Paint a watercolor portrait of Elon Musk from the attached reference photo, one brush stroke at a time, and record yourself doing it. Build the painting in code. Write your own brush engine (pigment, water, soft wet edges, granulation, glazing over dry layers), make the paper texture in code too, and lay down every mark as a separate stroke. Work in the order a portrait painter would: a faint sketch of the head and shoulders, pale skin washes, the shadows that shape the face, then hair, eyes and mouth, then the darkest accents. Mix every color on a virtual palette from a small set of base pigments. Use the photo the way a painter uses a reference: measure proportions and pick colors from it, but don't trace it or copy its pixels. Every mark has to be one of your own brush strokes. No image models, no other photos, no downloaded textures or brushes. Head and shoulders, a respectful likeness. Deliver: - painting.png: the finished portrait, at least 1600 px on the long side - timelapse.mp4: 1080x1080, 24 fps, 20 to 40 seconds, the portrait appearing stroke by stroke with no cuts - notes.md: the exact number of strokes, the pigments you started with and the colors you mixed, how long it took, the hardest moment, and how you used the photo Python and FFmpeg are installed and the machine has 4 CPU cores, so keep the whole job under 60 minutes. Don't keep frame images in the workspace: encode them into the MP4 and delete them (the saved workspace is limited to 500 files and 100 MB). Don't end your turn while a render is still running in the background.