diff --git a/tests/fixtures/architecture_wire_inventory.json b/tests/fixtures/architecture_wire_inventory.json
index 7016545f..fb5ecd92 100644
--- a/tests/fixtures/architecture_wire_inventory.json
+++ b/tests/fixtures/architecture_wire_inventory.json
@@ -631,6 +631,12 @@
       "classification": "behavior",
       "reason": "Imports the public Zustand facade; it should survive slice extraction unchanged."
     },
+    {
+      "file": "ui/tests/studioSubmissionFailure.test.ts",
+      "target": "ui/src/stores/useStore.ts",
+      "classification": "behavior",
+      "reason": "Imports the public Zustand facade; it should survive slice extraction unchanged."
+    },
     {
       "file": "ui/tests/tabFilterSearch.test.tsx",
       "target": "ui/src/stores/useStore.ts",
diff --git a/ui/src/components/Sidebar/Sidebar.tsx b/ui/src/components/Sidebar/Sidebar.tsx
index 2648e6fe..645dbab9 100644
--- a/ui/src/components/Sidebar/Sidebar.tsx
+++ b/ui/src/components/Sidebar/Sidebar.tsx
@@ -305,9 +305,11 @@ export function Sidebar() {
           <div className="flex-1 min-w-0">
             <ModelSelector />
           </div>
-          <div className="shrink-0">
-            <GenerateButton />
-          </div>
+          {!(isAudio && audioSubMode === 'mixer') && (
+            <div className="shrink-0">
+              <GenerateButton />
+            </div>
+          )}
         </div>
       </div>
       )}
diff --git a/ui/src/features/studio/studioSubmission.ts b/ui/src/features/studio/studioSubmission.ts
index 96486c2a..53d4e30e 100644
--- a/ui/src/features/studio/studioSubmission.ts
+++ b/ui/src/features/studio/studioSubmission.ts
@@ -1,4 +1,5 @@
 import * as api from '../../api/client'
+import i18n from '../../i18n'
 
 type Preparation = typeof import('./imageCommandSubmission').prepareStudioSubmission
 type Inputs = Parameters<Preparation>
@@ -33,6 +34,11 @@ export async function prepareStudioSubmission(
       const implementation = await import('./sfxCommandSubmission')
       return await implementation.prepareStudioSfxSubmission(params, before, current, context, referenceErrors)
     }
+    if (before.generationMode === 'audio') {
+      // Mixer is ffmpeg-only. An unknown Audio tab must not inherit leftover
+      // Speech/Music/SFX params and POST them to the legacy GPU endpoint.
+      throw new Error(i18n.t('studio:commands.audioNotGenerative'))
+    }
   } catch (error) {
     return { params, submit: () => Promise.reject(error) }
   }
diff --git a/ui/src/i18n/locales/en/studio.json b/ui/src/i18n/locales/en/studio.json
index 176e408b..ac2ec637 100644
--- a/ui/src/i18n/locales/en/studio.json
+++ b/ui/src/i18n/locales/en/studio.json
@@ -939,7 +939,8 @@
     "panelUnavailable": "The image request could not be shown. Open Studio and try again.",
     "pendingInvalid": "The saved submission could not be read. It has been preserved for recovery.",
     "referenceFailed": "A selected reference could not be prepared. No image was submitted.",
-    "conflictingGuides": "Conflicting image guides are selected. Choose one control image and mask before generating."
+    "conflictingGuides": "Conflicting image guides are selected. Choose one control image and mask before generating.",
+    "audioNotGenerative": "This Audio tab does not start a generation. Use Speech, Music or SFX, or Mix & Save in Mixer."
   },
   "speechCommands": {
     "prepared": "Speech request ready",
diff --git a/ui/src/i18n/locales/es/studio.json b/ui/src/i18n/locales/es/studio.json
index 9df6f163..d350b524 100644
--- a/ui/src/i18n/locales/es/studio.json
+++ b/ui/src/i18n/locales/es/studio.json
@@ -939,7 +939,8 @@
     "panelUnavailable": "No se pudo mostrar la solicitud de imagen. Abre Studio y vuelve a intentarlo.",
     "pendingInvalid": "No se pudo leer la solicitud guardada. Se ha conservado para recuperarla.",
     "referenceFailed": "No se pudo preparar una referencia seleccionada. No se ha enviado ninguna imagen.",
-    "conflictingGuides": "Hay guías de imagen distintas seleccionadas. Elige una imagen de control y una máscara antes de generar."
+    "conflictingGuides": "Hay guías de imagen distintas seleccionadas. Elige una imagen de control y una máscara antes de generar.",
+    "audioNotGenerative": "Esta pestaña de Audio no inicia una generación. Usa Voz, Música o SFX, o Mezclar y guardar en el Mezclador."
   },
   "speechCommands": {
     "prepared": "Solicitud de voz preparada",
diff --git a/ui/src/stores/useStore.ts b/ui/src/stores/useStore.ts
index b5178284..0eb80a95 100644
--- a/ui/src/stores/useStore.ts
+++ b/ui/src/stores/useStore.ts
@@ -4121,6 +4121,9 @@ export const useStore = create<AppState>((set, get) => {
   isGenerating: false,
   startGeneration: async (scheduledPrompt, submissionContext) => {
     const initialState = get()
+    if (initialState.generationMode === 'audio' && initialState.audioSubMode === 'mixer') {
+      throw new Error(i18n.t('studio:commands.audioNotGenerative'))
+    }
 
