modified: app.py

app.py

@@ -8,36 +8,62 @@ def set_client_for_session(request: gr.Request):
     ltx_client = Client("KingNish/ltx-video-distilled", headers={"x-ip-token": x_ip_token})
     return flux_client, ltx_client
 
-def text_to_image(prompt, image, client):
-    result = client.predict(
+def image_to_image(prompt, image, flux_client):
+    prompt = "You are an AI image-to-image editor that transforms a user-provided input image based on their prompt while maintaining consistency and fidelity to the original image. Generate an output image that accurately reflects the user's requested modifications, preserving key elements like style, composition, and context from the input image unless explicitly instructed otherwise. " + prompt
+    image = flux_client.predict(
         input_image=handle_file(image),
         prompt=prompt,
         seed=0,
         randomize_seed=True,
         guidance_scale=2.5,
-        steps=30,
+        steps=20,
         api_name="/infer"
     )
+    print(image)
+    return image[0]
+
+def image_to_video(prompt, image, ltx_client):
+    result = ltx_client.predict(
+            prompt=prompt,
+            input_image_url=None,
+            middle_image_url=handle_file(image),
+            final_image_url=None,
+            duration_ui=4,
+            api_name="/generate_video"
+    )
     print(result)
     return result
 
+def personalized_video(prompt, image, flux_client, ltx_client, request: gr.Request):
+    x_ip_token = request.headers['x-ip-token']
+    print(x_ip_token)
+    image = image_to_image(prompt, image, flux_client)
+    yield image, None
+    video = image_to_video(prompt, image, ltx_client)
+    yield image, video
+
 with gr.Blocks() as demo:
     flux_client = gr.State(None)
     ltx_client = gr.State(None)
-    image = gr.Image()
     input_image = gr.Image(type="filepath")
     prompt = gr.Textbox(max_lines=1)
-    prompt.submit(text_to_image, [prompt, input_image, flux_client], [image])
+    output_video = gr.Video()
+    submit_button = gr.Button("Submit")
+    submit_button.click(personalized_video, [prompt, input_image, flux_client, ltx_client], [output_video])
+    prompt.submit(personalized_video, [prompt, input_image, flux_client, ltx_client], [output_video])
 
-    # gr.Examples(
-    #     examples=[
-    #         ["A sunny day", "https://black-forest-labs-flux-1-kontext-dev.hf.space/gradio_api/file=/tmp/gradio/927b05b08b8ec2201f7d95722471d2089d4197f152c9c07d0bc9f9672962d03b/flowers.png"],
-    #     ],
-    #     inputs=[prompt, input_image],
-    #     outputs=image,
-    # )
+    gr.Examples(
+        examples=[
+            ["A sunny day", "https://black-forest-labs-flux-1-kontext-dev.hf.space/gradio_api/file=/tmp/gradio/927b05b08b8ec2201f7d95722471d2089d4197f152c9c07d0bc9f9672962d03b/flowers.png", None, None],
+        ],
+        inputs=[prompt, input_image, flux_client, ltx_client],
+        outputs=image,
+        fn=personalized_video,
+        cache_examples=True,
+        cache_mode = "eager"
+    )
 
-    demo.load(set_client_for_session, None, [flux_client, ltx_client])
+    # demo.load(set_client_for_session, None, [flux_client, ltx_client])
 
 demo.launch()