Add a Sleep GPU button that stops Comfy to free VRAM.

Co-authored-by: Cursor <cursoragent@cursor.com>

components/SettingsModal.vue

@@ -10,6 +10,36 @@
           <button type="button" class="rounded-full border border-white/10 px-3 py-1 text-sm text-zinc-400 hover:text-white" @click="emit('close')">Close</button>
         </div>
 
+        <section class="mt-6">
+          <h3 class="text-sm font-semibold uppercase tracking-wider text-zinc-500">GPU</h3>
+          <div class="mt-2 space-y-3 rounded-2xl border border-white/10 bg-zinc-950/50 px-4 py-3">
+            <p class="text-sm text-zinc-400">
+              Stops Comfy on the desktop GPU so it drops out of VRAM. The PC stays on. Generate or Wake GPU will start it again.
+            </p>
+            <div class="flex flex-wrap items-center gap-2">
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-sm disabled:cursor-not-allowed disabled:opacity-60"
+                :class="gpuSleeping ? 'border-amber-300/40 text-amber-100' : 'border-white/15 text-zinc-200 hover:border-amber-300/50 hover:text-white'"
+                :disabled="gpuSleeping || gpuWaking"
+                @click="sleepGpu"
+              >
+                {{ gpuSleeping ? 'Stopping GPU…' : 'Sleep GPU' }}
+              </button>
+              <button
+                type="button"
+                class="rounded-xl border px-3 py-2 text-sm disabled:cursor-not-allowed disabled:opacity-60"
+                :class="gpuWaking ? 'border-amber-300/40 text-amber-100' : 'border-amber-400/40 text-amber-200 hover:border-amber-300 hover:text-amber-50'"
+                :disabled="gpuWaking || gpuSleeping"
+                @click="wakeGpu"
+              >
+                {{ gpuWaking ? 'Starting GPU…' : 'Wake GPU' }}
+              </button>
+            </div>
+            <p v-if="gpuMessage" class="text-xs" :class="gpuError ? 'text-red-300' : 'text-zinc-500'">{{ gpuMessage }}</p>
+          </div>
+        </section>
+
         <section class="mt-6">
           <h3 class="text-sm font-semibold uppercase tracking-wider text-zinc-500">Storage</h3>
           <div class="mt-2 rounded-2xl border border-white/10 bg-zinc-950/50 px-4 py-3">
@@ -256,6 +286,10 @@ const loginNext = ref('')
 const drafts = reactive<Record<string, { name: string; protect: boolean; password: string; currentPassword: string }>>({})
 const storageLoading = ref(true)
 const storageError = ref('')
+const gpuSleeping = ref(false)
+const gpuWaking = ref(false)
+const gpuError = ref(false)
+const gpuMessage = ref('')
 const storage = ref<{
   libraryBytes: number
   breakdown: { clips: number; stills: number; tmp: number; other: number }
@@ -392,6 +426,38 @@ async function loadStorage() {
   }
 }
 
+async function sleepGpu() {
+  if (gpuSleeping.value || gpuWaking.value) return
+  gpuSleeping.value = true
+  gpuError.value = false
+  gpuMessage.value = 'Stopping Comfy on the GPU…'
+  try {
+    await $fetch('/api/comfy/sleep', { method: 'POST' })
+    gpuMessage.value = 'Comfy is stopped. VRAM is free until you generate or wake the GPU.'
+  } catch (error: any) {
+    gpuError.value = true
+    gpuMessage.value = error?.data?.statusMessage || error?.statusMessage || 'Could not stop Comfy.'
+  } finally {
+    gpuSleeping.value = false
+  }
+}
+
+async function wakeGpu() {
+  if (gpuWaking.value || gpuSleeping.value) return
+  gpuWaking.value = true
+  gpuError.value = false
+  gpuMessage.value = 'Starting Comfy on the GPU…'
+  try {
+    await $fetch('/api/comfy/wake', { method: 'POST' })
+    gpuMessage.value = 'Wake sent. The first job can still take a couple of minutes if Comfy is cold.'
+  } catch (error: any) {
+    gpuError.value = true
+    gpuMessage.value = error?.data?.statusMessage || error?.statusMessage || 'Could not wake the GPU.'
+  } finally {
+    gpuWaking.value = false
+  }
+}
+
 async function loadLoraKeys() {
   loraRowsLoading.value = true
   try {