     // Studio Prompt Scheduler: each non-empty line becomes its own normal
     // generation request. Submitting the requests one at a time preserves the
diff --git a/ui/tests/studioSubmissionFailure.test.ts b/ui/tests/studioSubmissionFailure.test.ts
index 98dd84af..8d6a1810 100644
--- a/ui/tests/studioSubmissionFailure.test.ts
+++ b/ui/tests/studioSubmissionFailure.test.ts
@@ -22,27 +22,71 @@ test('a missing image chunk produces a failed submission for the existing job er
   assert.equal(requests, 0)
 })
 
-for (const mode of ['audio', 'video']) {
-  test(`${mode} submits through its native API without loading image code`, async () => {
-    const params = { prompt: 'literal\nsecond line', generation_mode: mode, workspace: 'command-qa' }
-    const before = state(mode, params)
-    let loads = 0
-    let sent: unknown
-    globalThis.fetch = async (_url, options) => {
-      sent = JSON.parse(String(options?.body))
-      return new Response(JSON.stringify({ job_id: 'native-job', status: 'queued' }), { status: 200 })
+test('video submits through its native API without loading image code', async () => {
+  const params = { prompt: 'literal\nsecond line', generation_mode: 'video', workspace: 'command-qa' }
+  const before = state('video', params)
+  let loads = 0
+  let sent: unknown
+  globalThis.fetch = async (_url, options) => {
+    sent = JSON.parse(String(options?.body))
+    return new Response(JSON.stringify({ job_id: 'native-job', status: 'queued' }), { status: 200 })
+  }
+  const submission = await prepareStudioSubmission(params, before, () => before, undefined, [], async () => {
+    loads += 1
+    throw new Error('Image chunk unavailable')
+  })
+  const result = await submission.submit()
+  assert.equal(result.job_id, 'native-job')
+  assert.equal(loads, 0)
+  assert.deepEqual(sent, params)
+})
+
+function audioState(subMode: string, params: Record<string, unknown>) {
+  return { ...state('audio', params), audioSubMode: subMode } as Parameters<typeof prepareStudioSubmission>[1]
+}
+
+for (const subMode of ['mixer', '', undefined]) {
+  test(`audio ${subMode === undefined ? 'without a sub-mode' : `tab ${JSON.stringify(subMode)}`} does not fall back to the legacy GPU endpoint`, async () => {
+    const params = {
+      prompt: 'leftover speech lyrics',
+      model_type: 'kugelaudio_0_open',
+      generation_mode: 'audio',
+      workspace: 'command-qa',
+      _audio_sub_mode: subMode ?? 'mixer',
     }
-    const submission = await prepareStudioSubmission(params, before, () => before, undefined, [], async () => {
-      loads += 1
-      throw new Error('Image chunk unavailable')
-    })
-    const result = await submission.submit()
-    assert.equal(result.job_id, 'native-job')
-    assert.equal(loads, 0)
-    assert.deepEqual(sent, params)
+    const before = subMode === undefined ? state('audio', params) : audioState(subMode, params)
+    let requests = 0
+    globalThis.fetch = async () => { requests += 1; throw new Error('Unexpected POST') }
+    const submission = await prepareStudioSubmission(params, before, () => before)
+    await assert.rejects(submission.submit, /does not start a generation|no inicia una generación/)
+    assert.equal(requests, 0)
   })
 }
 
+test('startGeneration refuses Mixer before creating a job or POSTing', { concurrency: false }, async () => {
+  const { useStore } = await import('../src/stores/useStore.ts')
+  const before = useStore.getState()
+  let requests = 0
+  globalThis.fetch = async () => { requests += 1; throw new Error('Unexpected POST') }
+  useStore.setState({
+    generationMode: 'audio',
+    audioSubMode: 'mixer',
+    params: { ...before.params, model_type: 'kugelaudio_0_open', prompt: 'leftover speech lyrics' },
+    jobs: [],
+    llmStatus: { ...before.llmStatus, loaded: false },
+  })
+  try {
+    await assert.rejects(
+      () => useStore.getState().startGeneration(),
+      /does not start a generation|no inicia una generación/,
+    )
+    assert.equal(useStore.getState().jobs.length, 0)
+    assert.equal(requests, 0)
+  } finally {
+    useStore.setState(before)
+  }
+})
+
 test('legacy image control and mask fields are translated in the detached V2 snapshot', async () => {
   const params = { workspace: 'command-qa', prompt: 'Use this control image literally', model_type: 'pi_flux2',
     resolution: '512x512', num_inference_steps: 4, seed: 42, guidance_scale: 1,