Feature: z-image + metadata node

invokeai/app/invocations/flux_denoise.py

@@ -47,6 +47,7 @@
     pack,
     unpack,
 )
+from invokeai.backend.flux.schedulers import FLUX_SCHEDULER_LABELS, FLUX_SCHEDULER_MAP, FLUX_SCHEDULER_NAME_VALUES
 from invokeai.backend.flux.text_conditioning import FluxReduxConditioning, FluxTextConditioning
 from invokeai.backend.model_manager.taxonomy import BaseModelType, FluxVariantType, ModelFormat, ModelType
 from invokeai.backend.patches.layer_patcher import LayerPatcher
@@ -63,7 +64,7 @@
     title="FLUX Denoise",
     tags=["image", "flux"],
     category="image",
-    version="4.1.0",
+    version="4.2.0",
 )
 class FluxDenoiseInvocation(BaseInvocation):
     """Run denoising process with a FLUX transformer model."""
@@ -132,6 +133,12 @@ class FluxDenoiseInvocation(BaseInvocation):
     num_steps: int = InputField(
         default=4, description="Number of diffusion steps. Recommended values are schnell: 4, dev: 50."
     )
+    scheduler: FLUX_SCHEDULER_NAME_VALUES = InputField(
+        default="euler",
+        description="Scheduler (sampler) for the denoising process. 'euler' is fast and standard. "
+        "'heun' is 2nd-order (better quality, 2x slower). 'lcm' is optimized for few steps.",
+        ui_choice_labels=FLUX_SCHEDULER_LABELS,
+    )
     guidance: float = InputField(
         default=4.0,
         description="The guidance strength. Higher values adhere more strictly to the prompt, and will produce less diverse images. FLUX dev only, ignored for schnell.",
@@ -242,6 +249,12 @@ def _run_diffusion(
             shift=not is_schnell,
         )
 
+        # Create scheduler if not using default euler
+        scheduler = None
+        if self.scheduler in FLUX_SCHEDULER_MAP:
+            scheduler_class = FLUX_SCHEDULER_MAP[self.scheduler]
+            scheduler = scheduler_class(num_train_timesteps=1000)
+
         # Clip the timesteps schedule based on denoising_start and denoising_end.
         timesteps = clip_timestep_schedule_fractional(timesteps, self.denoising_start, self.denoising_end)
 
@@ -426,6 +439,7 @@ def _run_diffusion(
                 img_cond=img_cond,
                 img_cond_seq=img_cond_seq,
                 img_cond_seq_ids=img_cond_seq_ids,
+                scheduler=scheduler,
             )
 
         x = unpack(x.float(), self.height, self.width)

invokeai/app/invocations/metadata_linked.py

@@ -52,6 +52,7 @@
 )
 from invokeai.app.invocations.scheduler import SchedulerOutput
 from invokeai.app.invocations.t2i_adapter import T2IAdapterField, T2IAdapterInvocation
+from invokeai.app.invocations.z_image_denoise import ZImageDenoiseInvocation
 from invokeai.app.services.shared.invocation_context import InvocationContext
 from invokeai.backend.model_manager.taxonomy import BaseModelType, ModelType, SubModelType
 from invokeai.backend.stable_diffusion.schedulers.schedulers import SCHEDULER_NAME_VALUES
@@ -729,6 +730,52 @@ def _loras_to_json(obj: Union[Any, list[Any]]):
         return LatentsMetaOutput(**params, metadata=MetadataField.model_validate(md))
 
 
+@invocation(
+    "z_image_denoise_meta",
+    title=f"{ZImageDenoiseInvocation.UIConfig.title} + Metadata",
+    tags=["z-image", "latents", "denoise", "txt2img", "t2i", "t2l", "img2img", "i2i", "l2l"],
+    category="latents",
+    version="1.0.0",
+)
+class ZImageDenoiseMetaInvocation(ZImageDenoiseInvocation, WithMetadata):
+    """Run denoising process with a Z-Image transformer model + metadata."""
+
+    def invoke(self, context: InvocationContext) -> LatentsMetaOutput:
+        def _loras_to_json(obj: Union[Any, list[Any]]):
+            if not isinstance(obj, list):
+                obj = [obj]
+
+            output: list[dict[str, Any]] = []
+            for item in obj:
+                output.append(
+                    LoRAMetadataField(
+                        model=item.lora,
+                        weight=item.weight,
+                    ).model_dump(exclude_none=True, exclude={"id", "type", "is_intermediate", "use_cache"})
+                )
+            return output
+
+        obj = super().invoke(context)
+
+        md: Dict[str, Any] = {} if self.metadata is None else self.metadata.root
+        md.update({"width": obj.width})
+        md.update({"height": obj.height})
+        md.update({"steps": self.steps})
+        md.update({"guidance": self.guidance_scale})
+        md.update({"denoising_start": self.denoising_start})
+        md.update({"denoising_end": self.denoising_end})
+        md.update({"scheduler": self.scheduler})
+        md.update({"model": self.transformer.transformer})
+        md.update({"seed": self.seed})
+        if len(self.transformer.loras) > 0:
+            md.update({"loras": _loras_to_json(self.transformer.loras)})
+
+        params = obj.__dict__.copy()
+        del params["type"]
+
+        return LatentsMetaOutput(**params, metadata=MetadataField.model_validate(md))
+
+
 @invocation(
     "metadata_to_vae",
     title="Metadata To VAE",