Upload app.py

app.py

@@ -1887,37 +1887,32 @@ with gr.Blocks(theme=args.theme, elem_id="main", fill_width=True, fill_height=Fa
         copy_prompt_btn_pony.click(gradio_copy_prompt, inputs=[output_text_pony], outputs=[prompt_gui], show_api=False)
 
     # Stable programmatic endpoint
-    from typing import Any, Dict, List, Union, Generator
-    Payload = Union[List[Any], Dict[str, Any]]
-
-    def generate_minimal(payload: Payload) -> Generator[Any, None, None]:
-        # One-call wrapper: optionally preload a model, then delegate to the original generator.
-        # Accepts EITHER:
-        #  - dict: {"args":[...full positional array...], "model_name":..., "vae_model":..., "task":..., "controlnet_model":...}
-        #  - list: [...full positional array...] (legacy, forwarded as-is)
-        # Normalize input
-        if isinstance(payload, dict):
-            args = payload.get("args") or payload.get("data") or []
-            m = payload.get("model_name")
-            v = payload.get("vae_model")
-            t = payload.get("task")
-            c = payload.get("controlnet_model")
-            # Optional preload if all four fields are present
-            if all(x is not None for x in (m, v, t, c)):
-                for _ in sd_gen.load_new_model(m, v, t, c):
-                    # If you want to stream loader status, you could `yield {"status": _}`
-                    pass
-        else:
-            # Legacy: payload is the full positional array already
-            args = payload
-        # Delegate to the existing streaming generator; preserves SSE behavior
-        yield from sd_gen_generate_pipeline(*args)
+    # --- minimal sync wrapper + stable API ---
+    from typing import Any, List, Tuple, Optional
+
+    def generate_minimal(
+        args: List[Any],
+        model_name: str,
+        vae_model: str,
+        task: str,
+        controlnet_model: str
+    ) -> Tuple[str, Optional[List[str]], Optional[str]]:
+        # Preload the requested model (drain loader yields so the model is ready)
+        for _ in sd_gen.load_new_model(model_name, vae_model, task, controlnet_model):
+            pass
+        # Run the existing streaming generator and collapse to the final triple
+        last_status, last_imgs, last_info = "START", None, None
+        for status, imgs, info in sd_gen_generate_pipeline(*args):
+            last_status, last_imgs, last_info = status, imgs, info
+        return last_status or "COMPLETE", last_imgs, last_info
 
     gr.api(
         generate_minimal,
-        api_name="generate",
-        api_description="Accepts a list (positional) or a dict with {'args': [...]} + optional loader fields.",
-        queue=True,
+        api_name="generate_image",        # => POST /gradio_api/call/generate_image
+        api_description="Preload model then generate with full positional args. Returns final (status, images, info).",
+        show_api=True,                    # force listing on the API page
+        queue=True,                       # reuse app queue
+        concurrency_id="gpu"              # share GPU queue with other heavy events
     )
 
     gr.LoginButton()