feat(z-image): add metadata recall for VAE and Qwen3 encoder

Add support for saving and recalling Z-Image component models (VAE and
Qwen3 Encoder) in image metadata.

Backend:
- Add qwen3_encoder field to CoreMetadataInvocation (version 2.1.0)

Frontend:
- Add vae and qwen3_encoder to Z-Image graph metadata
- Add Qwen3EncoderModel metadata handler for recall
- Add ZImageVAEModel metadata handler (uses zImageVaeModelSelected
  instead of vaeSelected to set Z-Image-specific VAE state)
- Add qwen3Encoder translation key

This enables "Recall Parameters" / "Remix Image" to restore the VAE
and Qwen3 Encoder settings used for Z-Image generations.

invokeai/app/invocations/metadata.py

@@ -170,7 +170,7 @@ GENERATION_MODES = Literal[
     title="Core Metadata",
     tags=["metadata"],
     category="metadata",
-    version="2.0.0",
+    version="2.1.0",
     classification=Classification.Internal,
 )
 class CoreMetadataInvocation(BaseInvocation):
@@ -221,6 +221,10 @@ class CoreMetadataInvocation(BaseInvocation):
         default=None,
         description="The VAE used for decoding, if the main model's default was not used",
     )
+    qwen3_encoder: Optional[ModelIdentifierField] = InputField(
+        default=None,
+        description="The Qwen3 text encoder model used for Z-Image inference",
+    )
 
     # High resolution fix metadata.
     hrf_enabled: Optional[bool] = InputField(

invokeai/frontend/web/public/locales/en.json

@@ -839,6 +839,7 @@
         "parameterSet": "Parameter {{parameter}} set",
         "parsingFailed": "Parsing Failed",
         "positivePrompt": "Positive Prompt",
+        "qwen3Encoder": "Qwen3 Encoder",
         "recallParameters": "Recall Parameters",
         "recallParameter": "Recall {{label}}",
         "scheduler": "Scheduler",

invokeai/frontend/web/src/features/metadata/parsing.tsx

@@ -31,6 +31,8 @@ import {
   setSteps,
   vaeSelected,
   widthChanged,
+  zImageQwen3EncoderModelSelected,
+  zImageVaeModelSelected,
 } from 'features/controlLayers/store/paramsSlice';
 import { refImagesRecalled } from 'features/controlLayers/store/refImagesSlice';
 import type { CanvasMetadata, LoRA, RefImageState } from 'features/controlLayers/store/types';
@@ -694,6 +696,51 @@ const VAEModel: SingleMetadataHandler<ParameterVAEModel> = {
 };
 //#endregion VAEModel
 
+//#region Qwen3EncoderModel
+const Qwen3EncoderModel: SingleMetadataHandler<ModelIdentifierField> = {
+  [SingleMetadataKey]: true,
+  type: 'Qwen3EncoderModel',
+  parse: async (metadata, store) => {
+    const raw = getProperty(metadata, 'qwen3_encoder');
+    const parsed = await parseModelIdentifier(raw, store, 'qwen3_encoder');
+    assert(parsed.type === 'qwen3_encoder');
+    return Promise.resolve(parsed);
+  },
+  recall: (value, store) => {
+    store.dispatch(zImageQwen3EncoderModelSelected(value));
+  },
+  i18nKey: 'metadata.qwen3Encoder',
+  LabelComponent: MetadataLabel,
+  ValueComponent: ({ value }: SingleMetadataValueProps<ModelIdentifierField>) => (
+    <MetadataPrimitiveValue value={`${value.name} (${value.base.toUpperCase()})`} />
+  ),
+};
+//#endregion Qwen3EncoderModel
+
+//#region ZImageVAEModel
+const ZImageVAEModel: SingleMetadataHandler<ModelIdentifierField> = {
+  [SingleMetadataKey]: true,
+  type: 'ZImageVAEModel',
+  parse: async (metadata, store) => {
+    const raw = getProperty(metadata, 'vae');
+    const parsed = await parseModelIdentifier(raw, store, 'vae');
+    assert(parsed.type === 'vae');
+    // Only recall if the current main model is Z-Image
+    const base = selectBase(store.getState());
+    assert(base === 'z-image', 'ZImageVAEModel handler only works with Z-Image models');
+    return Promise.resolve(parsed);
+  },
+  recall: (value, store) => {
+    store.dispatch(zImageVaeModelSelected(value));
+  },
+  i18nKey: 'metadata.vae',
+  LabelComponent: MetadataLabel,
+  ValueComponent: ({ value }: SingleMetadataValueProps<ModelIdentifierField>) => (
+    <MetadataPrimitiveValue value={`${value.name} (${value.base.toUpperCase()})`} />
+  ),
+};
+//#endregion ZImageVAEModel
+
 //#region LoRAs
 const LoRAs: CollectionMetadataHandler<LoRA[]> = {
   [CollectionMetadataKey]: true,
@@ -928,6 +975,8 @@ export const ImageMetadataHandlers = {
   RefinerDenoisingStart,
   MainModel,
   VAEModel,
+  Qwen3EncoderModel,
+  ZImageVAEModel,
   LoRAs,
   CanvasLayers,
   RefImages,

invokeai/frontend/web/src/features/nodes/util/graph/generation/buildZImageGraph.ts

@@ -123,6 +123,8 @@ export const buildZImageGraph = async (arg: GraphBuilderArg): Promise<GraphBuild
     negative_prompt: prompts.negative,
     model: Graph.getModelMetadataField(modelConfig),
     steps,
+    vae: zImageVaeModel ?? undefined,
+    qwen3_encoder: zImageQwen3EncoderModel ?? undefined,
   });
   g.addEdgeToMetadata(seed, 'value', 'seed');
   g.addEdgeToMetadata(positivePrompt, 'value', 'positive_prompt');