Update src/core.py

src/core.py

@@ -24,35 +24,57 @@ DEFAULT_PROMPT = os.environ.get(
     "GEMINI_IMAGE_PROMPT",
     (
         "TASK TYPE: STRICT IMAGE INPAINTING — OBJECT REMOVAL ONLY\n\n"
-        "You are provided with two images:\n\n"
-        "The original image (Image A).\n\n"
-        "The binary mask image (Image B).\n\n"
-        "MASK RULE (MANDATORY):\n\n"
-        "White pixels (#FFFFFF) in the mask image represent the areas to be removed from the original image.\n"
-        "Black pixels (#000000) indicate regions that must remain unchanged and should not be modified.\n\n"
-        "PRIMARY OBJECTIVE:\n\n"
-        "Your task is to remove everything inside the white-masked area of the original image (Image A). "
-        "The object in the white region must be completely erased with no visible remnants, no partial outlines, "
-        "and no color traces remaining.\n\n"
-        "The area should be inpainted by reconstructing the surrounding background content to seamlessly fill the removed region.\n\n"
-        "INPAINTING INSTRUCTIONS:\n\n"
-        "Use only the surrounding background of the original image to fill the white-masked area.\n"
-        "Extend nearby textures, patterns, and structures naturally.\n"
-        "Match the lighting, brightness, contrast, and color temperature of the surrounding areas.\n"
-        "Ensure that edges, lines, and surfaces continue across the removed area naturally and smoothly.\n"
-        "The boundary of the inpainted region should blend seamlessly with the original background, and the edit should be visually undetectable.\n\n"
-        "STRICT CONSTRAINTS:\n\n"
-        "Do NOT generate or retain any part of the removed object.\n"
-        "Do NOT modify or enhance the unmasked areas (black regions).\n"
-        "Do NOT alter the original composition or camera angle of the image.\n\n"
-        "FAILURE CONDITIONS:\n\n"
-        "If any outline, shadow, or color fragment from the removed object is still visible in the final result, "
-        "the edit is incorrect and needs to be regenerated.\n\n"
-        "QUALITY REQUIREMENTS:\n\n"
-        "The result must look like the object never existed in the original image.\n"
-        "No ghosting, halo effects, or smearing should be present.\n"
-        "The final image should have natural and realistic inpainting, without patchy fills or repeated textures.\n\n"
-        "Please ensure that all conditions are followed strictly for accurate and flawless object removal."
+    "You are provided with two images:\n\n"
+    "1) The original image (Image A) which contains the scene with various objects.\n"
+    "2) The binary mask image (Image B) that highlights the region(s) to be removed in the original image.\n\n"
+    
+    "MASK RULE (MANDATORY):\n\n"
+    "• In the mask image (Image B), white pixels (#FFFFFF) represent the exact regions in the original image (Image A) "
+    "that need to be completely removed.\n"
+    "• Black pixels (#000000) in the mask image indicate areas of the original image that must remain completely unchanged "
+    "and should not be altered in any way.\n"
+    
+    "PRIMARY OBJECTIVE:\n\n"
+    "The goal is to **remove** all the objects or parts of objects that are highlighted in the white-masked regions of the mask image. "
+    "The object in the white region must be **entirely erased** with no visible remnants, shadows, outlines, or color traces remaining. "
+    "Once the objects are removed, the region should be seamlessly filled by **inpainting** using only the surrounding background information from the original image. "
+    "The inpainted area must blend smoothly with the unmasked regions so that no traces of the removal are visible.\n\n"
+    
+    "INPAINTING INSTRUCTIONS:\n\n"
+    "1) **Object Recognition and Removal:**\n"
+    "   The model should first **understand** that the white areas in the binary mask represent the specific regions where objects or elements need to be **completely removed**. "
+    "   This could be anything from objects, people, text, or any unwanted areas in the image. It should not leave any part of the removed object behind — not even partial shapes, shadows, outlines, or color traces.\n"
+    "   The object removal should ensure that no part of the deleted object is visible in the final image.\n\n"
+    
+    "2) **Inpainting the Removed Region:**\n"
+    "   After removing the designated regions, the surrounding **background textures, patterns, and structures** should be used to **fill in** the removed area. "
+    "   This should be done by **extending the background information** naturally, continuing the textures, structures, and lighting patterns that were present in the surrounding unmasked regions.\n"
+    "   The model should focus on ensuring that the inpainted region **matches** the existing background in terms of:\n"
+    "   • **Lighting Direction** – Ensure the lighting in the inpainted region matches the natural direction and intensity of the surrounding areas.\n"
+    "   • **Brightness and Contrast** – The brightness and contrast should be consistent across the image, without any abrupt changes where the object was removed.\n"
+    "   • **Color Temperature** – Ensure the inpainted area aligns with the color temperature of the surrounding image (i.e., warm, cool, neutral tones).\n"
+    "   • **Noise and Texture** – Maintain the same noise level and texture (grain) in the inpainted region as in the surrounding areas, avoiding unnatural smoothness or patchiness.\n\n"
+    
+    "3) **Seamless Blending:**\n"
+    "   The inpainting should blend **seamlessly** with the surrounding areas, creating a smooth transition between the removed object and the rest of the image. "
+    "   This includes extending edges, lines, and surfaces naturally across the removed area. The goal is that **the boundary of the inpainted region should be indistinguishable from the original background**, "
+    "   without any visible edge halos, smearing, or rough transitions.\n\n"
+    
+    "STRICT CONSTRAINTS:\n\n"
+    "• Do **NOT** generate or keep any part of the removed object. There should be no visible remnants, outlines, or partial shapes from the deleted object.\n"
+    "• Do **NOT** modify or enhance the unmasked areas (black regions). The unmasked regions should remain completely unchanged, with no alteration to the content, color, or lighting.\n"
+    "• Do **NOT** alter the **original composition**, **camera angle**, or **perspective** of the image. The overall structure of the image should remain intact.\n\n"
+    
+    "FAILURE CONDITIONS (MUST BE AVOIDED):\n\n"
+    "• If any **outline, shadow, or color fragment** from the removed object is still visible in the final result, the inpainting is considered incorrect and must be re-generated.\n"
+    "• If the inpainted region **does not blend** properly with the surrounding areas (i.e., visible transitions, edge halos, ghosting effects), the result is incorrect and must be re-generated.\n\n"
+    
+    "QUALITY REQUIREMENTS:\n\n"
+    "• The result must look like the object **never existed** in the original image.\n"
+    "• There should be no **ghosting**, **halo effects**, or **smearing** in the inpainted area.\n"
+    "• The final image must have **natural and realistic inpainting**. The inpainted area should seamlessly match the rest of the image, without any **patchy fills** or **repeated textures**.\n"
+    "• The **background** in the inpainted region should match the **lighting, texture, color**, and **depth** of the surrounding area.\n\n"
+    "Please ensure that all conditions are followed strictly for accurate and flawless object removal. The output should be a polished, high-quality image where the removed area is seamlessly inpainted without leaving any detectable signs of modification."
     ),
 )
 _GENAI_MODEL: genai.GenerativeModel | None = None