Rain Station Animated Short
By @zhu185178. Made with Claude Opus 5.5. Watch the original post.
Prompt
Make me a 30-second 2D animated previs short, "Rain Station", with benchmark-level image quality. Do not install HyperFrames/Remotion; write all the code yourself (Python + ffmpeg) to get it done. Story (24fps, 2.39:1, 4K): An elevated train platform on a rainy night. She wears a beige trench coat and holds a clear umbrella, standing on the near platform; on the opposite side, he stands under a streetlight, in a long dark gray overcoat and a scarf. 1. 0–6.4s Close-up of raindrops on the clear umbrella → pull back, the focus shifts across to reveal him (over-the-shoulder shot) 2. 6.4–9.4s Medium close-up of him: head down → looks up toward frame left 3. 9.4–12.4s Frontal close-up of her: blinks → recognizes him; a horn sounds in the distance 4. 12.4–17s Wide shot: the headlight glow arrives first, then a 5-car train passes; in the first gap between cars he is still there, the second is already empty 5. 17–18.8s Close-up of her, eyes welling with tears, bands of light from the train windows sweep across her face, her hair blown up by the wind 6. 18.8–22.4s The taillights recede, the opposite platform is empty; his lamp flickers a few times and goes out 7. 22.4–28.6s Her from behind, the focus pulls back from the empty platform to her; footsteps behind her, a black umbrella comes over her head, she turns and looks into the camera; a male voice: "Long time no see.", with subtitles 8. 28.6–30s Black screen, title "雨站 / RAIN STATION" Method: - Source art: use GPT Image 2.5 to draw the wide background, him, her, and the train (lead car / middle car / rear car). Put every character on a pure green #00FF00 background and key them out yourself (asking for a transparent background while giving a reference image gets you a fake checkerboard). Keep all poses of the same character in the same session, editing continuously with the base image as the reference; when changing poses, composite only the region that changed. - Image quality: use Real-ESRGAN x4plus anime 6B (the safetensors version on HF; write your own RRDBNet and run it on MPS) to upscale all the source art 4x; use Depth Anything V2 to estimate the depth of the wide background for per-pixel parallax. - Rendering: write your own 2.5D compositor. Place each layer at its true depth, and move a pinhole camera with a real focal length; do circular bokeh, focus pulls, 3D rain streaks, train motion blur, volumetric light, bloom and film halation, fine grain, and color grading. Use mesh deformation on the characters for breathing, wind in the hair tips and umbrella shake, and switch key poses Japanese-style, "on twos". 4K supersampling, cache every frame to disk, and split the work into 3 processes running in parallel by time range (16GB RAM). - Sound: real Mixkit sound effects (rain, rain on the umbrella, train, horn, footsteps); write a score to the story beats and render it with the Steinway piano and string samples built into Logic/GarageBand; voice the line with a Kokoro English male voice; set the sound distance per shot, loudness -16 LUFS. - At every step, render preview frames, tile them into a sheet to self-check, fix things, and only then output the film. Deliver a 4K master and a 1080p sharing version.