luckiest-co 1.0.65 → 1.0.67

package/skills/photographer/references/video-generation.md

@@ -0,0 +1,98 @@
+# Phase 5: Video Ad Generation
+
+After the user is satisfied with the campaign image set, offer to create a short cinematic video ad.
+
+Ask:
+
+```markdown
+Generate a short video ad from the selected campaign images?
+
+- Yes, generate video
+- No, skip
+```
+
+If yes, let the user pick which campaign image should be used as the video's first frame:
+
+```markdown
+Which campaign image should become the video's first frame?
+
+- Shot 1: Hero
+- Shot 2: Detail
+- Shot 3: Environment
+- Other selected campaign image
+```
+
+## Video Motion Prompt
+
+Construct a motion-focused prompt that describes camera movement, lighting animation, environmental motion, and product fidelity.
+
+```markdown
+{camera movement} shot of the selected product campaign image.
+Preserve the product's exact silhouette, closure geometry, material, label placement, and luxury set design from the first frame.
+{lighting animation}. {environmental motion}.
+Cinematic beauty commercial, luxury fragrance advertisement, smooth motion, professional color grading, premium campaign finish.
+```
+
+## Camera Movement Options
+
+- **Slow dolly zoom in**: slow push-in that emphasizes product detail
+- **Gentle orbit around**: subtle camera arc that reveals dimensionality
+- **Slow pull-back reveal**: starts close, then reveals the full set
+- **Static with subtle parallax**: premium stillness with micro-depth movement
+- **Crane down**: downward reveal that feels dramatic and editorial
+
+## Lighting Animation
+
+- Golden light slowly shifting across the glass or packaging surface
+- Soft caustics dancing on nearby marble, water, or reflective surfaces
+- Light gradually transitioning from cool to warm
+- Rim light subtly tracing the bottle or container edge
+- Highlight line gliding across the cap, label, or curved body
+
+Match the lighting physics to the product material. Do not describe caustics or transmission for frosted, matte, ceramic, or opaque packaging.
+
+## Environmental Motion
+
+- Particles floating gently in the air
+- Petals drifting slowly through the frame
+- Water ripples expanding outward
+- Silk fabric undulating in a gentle breeze
+- Mist swirling around the base
+- Subtle shadow movement from leaves, curtains, or sculptural props
+
+## Generation Guidance
+
+Use whichever image-to-video tool this session has (for example a connected
+image-to-video MCP tool). If the user wants more than a single 3-5 second clip
+(a cut with several shots, captions, music, or end card), hand off to
+`luckiest-video-studio` with the chosen stills as references instead.
+
+Generate a short video from the selected campaign image as the first frame. Keep motion elegant and restrained; the product should remain stable, recognizable, and premium.
+
+Recommended defaults:
+
+| Setting | Recommendation |
+|---------|----------------|
+| Duration | 3-5 seconds |
+| Aspect ratio | Match the selected campaign image |
+| Variants | 1-2 options unless the user asks for more |
+| Motion intensity | Low to medium; avoid chaotic movement |
+
+## QA Before Presenting
+
+Review the video before showing it:
+
+| Check | What to verify |
+|-------|----------------|
+| Product identity | Product remains recognizable throughout |
+| Shape stability | Bottle/container does not melt, stretch, or morph |
+| Closure fidelity | Cap, pump, stopper, or lid remains consistent |
+| Label stability | Label does not drift unnaturally; minor AI text changes may be acceptable |
+| Motion quality | Camera movement is smooth and premium |
+| Material truth | Light behavior matches the product material |
+
+If the product shape changes significantly, regenerate with stronger fidelity language and calmer motion.
+
+## Presentation
+
+Show the final video result as playable media or a link, plus a one-sentence creative description. Keep logs and request payloads out of the main reply, but answer plainly if the user asks which model, settings, or prompt made it.

package/skills/photographer/scripts/compress_product_img.py

@@ -0,0 +1,84 @@
+#!/usr/bin/env python3
+"""
+Shrink local product photos so an agent can read them as images.
+
+Fits each image inside 1024x1024, converts it to JPEG (quality 85), then lowers
+quality and scale until the file is at most 145 KB. Reads local files only and
+makes no network calls. Needs Pillow (pip install pillow).
+
+Usage:
+  python3 compress_product_img.py --images photo.jpg [more.jpg ...] [--out-dir DIR]
+
+Prints one line per image:
+  [COMPRESSED] <out path> (1200KB -> 140KB, 1024x768)
+"""
+
+import argparse
+import io
+import os
+import sys
+import tempfile
+import uuid
+
+try:
+    from PIL import Image, ImageOps
+except ImportError:
+    sys.exit("[ERROR] Pillow is not installed. Run: pip install pillow")
+
+MAX_DIM = 1024
+MAX_BYTES = 148_480  # 145 KB, under the agent Read tool's image size limit
+
+
+def _open_rgb(raw: bytes) -> Image.Image:
+    img = Image.open(io.BytesIO(raw))
+    img = ImageOps.exif_transpose(img) or img
+    return img.convert("RGB") if img.mode != "RGB" else img
+
+
+def compress(raw: bytes) -> tuple[bytes, int, int]:
+    """Return (jpeg_bytes, width, height), at most MAX_BYTES when reachable."""
+    img = _open_rgb(raw)
+    img.thumbnail((MAX_DIM, MAX_DIM), Image.Resampling.LANCZOS)
+    base_w, base_h = img.size
+
+    quality, scale = 85, 1.0
+    while True:
+        size = (max(1, round(base_w * scale)), max(1, round(base_h * scale)))
+        frame = img if scale == 1.0 else img.resize(size, Image.Resampling.LANCZOS)
+        buf = io.BytesIO()
+        frame.save(buf, format="JPEG", quality=quality)
+        out = buf.getvalue()
+        if len(out) <= MAX_BYTES or (quality <= 30 and scale <= 0.3):
+            return out, frame.width, frame.height
+        if quality > 30:
+            quality -= 10
+        else:
+            scale -= 0.1
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Shrink product photos for agent vision")
+    parser.add_argument("--images", nargs="+", required=True, help="Local image paths")
+    parser.add_argument("--out-dir", default=tempfile.gettempdir(), help="Where to write JPEGs")
+    args = parser.parse_args()
+    os.makedirs(args.out_dir, exist_ok=True)
+
+    for src in args.images:
+        if not os.path.isfile(src):
+            print(f"[ERROR] Not a local file: {src}")
+            continue
+        with open(src, "rb") as f:
+            raw = f.read()
+        try:
+            data, w, h = compress(raw)
+        except Exception as e:  # bad or unsupported image
+            print(f"[ERROR] Could not compress {src}: {e}")
+            continue
+        out = os.path.join(args.out_dir, f"product_{uuid.uuid4().hex[:12]}.jpg")
+        with open(out, "wb") as f:
+            f.write(data)
+        print(f"[COMPRESSED] {out} ({len(raw) // 1024}KB -> {len(data) // 1024}KB, {w}x{h})")
+
+
+if __name__ == "__main__":
+    main()