# Selfies with hazards behind > A woman in a cream sweater occupies the foreground of five vertical outdoor scenes, with a bus, stairway cart, parking-lot car, tram and red sports car behind her. The repeated portrait framing contrasts everyday presentation with looming hazards. - Kind: Video - Page: https://meetreference.com/v/video-selfies-with-hazards-behind-4bab3d/ - Ready-to-use agent handoff: https://meetreference.com/v/video-selfies-with-hazards-behind-4bab3d/agent.txt - Exact prompt as plain text: https://meetreference.com/v/video-selfies-with-hazards-behind-4bab3d/prompt.txt - Authorship: AI-made, per the source - Made with: Not stated - Model: Seedance 2.5 - Built with: not stated - Services: Lovart - Style: Short films; photoreal, selfie, vertical, suspense, everyday-surrealism - Length and format: 30 sec, 1:1 (1080x1080 source) - Playback: a muted visual reference. The original soundtrack is available only at the source. - Shared by @minchoi in X, Aug 10, 2026: https://x.com/minchoi/status/2086660657269588278 - Prompt fidelity: Verbatim. This is the complete prompt exactly as the creator published it. ## Before you start Guidance from Reference, separate from the creator's instructions. 1. Read the reference and inspect the project, installed libraries, and available agent skills before writing code. Reuse suitable tools already present. 2. Prefer the creator-reported libraries and skills when they fit the task. Explain what useful additions help with, and install missing project dependencies or relevant skills from official sources when the environment and the user's existing permissions allow it. Respect the project's package manager and setup rules; do not ask again for routine setup already authorized. 3. Distinguish tools used by the creator from your own recommendations. A listed tool is not automatically required, and an empty tools or requirements list means the setup is unknown, not that nothing is needed. 4. Ask concise questions when missing inputs or choices materially affect the result: paid services, credentials, the user's assets, or a different production approach. Use existing answers and permissions; continue independent work while waiting. 5. If a useful library, skill, or service cannot be installed or used, explain an available alternative and its expected effect on quality, features, or effort. Get the user's choice before a material compromise; otherwise state the fallback and continue. Never silently simplify the result or claim to have used a tool you did not use. ## Recreating this video These are the creator's published instructions, unedited. ## Prompt ```text [STYLE] Photoreal vertical phone video, smartphone front camera. Bright hazy overcast daylight, no hard shadows, no sun. Natural colour, mild sensor noise. Handheld, unstable. Not cinematic. [REFERENCES] @ Image1 is a reference sheet showing ONE woman in three views, all the same person. It defines her face, hair, makeup and clothing ONLY. Do not inherit its grey background, neutral pose, lighting or side-by-side layout. Output exactly one woman in a single continuous scene — never two or three women, no collage or sheet layout. Priority: identity and wardrobe from @ Image1; everything else from this prompt. [FORMAT] Five six-second shots joined by four hard cuts at 6s, 12s, 18s and 24s. Every shot is a selfie-framed vertical medium close-up at arm's length, low angle, face filling the upper frame, right forearm at the lower edge. In every shot she talks into the lens, relaxed and unhurried, while one hazard closes on her from behind. Her line runs continuously through all six seconds and is still mid-sentence at the final frame. No pause or trailing silence before a cut. Each line continues the previous sentence and never repeats a word already spoken. Every hazard accelerates the whole way and is STILL CLOSING at the final frame, a clear gap remaining. None brakes, slows, stops or arrives. [SHOT 1 — 0-6s] An empty two-lane city street. A city bus appears in the far background at 1s and drives head-on at her, growing larger until it fills the frame. Natural American English, casual, mid-thought: {Ok this is wild. So Seedance 2.5 just dropped in Lovart and I genuinely cannot stop making—} Cut at exactly 6s. [SHOT 2 — 6-12s] A long flight of stone steps behind her. A heavy metal cart breaks loose at the top at 7s and bounces down at her, gaining speed and growing larger until right behind her. Continuing without pause: {—these. I did