add : structured output

.gradio/cached_examples/19/indices.csv

@@ -0,0 +1,2 @@
+1
+0

.gradio/cached_examples/19/log.csv

@@ -0,0 +1,3 @@
+Result,Seed,timestamp
+"{""path"": "".gradio/cached_examples/19/Result/b62d7595ca8cb25f5456/image.webp"", ""url"": ""/gradio_api/file=/tmp/gradio/6414decced94e8faa0bebd4d105bf6fbcbfb496d221122a4cf1b1a449de31f87/image.webp"", ""size"": null, ""orig_name"": ""image.webp"", ""mime_type"": null, ""is_stream"": false, ""meta"": {""_type"": ""gradio.FileData""}}",42,2025-03-10 06:01:40.924059
+"{""path"": "".gradio/cached_examples/19/Result/36f4c25c9f9b3f438281/image.webp"", ""url"": ""/gradio_api/file=/tmp/gradio/04be7c4b38d246744b99e6ac36b7ff571528369078988e11fbf5f63b9dee6d68/image.webp"", ""size"": null, ""orig_name"": ""image.webp"", ""mime_type"": null, ""is_stream"": false, ""meta"": {""_type"": ""gradio.FileData""}}",42,2025-03-10 06:07:53.994344

app.py

@@ -46,13 +46,14 @@ def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidan
         seed = random.randint(0, MAX_SEED)
     generator = torch.Generator().manual_seed(seed)
 
-    # refined_prompt = run_gemini(
-    #     target_prompt=prompt,
-    #     prompt_in_path="prompt.json",
-    # )
+    refined_prompt = run_gemini(
+        target_prompt=prompt,
+        prompt_in_path="prompt.json",
+        output_structure=RefinedPrompt,
+    )
     
     for img in pipe.flux_pipe_call_that_returns_an_iterable_of_images(
-            prompt=prompt,
+            prompt=refined_prompt.prompt,
             guidance_scale=guidance_scale,
             num_inference_steps=num_inference_steps,
             width=width,

llm_wrapper.py

@@ -60,6 +60,7 @@ def encode_image(image_source):
 def run_gemini(
     target_prompt: str,
     prompt_in_path: str,
+    output_structure
     img_in_data: str = None,
     model: str = "gemini-2.0-flash",
 ) -> str:
@@ -88,6 +89,11 @@ def run_gemini(
     chat_completion = client.models.generate_content(
         model=model,
         contents=input_content,
+        config={
+            "system_instruction": system_prompt,
+            "response_mime_type": "application/json",
+            "response_schema": output_structure
+        }
     )
 
     chat_output = chat_completion.parsed