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

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

pages/index.vue

@@ -30,6 +30,17 @@
             >
               {{ gpuWaking ? 'Starting GPU…' : 'Wake GPU' }}
             </button>
+            <button
+              v-if="showSleepGpu"
+              type="button"
+              class="rounded-full border px-3 py-1 hover:border-amber-300/50 hover:text-white disabled:cursor-not-allowed disabled:opacity-80"
+              :class="gpuSleeping ? 'border-amber-300/40 text-amber-100' : 'border-white/15 text-zinc-300'"
+              :disabled="gpuSleeping || rigBusy"
+              :title="gpuWakeError || (rigBusy ? 'Cancel the running job first' : (gpuSleeping ? 'Stopping Comfy on the GPU' : 'Stop Comfy on the GPU to free VRAM. Does not shut the PC down.'))"
+              @click="sleepGpu"
+            >
+              {{ gpuSleeping ? 'Stopping GPU…' : 'Sleep GPU' }}
+            </button>
           </div>
           <span class="hidden md:block text-zinc-400">{{ userName }}</span>
           <button type="button" class="rounded-full border border-white/10 px-3 py-1 text-zinc-300 hover:border-amber-300/50 hover:text-white" @click="settingsOpen = true">
@@ -2453,6 +2464,7 @@ const comfyOk = ref(false)
 const comfyBusy = ref(false)
 const comfyAsleep = ref(false)
 const gpuWaking = ref(false)
+const gpuSleeping = ref(false)
 const gpuWakeError = ref('')
 const userName = ref('')
 const authEnabled = ref(false)
@@ -2726,8 +2738,17 @@ const liveMaxStep = computed(() => (
 const rigBusy = computed(() => (
   comfyBusy.value || imageComfyBusy.value || videoBusy.value || editBusy.value
 ))
-const showWakeGpu = computed(() => gpuWaking.value || (!comfyOk.value && !imageComfyOk.value))
+const showWakeGpu = computed(() => gpuWaking.value || (!gpuSleeping.value && !comfyOk.value && !imageComfyOk.value))
+const showSleepGpu = computed(() => gpuSleeping.value || ((comfyOk.value || imageComfyOk.value) && !gpuWaking.value))
 const rigBadge = computed(() => {
+  if (gpuSleeping.value) {
+    return {
+      label: 'Stopping…',
+      className: 'border-amber-400/40 text-amber-200',
+      dot: 'bg-amber-400 animate-pulse',
+      title: 'Stopping Comfy on the GPU'
+    }
+  }
   if (gpuWaking.value) {
     return {
       label: 'Starting…',
@@ -2857,6 +2878,7 @@ const recommendHint = computed(() => {
   return 'Send this still and draft to Beast Qwen VL.'
 })
 const gpuSleepHint = computed(() => {
+  if (gpuSleeping.value) return 'Stopping Comfy on the GPU to free VRAM. The PC stays on.'
   if (comfyOk.value || imageComfyOk.value) return ''
   if (gpuWaking.value) return 'Starting Comfy on the GPU. Generate will wait in the job queue until it answers.'
   if (comfyAsleep.value) return 'GPU is asleep. Generate will wake it — the first job can take a few minutes. Or click Wake GPU to pre-warm.'
@@ -3528,7 +3550,7 @@ async function pollComfyHealth() {
 }
 
 async function pokeGpu() {
-  if (gpuWaking.value) return
+  if (gpuWaking.value || gpuSleeping.value) return
   gpuWaking.value = true
   gpuWakeError.value = ''
   try {
@@ -3550,6 +3572,24 @@ async function pokeGpu() {
   }
 }
 
+async function sleepGpu() {
+  if (gpuSleeping.value || gpuWaking.value || rigBusy.value) return
+  gpuSleeping.value = true
+  gpuWakeError.value = ''
+  try {
+    await $fetch('/api/comfy/sleep', { method: 'POST' })
+    comfyOk.value = false
+    imageComfyOk.value = false
+    imageBeastOk.value = false
+    comfyAsleep.value = true
+    await pollComfyHealth()
+  } catch (error: any) {
+    gpuWakeError.value = error?.data?.statusMessage || error?.statusMessage || error?.message || 'Could not stop Comfy.'
+  } finally {
+    gpuSleeping.value = false
+  }
+}
+
 onMounted(async () => {
   try {
     withSound.value = localStorage.getItem('aigen-generate-sound') !== 'false'