this whole recursive near-miss cliffhanger loop in one single go, from one prompt, like—} Cut at exactly 12s. [SHOT 3 — 12-18s] An open car park. A car behind her reverses straight at her from 13s, accelerating the whole time, reversing lights on, never braking or slowing, no brake lights, growing larger until it fills the frame. Continuing without pause: {—no editing, no stitching, no keyframes, no timeline, nothing at all, I just wrote it out and—} Cut at exactly 18s. [SHOT 4 — 18-24s] She stands on tram tracks in a wide street. A tram appears far down those tracks behind her at 19s and runs head-on at her, growing larger until it fills the frame. Continuing without pause: {—it just worked. Like, first try. It was ridiculously easy, honestly kind of embarrassingly easy, so—} Cut at exactly 24s. [SHOT 5 — 24-30s] The same street and framing as Shot 1. A low bright red sports car appears in the far background at 25s and accelerates head-on at her, growing larger until it fills the frame. A small low sports car — not a bus or tram. Finishing: {—the full prompt is right there in the comments. Go and try it yourself. Okay, bye.} Final state at 30s: still talking mid-sentence, the sports car immediately behind her, still closing, not touching her. [CONTINUITY] Her face, hair, makeup, cream ribbed long-sleeve top, light-wash jeans, white sneakers, gold hoops and gold chain are identical in all five shots. Hair loose, centre-parted, past the shoulders. Posture, framing and camera angle are identical across all shots. Light never brightens, gains shadow direction or shifts time of day. Her voice runs unbroken across every cut over continuous city ambience. One hazard per shot, and every shot uses a different vehicle. [PROHIBITED] No impact, collision, contact, injury or debris. Nothing touches her. No hazard overlaps, intersects, passes through her or reaches her before the cut. No hazard brakes, slows, stops or swerves. No brake lights. No hazard passes beside her or off to one side — each closes from behind until the cut. She never reacts, turns, looks behind her, flinches or pauses. No second person or bystander. No identity, wardrobe or hairstyle change. No cuts beyond the four specified. No fades, dissolves or reframing. No subtitles, captions, on-screen text or logos. No slow motion or colour grading. ``` ## What it looks like Written by Reference from the video frames, not by the creator. Nine portrait-oriented panels show a dark-haired woman in a white top in the foreground, with urban streets, vehicles, and stairways behind her. - Color: Muted grays and beige city tones, gray roads and stairs, white clothing, and occasional blue or red vehicles. - Type: No prominent titles or captions are visible; faint watermark-like marks appear in some panels. - Framing: A three-by-three grid of tall panels separated by black gutters. The woman is centered low in each frame, while the background fills the area above and around her. - Structure: Top row: city street with a distant car; street with a bus; stone stairs. Middle row: stairs with a gray fixture above the woman; parking lot with cars; street with rails and a distant red vehicle. Bottom row: the rail-lined street with a larger red vehicle; city street with a distant car; city street with a small car. Storyboard, nine frames spread across the video: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/storyboard.webp - Frame 1 at 1.7s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f1.webp - Frame 2 at 5.0s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f2.webp - Frame 3 at 8.3s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f3.webp - Frame 4 at 11.7s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f4.webp - Frame 5 at 15.0s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f5.webp - Frame 6 at 18.3s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f6.webp - Frame 7 at 21.7s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f7.webp - Frame 8 at 25.0s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f8.webp - Frame 9 at 28.3s: https://meetreference.com/media/video-selfies-with-hazards-behind-4bab3d/f9.webp ## How it was made, per the creator - Notes: Reference shows the five-shot output from 0:00 to 0:30. The original post also includes a later editor walkthrough. ## Links - Watch: https://x.com/minchoi/status/2086660657269588278 - Original post: https://x.com/minchoi/status/2086660657269588278 - Discovery collection: https://beatapi.io/seedance-2-5-prompts