auto-git:

[change] generate_equirect.py

generate_equirect.py

@@ -485,7 +485,6 @@ def generate(
 
     device = select_device()
     is_sdxl = "sdxl" in model_path.lower()
-    scale = guidance  # keep inpaint guidance in sync with cfg guidance
     enable_upscale = bool(upscale and upscale != "none")
 
     os.makedirs(work_dir, exist_ok=True)
@@ -613,23 +612,11 @@ def generate(
                     f"Unsupported scheduler '{scheduler}'. "
                     "Try one of: euler, euler_a, heun, ddim, dpmsolver, dpmsolver-sde."
                 )
-        gen_pipe.enable_attention_slicing()
+        configure_pipeline_memory(gen_pipe)
-        if hasattr(gen_pipe, "enable_vae_tiling"):
-            gen_pipe.enable_vae_tiling()
-        if hasattr(gen_pipe, "enable_vae_slicing"):
-            gen_pipe.enable_vae_slicing()
         if "pipe_kwargs" in locals():
             del pipe_kwargs
 
-        def progress_cb(phase: str, current: int, total: int):
+        progress_cb = make_progress_cb(enable_upscale, seam_inpaint)
-            payload = {
-                "phase": phase,
-                "current": current,
-                "total": total,
-                "upscale": enable_upscale,
-                "seamInpaint": seam_inpaint,
-            }
-            print(f"PROGRESS {json.dumps(payload)}", flush=True)
 
         print("→ Generating equirectangular HDRI…")
         progress_cb("gen", 0, steps)
@@ -659,71 +646,35 @@ def generate(
         image.save(gen_path)
         print(f"→ Saved initial image to {gen_path}")
 
-        seamless_path = os.path.join(tempdir, os.path.basename(output_path))
         if seam_inpaint:
-            shift_amt = width // 2
+            del image
-            mask_w = width // 8
+            clear_torch_cache(device)
+            restart_for_postprocess(
+                prompt,
+                gen_path,
+                output_path,
+                tempdir,
+                work_dir,
+                upscale,
+                steps,
+                guidance,
+                width,
+                height,
+                seam_inpaint,
+            )
 
-            shifted = shift_image(image, shift_amt)
+        return postprocess_image(
-            mask = create_mask(width, height, mask_w)
+            prompt,
-
+            gen_path,
-            inpaint_pipe = StableDiffusionInpaintPipeline.from_pretrained(
+            output_path,
-                INPAINT_MODEL,
+            tempdir,
-                torch_dtype=torch.float32
+            upscale=upscale,
-            ).to(device)
+            steps=steps,
-            inpaint_pipe.enable_attention_slicing()
+            guidance=guidance,
-
-            print("→ Inpainting seam for seamless tiling…")
-            progress_cb("inpaint", 0, steps)
-            inpainted = inpaint_pipe(
-                prompt=prompt,
-                image=shifted,
-                mask_image=mask,
-                num_inference_steps=steps,
-                guidance_scale=scale,
             width=width,
             height=height,
-                callback_steps=1,
+            seam_inpaint=False,
-                callback=lambda step, timestep, kwargs: progress_cb("inpaint", step + 1, steps),
-            ).images[0]
-
-            inpainted = unshift_image(inpainted, shift_amt)
-            inpainted.save(seamless_path)
-            print(f"→ Crafted seamless image: {seamless_path}")
-            final_source = inpainted
-        else:
-            image.save(seamless_path)
-            print(f"→ Using raw output (seam inpaint disabled): {seamless_path}")
-            final_source = image
-
-        final_path = seamless_path
-
-        if upscale and upscale != "none":
-            try:
-                if upscale is True or upscale == "topaz":
-                    final_path = run_topaz(seamless_path, tempdir)
-                elif upscale == "realesrgan":
-                    final_path = run_realesrgan(
-                        final_source,
-                        tempdir,
-                        scale=REALESRGAN_SCALE,
-                        model_path=REALESRGAN_MODEL,
-                        progress_cb=progress_cb
         )
-                else:
-                    raise ValueError(f"Unknown upscale option '{upscale}'")
-            except Exception as e:  # noqa: BLE001
-                print(f"Upscaling failed ({upscale}); keeping seamless image: {e}")
-
-        with Image.open(final_path) as final_img:
-            final_img.load()
-            save_png_with_prompt(final_img, output_path, prompt)
-        try:
-            with Image.open(output_path) as _im:
-                print(f"→ Final image written to {output_path} [{_im.size[0]}x{_im.size[1]}]")
-        except Exception:
-            print(f"→ Final image written to {output_path}")
-        return output_path
 
 
 def main():