fix missing infotext cased by conda cache

some generation params such as TI hashes or Emphasis is added in sd_hijack / sd_hijack_clip
if conda are fetche from cache sd_hijack_clip will not be executed and it won't have a chance to to add generation params

the generation params will also be missing if in non low-vram mode because the hijack.extra_generation_params was never read after calculate_hr_conds

Author: w-e-w

modules/processing.py

@@ -187,6 +187,7 @@ class StableDiffusionProcessing:
 
     cached_uc = [None, None]
     cached_c = [None, None]
+    hijack_generation_params_state_list = []
 
     comments: dict = None
     sampler: sd_samplers_common.Sampler | None = field(default=None, init=False)
@@ -480,16 +481,36 @@ def get_conds_with_caching(self, function, required_prompts, steps, caches, extr
 
         for cache in caches:
             if cache[0] is not None and cached_params == cache[0]:
+                if len(cache) == 3:
+                    generation_params_state, cached_params_2 = cache[2]
+                    if cached_params == cached_params_2:
+                        self.hijack_generation_params_state_list.extend(generation_params_state)
                 return cache[1]
 
         cache = caches[0]
 
         with devices.autocast():
             cache[1] = function(shared.sd_model, required_prompts, steps, hires_steps, shared.opts.use_old_scheduling)
 
+        generation_params_state = model_hijack.capture_generation_params_state()
+        self.hijack_generation_params_state_list.extend(generation_params_state)
+        if len(cache) == 2:
+            cache.append((generation_params_state, cached_params))
+        else:
+            cache[2] = (generation_params_state, cached_params)
+
         cache[0] = cached_params
         return cache[1]
 
+    def apply_hijack_generation_params(self):
+        self.extra_generation_params.update(model_hijack.extra_generation_params)
+        for func in self.hijack_generation_params_state_list:
+            try:
+                func(self.extra_generation_params)
+            except Exception:
+                errors.report(f"Failed to apply hijack generation params state", exc_info=True)
+        self.hijack_generation_params_state_list.clear()
+
     def setup_conds(self):
         prompts = prompt_parser.SdConditioning(self.prompts, width=self.width, height=self.height)
         negative_prompts = prompt_parser.SdConditioning(self.negative_prompts, width=self.width, height=self.height, is_negative_prompt=True)
@@ -502,6 +523,8 @@ def setup_conds(self):
         self.uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, total_steps, [self.cached_uc], self.extra_network_data)
         self.c = self.get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, prompts, total_steps, [self.cached_c], self.extra_network_data)
 
+        self.apply_hijack_generation_params()
+
     def get_conds(self):
         return self.c, self.uc
 
@@ -965,8 +988,6 @@ def process_images_inner(p: StableDiffusionProcessing) -> Processed:
 
             p.setup_conds()
 
-            p.extra_generation_params.update(model_hijack.extra_generation_params)
-
             # params.txt should be saved after scripts.process_batch, since the
             # infotext could be modified by that callback
             # Example: a wildcard processed by process_batch sets an extra model
@@ -1513,6 +1534,8 @@ def calculate_hr_conds(self):
         self.hr_uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, hr_negative_prompts, self.firstpass_steps, [self.cached_hr_uc, self.cached_uc], self.hr_extra_network_data, total_steps)
         self.hr_c = self.get_conds_with_caching(prompt_parser.get_multicond_learned_conditioning, hr_prompts, self.firstpass_steps, [self.cached_hr_c, self.cached_c], self.hr_extra_network_data, total_steps)
 
+        self.apply_hijack_generation_params()
+
     def setup_conds(self):
         if self.is_hr_pass:
             # if we are in hr pass right now, the call is being made from the refiner, and we don't need to setup firstpass cons or switch model

modules/sd_hijack.py

@@ -6,6 +6,7 @@
 from modules.hypernetworks import hypernetwork
 from modules.shared import cmd_opts
 from modules import sd_hijack_clip, sd_hijack_open_clip, sd_hijack_unet, sd_hijack_xlmr, xlmr, xlmr_m18
+from modules.util import GenerationParamsState
 
 import ldm.modules.attention
 import ldm.modules.diffusionmodules.model
@@ -321,6 +322,13 @@ def clear_comments(self):
         self.comments = []
         self.extra_generation_params = {}
 
