_with_negpip = _encoded_with_negpip(encode_func, m_token, length)
-
-        def _encode_func(tokens):
-            emb, pooled = encode_func(tokens)
-            if encoded_with_negpip:
-                return emb[:, 0::2, :], pooled
-            return emb, pooled
-
-        def _apply_negpip(weights_sign, emb):
-            emb_negpip = torch.empty_like(emb).repeat(1, 2, 1)
-            emb_negpip[:, 0::2, :] = emb
-            emb_negpip[:, 1::2, :] = from_zero(weights_sign, emb)
-            return emb_negpip
-
-        weighted_emb, pooled = advanced_encode_from_tokens(
-            tokenized_abs,
-            token_normalization,
-            weight_interpretation,
-            _encode_func,
-            m_token,
-            length,
-            w_max,
-            return_pooled,
-            apply_to_pooled,
-        )
-
-        if encoded_with_negpip:
-            weighted_emb = _apply_negpip(weights_sign, weighted_emb)
-
-        return weighted_emb, pooled
-
-    return advanced_encode_from_tokens_negpip
-
-
 class CLIPNegPip:
     @classmethod
     def INPUT_TYPES(s):
@@ -169,22 +107,3 @@ def patch(self, model: ModelPatcher, clip: CLIP):
                 m.set_model_attn2_patch(_negpip_attn)
 
         return (m, c)
-
-
-def hijack_adv_encode():
-    global INITIALIZED
-    if not INITIALIZED:
-        try:
-            import custom_nodes.ComfyUI_ADV_CLIP_emb.adv_encode as adv_encode
-
-            advanced_encode_from_tokens_negpip = _advanced_encode_from_tokens_negpip_wrapper(
-                adv_encode.advanced_encode_from_tokens, adv_encode.from_zero
-            )
-
-            adv_encode.advanced_encode_from_tokens = advanced_encode_from_tokens_negpip
-
-        except ImportError:
-            pass
-
-        finally:
-            INITIALIZED = True

compat/advanced_encode.py

@@ -0,0 +1,94 @@
+import torch
+from math import copysign
+
+INITIALIZED = False
+
+
+def _advanced_encode_from_tokens_negpip_wrapper(advanced_encode_from_tokens, from_zero):
+
+    def advanced_encode_from_tokens_negpip(
+        tokenized,
+        token_normalization,
+        weight_interpretation,
+        encode_func,
+        m_token=266,
+        length=77,
+        w_max=1.0,
+        return_pooled=False,
+        apply_to_pooled=False,
+    ):
+        tokenized_abs = [[(t, abs(w), p) for t, w, p in x] for x in tokenized]
+        weights_sign = [[copysign(1, w) for _, w, _ in x] for x in tokenized]
+
+        def _encoded_with_negpip(encode_func, m_token=266, length=77):
+            tokens = [[(m_token, 1.0) for _ in range(length)]]
+            emb, _ = encode_func(tokens)
+            if emb.shape[1] == length:
+                return False
+            elif emb.shape[1] == length * 2:
+                return True
+            raise ValueError("Unknown tensor shape - perhaps you've applied NegPip node more than once")
+
+        encoded_with_negpip = _encoded_with_negpip(encode_func, m_token, length)
+
+        def _encode_func(tokens):
+            emb, pooled = encode_func(tokens)
+            if encoded_with_negpip:
+                return emb[:, 0::2, :], pooled
+            return emb, pooled
+
+        def _apply_negpip(weights_sign, emb):
+            emb_negpip = torch.empty_like(emb).repeat(1, 2, 1)
+            emb_negpip[:, 0::2, :] = emb
+            emb_negpip[:, 1::2, :] = from_zero(weights_sign, emb)
+            return emb_negpip
+
+        weighted_emb, pooled = advanced_encode_from_tokens(
+            tokenized_abs,
+            token_normalization,
+            weight_interpretation,
+            _encode_func,
+            m_token,
+            length,
+            w_max,
+            return_pooled,
+            apply_to_pooled,
+        )
+
+        if encoded_with_negpip:
+            weighted_emb = _apply_negpip(weights_sign, weighted_emb)
+
+        return weighted_emb, pooled
+
+    return advanced_encode_from_tokens_negpip
+
+
+def hijack_adv_encode():
+    global INITIALIZED
+    if not INITIALIZED:
+        import sys
+        import pathlib
+
+        custom_nodes = pathlib.Path(__file__).parent.parent.parent
+        assert custom_nodes.name == "custom_nodes"
+
+        sys.path.insert(0, str(custom_nodes))
+
+        try:
+
+            import custom_nodes.ComfyUI_ADV_CLIP_emb.adv_encode as adv_encode
+            import ComfyUI_ADV_CLIP_emb.adv_encode as adv_encode_inner
+
+            advanced_encode_from_tokens_negpip = _advanced_encode_from_tokens_negpip_wrapper(
+                adv_encode.advanced_encode_from_tokens, adv_encode.from_zero
+            )
+
+            adv_encode.advanced_encode_from_tokens = advanced_encode_from_tokens_negpip
+            adv_encode_inner.advanced_encode_from_tokens = advanced_encode_from_tokens_negpip
+
+        except ImportError:
+            pass
+
+        finally:
+            sys.path.pop(0)
+            INITIALIZED = True

pyproject.toml

@@ -1,7 +1,7 @@
 [project]
 name = "comfyui-ppm"
 description = "Fixed AttentionCouple/NegPip(negative weights in prompts), more CFG++ samplers, etc."
-version = "1.0.4"
+version = "1.0.5"
 license = "AGPL-3.0"
 
 [project.urls]