A detailed prompt for creating a 30-second 2D animated short film titled 'Rain Station' using Python, ffmpeg, and GPT Image 2.5 for assets, rendered via custom code by Claude Opus 5.5.
Prompt
Help me create a 30-second 2D animated storyboard preview short film titled 'Rain Station', with benchmark-level quality. Do not use HyperFrames/Remotion; write all code yourself (Python + ffmpeg).
Story (24fps, 2.39:1, 4K): A rainy night at an elevated station platform. She wears a beige trench coat and holds a transparent umbrella, standing on the near-side platform; he stands under a streetlight on the opposite side, wearing a dark grey long coat and a scarf.
1. 0–6.4s Close-up of raindrops on the transparent umbrella → Pull back, focus shifts to him appearing on the opposite side (over-the-shoulder shot)
2. 6.4–9.4s Medium close-up of him: Head down → Looks up towards the left of the frame
3. 9.4–12.4s Frontal close-up of her: Blinks → Recognizes him; distant horn sound
4. 12.4–17s Long shot: Headlights arrive first, a 5-car train passes; he is still visible in the gap of the first car, but the second is empty
5. 17–18.8s Close-up of her teary eyes, light bands from windows sweep across her face, hair blown by wind
6. 18.8–22.4s Tail lights recede, opposite platform is empty; his streetlight flickers a few times then goes out
7. 22.4–28.6s Her back view, focus pulls from empty platform back to her; footsteps behind, a black umbrella covers her head, she turns to look at the camera; male voice says 'Long time no see.', with subtitles
8. 28.6–30s Black screen title 'Rain Station / RAIN STATION'
Method:
- Original Art: Use GPT Image 2.5 to draw the background, him, her, and the train (head/middle/tail cars). Characters must have pure green #00FF00 backgrounds for self-matting (using reference images with transparent backgrounds may result in fake checkerboards). Keep various poses of the same character in one session, editing continuously based on a base image; only composite changed areas when switching poses.
- Quality: Use Real-ESRGAN x4plus anime 6B (safetensors version from HF, write your own RRDBNet, run on MPS) to upscale all original art by 4x; use Depth Anything V2 to estimate background depth for per-pixel parallax.
- Rendering: Write your own 2.5D compositor. Place layers according to real depth, move pinhole camera according to real focal length; implement circular bokeh, rack focus, 3D rain streaks, train motion blur, volumetric light, bloom/glow/film halo, fine grain, and color grading. Use mesh deformation for characters to simulate breathing, wind-blown hair, and umbrella shake, switching key poses in Japanese 'one-frame-two' style. 4K supersampling, cache frames to disk, parallelize rendering into 3 processes by time segment (16GB memory).
- Sound: Mixkit real SFX (rain, umbrella rain, train, horn, footsteps); compose music according to plot nodes, render with Steinway piano and string samples from Logic/GarageBand; dialogue uses Kokoro Chinese male voice; apply distance-based audio panning/volume, loudness -16 LUFS.
- Render preview frames at each step, stitch them together for self-checking, fix issues before final output. Deliver 4K master and 1080p share version.
How to use this prompt
1
Copy the full prompt above.
2
Open a platform that supports Claude Opus 5.5, such as YouMind, and paste the prompt in.
3
Swap the subject, style, or details to fit your idea, then generate.
This is a free AI prompt from YouMind's prompt library. Explore thousands more video prompts, all free to copy and adapt.
Let AI search through tens of thousands of prompts. Filter by model, time range, keywords, and sort by engagement — views, bookmarks, reposts and more.
Turn any photo into a detailed AI image prompt. The free image to prompt converter analyzes composition, style, and lighting so you can recreate any look in seconds.