+    def capture_generation_params_state(self):
+        state = []
+        for key in list(self.extra_generation_params):
+            if isinstance(self.extra_generation_params[key], GenerationParamsState):
+                state.append(self.extra_generation_params.pop(key))
+        return state
+
     def get_prompt_lengths(self, text):
         if self.clip is None:
             return "-", "-"

modules/sd_hijack_clip.py

@@ -5,6 +5,7 @@
 
 from modules import prompt_parser, devices, sd_hijack, sd_emphasis
 from modules.shared import opts
+from modules.util import GenerationParamsState
 
 
 class PromptChunk:
@@ -27,6 +28,31 @@ def __init__(self):
 are applied by sd_hijack.EmbeddingsWithFixes's forward function."""
 
 
+class EmbeddingHashes(GenerationParamsState):
+    def __init__(self, hashes: list):
+        super().__init__()
+        self.hashes = hashes
+
+    def __call__(self, extra_generation_params):
+        unique_hashes = dict.fromkeys(self.hashes)
+        if existing_ti_hashes := extra_generation_params.get('TI hashes'):
+            unique_hashes.update(dict.fromkeys(existing_ti_hashes.split(', ')))
+        extra_generation_params['TI hashes'] = ', '.join(unique_hashes)
+
+
+class EmphasisMode(GenerationParamsState):
+    def __init__(self, texts):
+        super().__init__()
+        if opts.emphasis != 'Original' and any(x for x in texts if '(' in x or '[' in x):
+            self.emphasis = opts.emphasis
+        else:
+            self.emphasis = None
+
+    def __call__(self, extra_generation_params):
+        if self.emphasis:
+            extra_generation_params['Emphasis'] = self.emphasis
+
+
 class TextConditionalModel(torch.nn.Module):
     def __init__(self):
         super().__init__()
@@ -238,12 +264,9 @@ def forward(self, texts):
                 hashes.append(f"{name}: {shorthash}")
 
             if hashes:
-                if self.hijack.extra_generation_params.get("TI hashes"):
-                    hashes.append(self.hijack.extra_generation_params.get("TI hashes"))
-                self.hijack.extra_generation_params["TI hashes"] = ", ".join(hashes)
+                self.hijack.extra_generation_params["TI hashes"] = EmbeddingHashes(hashes)
 
-        if any(x for x in texts if "(" in x or "[" in x) and opts.emphasis != "Original":
-            self.hijack.extra_generation_params["Emphasis"] = opts.emphasis
+        self.hijack.extra_generation_params["Emphasis"] = EmphasisMode(texts)
 
         if self.return_pooled:
             return torch.hstack(zs), zs[0].pooled

modules/util.py

@@ -288,3 +288,18 @@ def compare_sha256(file_path: str, hash_prefix: str) -> bool:
         for chunk in iter(lambda: f.read(blksize), b""):
             hash_sha256.update(chunk)
     return hash_sha256.hexdigest().startswith(hash_prefix.strip().lower())
+
+
+class GenerationParamsState:
+    """A custom class used in StableDiffusionModelHijack for assigning extra_generation_params
+    generation_params assigned using this class will work properly with StableDiffusionProcessing.get_conds_with_caching()
+        if assigned directly the generation_params will not be populated if conda cache is used
+
+    Generation_params of this class will be captured (see StableDiffusionModelHijack.capture_generation_params_state) and stored with conda cache, and will be extracted in StableDiffusionProcessing.apply_hijack_generation_params()
+
+    To use this class, create a subclass with a __call__ method that takes extra_generation_params: dict as input
+
+    Example usage: sd_hijack_clip.EmbeddingHashes, sd_hijack_clip.EmphasisMode
+    """
+    def __call__(self, extra_generation_params: dict):
+        raise NotImplementedError