perf: generate even faster with one step prediction (#1)

* perf: generate even faster with one step prediction

* upscale up to 8x

app/page.tsx

@@ -25,7 +25,7 @@ const INPUT_DEFAULTS = {
   image_size: "square_hd",
   sync_mode: true,
   num_images: 1,
-  num_inference_steps: "2",
+  num_inference_steps: "1",
 };
 
 export default function Lightning() {
@@ -44,22 +44,36 @@ export default function Lightning() {
     },
   });
 
-  const timer = useRef<any | undefined>(undefined);
+  const timer_2x = useRef<any | undefined>(undefined);
+  const timer_4x = useRef<any | undefined>(undefined);
+  const timer_8x = useRef<any | undefined>(undefined);
+  const timers = {
+    2: timer_2x,
+    4: timer_4x,
+    8: timer_8x,
+  };
 
   const handleOnChange = async (prompt: string) => {
-    if (timer.current) {
-      clearTimeout(timer.current);
-    }
+
+    Object.values(timers).forEach((t) => {
+      if (t.current) {
+        clearTimeout(t.current);
+      }
+    });
     setPrompt(prompt);
     const input = {
       ...INPUT_DEFAULTS,
       prompt: prompt,
       seed: seed ? Number(seed) : Number(randomSeed()),
     };
     connection.send(input);
-    timer.current = setTimeout(() => {
-      connection.send({ ...input, num_inference_steps: "4" });
-    }, 500);
+    // Iterate over the timers and set a timeout for each with
+    // the inference step count as the key
+    for (const [key, value] of Object.entries(timers)) {
+      value.current = setTimeout(() => {
+        connection.send({ ...input, num_inference_steps: key });
+      }, 100 * Number(key));
+    }
   };
 
   useEffect(() => {
@@ -69,7 +83,7 @@ export default function Lightning() {
     // initial image
     connection.send({
       ...INPUT_DEFAULTS,
-      num_inference_steps: "4",
+      num_inference_steps: "8",
       prompt: prompt,
       seed: seed ? Number(seed) : Number(randomSeed()),
     });