Don't zero out conditioning for SD1.5

- it does not like it.

Author: Acly

ai_diffusion/workflow.py

@@ -153,6 +153,8 @@ def load_checkpoint_with_lora(w: ComfyWorkflow, checkpoint: CheckpointInput, mod
     if arch.supports_attention_guidance and checkpoint.self_attention_guidance:
         model = w.apply_self_attention_guidance(model)
 
+    return model, Clip(clip, arch), vae
+
 
 def vae_decode(w: ComfyWorkflow, vae: Output, latent: Output, tiled: bool):
     if tiled:
@@ -253,26 +255,31 @@ def from_input(i: ControlInput):
         return Control(i.mode, ImageOutput(i.image), None, i.strength, i.range)
 
 
+class Clip(NamedTuple):
+    model: Output
+    arch: Arch
+
+
 class TextPrompt:
     text: str
     language: str
     # Cached values to avoid re-encoding the same text for multiple regions and passes
     _output: Output | None = None
-    _clip: Output | None = None  # can be different due to Lora hooks
+    _clip: Clip | None = None  # can be different due to Lora hooks
 
     def __init__(self, text: str, language: str):
         self.text = text
         self.language = language
 
-    def encode(self, w: ComfyWorkflow, clip: Output, style_prompt: str | None = None):
+    def encode(self, w: ComfyWorkflow, clip: Clip, style_prompt: str | None = None):
         text = self.text
         if text != "" and style_prompt:
             text = merge_prompt(text, style_prompt, self.language)
         if self._output is None or self._clip != clip:
             if text and self.language:
                 text = w.translate(text)
-            self._output = w.clip_text_encode(clip, text)
-            if text == "":
+            self._output = w.clip_text_encode(clip.model, text)
+            if text == "" and clip.arch is not Arch.sd15:
                 self._output = w.conditioning_zero_out(self._output)
             self._clip = clip
         return self._output
@@ -286,7 +293,7 @@ class Region:
     control: list[Control] = field(default_factory=list)
     loras: list[LoraInput] = field(default_factory=list)
     is_background: bool = False
-    clip: Output | None = None
+    clip: Clip | None = None
 
     @staticmethod
     def from_input(i: RegionInput, index: int, language: str):
@@ -301,15 +308,15 @@ def from_input(i: RegionInput, index: int, language: str):
             is_background=index == 0,
         )
 
-    def patch_clip(self, w: ComfyWorkflow, clip: Output):
+    def patch_clip(self, w: ComfyWorkflow, clip: Clip):
         if self.clip is None:
             self.clip = clip
             if len(self.loras) > 0:
                 hooks = w.create_hook_lora([(lora.name, lora.strength) for lora in self.loras])
-                self.clip = w.set_clip_hooks(clip, hooks)
+                self.clip = Clip(w.set_clip_hooks(clip.model, hooks), clip.arch)
         return self.clip
 
-    def encode_prompt(self, w: ComfyWorkflow, clip: Output, style_prompt: str | None = None):
+    def encode_prompt(self, w: ComfyWorkflow, clip: Clip, style_prompt: str | None = None):
         return self.positive.encode(w, self.patch_clip(w, clip), style_prompt)
 
     def copy(self):
@@ -384,7 +391,7 @@ def downscale_all_control_images(cond: ConditioningInput, original: Extent, targ
 def encode_text_prompt(
     w: ComfyWorkflow,
     cond: Conditioning,
-    clip: Output,
+    clip: Clip,
     regions: Output | None,
 ):
     if len(cond.regions) <= 1 or all(len(r.loras) == 0 for r in cond.regions):
@@ -413,7 +420,7 @@ def apply_attention_mask(
     w: ComfyWorkflow,
     model: Output,
     cond: Conditioning,
-    clip: Output,
+    clip: Clip,
     shape: Extent | ImageReshape = no_reshape,
 ):
     if len(cond.regions) == 0:
@@ -643,7 +650,7 @@ def scale_refine_and_decode(
     sampling: SamplingInput,
     latent: Output,
     model: Output,
-    clip: Output,
+    clip: Clip,
     vae: Output,
     models: ModelDict,
     tiled_vae: bool,
@@ -1240,7 +1247,7 @@ def get_param(node: ComfyNode, expected_type: type | tuple[type, type] | None =
                 sampling = _sampling_from_style(style, 1.0, is_live)
                 model, clip, vae = load_checkpoint_with_lora(w, checkpoint_input, models)
                 outputs[node.output(0)] = model
-                outputs[node.output(1)] = clip
+                outputs[node.output(1)] = clip.model
                 outputs[node.output(2)] = vae
                 outputs[node.output(3)] = style.style_prompt
                 outputs[node.output(4)] = style.negative_prompt