# FPV Route Desk > Write one-shot first-person FPV video prompts that hold together, and check the ones you > already have. Five lanes over one scene, plus a free browser-side checker that costs nothing. URL: https://fpv-route-desk.skillsafe.ai/ API: https://fpv-route-desk.skillsafe.ai/api.html Derived from the open-source MIT agent skill zhouwei713/fpv-immersive-video-prompting (https://github.com/zhouwei713/fpv-immersive-video-prompting): its POV identity table, variable-count table, numbered-stop and red-line path-control rules, video prompt structure, common pitfalls and verification checklist. Not affiliated with or endorsed by zhouwei713, Hermes Agent, ByteDance, Kuaishou, Runway or Google. ## What the free checker enforces All of it runs in the browser, on text you paste, with no account and no credits. - **The timeline must tile the whole shot.** Windows are written `0.00-2.50：what happens`. The first starts at zero, each begins where the previous ended, and the last ends at the stated length. A gap is a stretch of film with no instruction; an overlap gives the camera two instructions at once. This is a one-shot video, so there is no cut to hide either. - **Numbered stops must ascend** from 1, by one, with exactly as many stops as declared targets. A marker drawn on the first frame that the route never visits is still in the image. - **The POV must be able to do what it is asked to do.** Seven identities, each carrying the movements it cannot physically perform. - **The image pack must add up**: 1 numbered first frame + one reference per target + an optional clean frame, with the total stated as a number in the prompt. - **The avoid list must forbid the right things**, and it is read as an avoid list: a clause saying what the numbers mean is not a clause forbidding them from surviving into the film. - **The target count must fit the length**, against the documented bands. ## POV identities and what each one cannot do - `guest` - Guest / visitor. Camera at eye level. Good for: palace, mansion, apartment, party, museum, showroom. Cannot: a person on foot cannot fly, hover or bank; a person has no propellers; a person cannot pass through walls, glass or furniture. - `pet` - Pet cat or dog. Camera at low, near the floor. Good for: living room, cafe, cozy home, indoor close range. Cannot: a cat or dog cannot fly or hover; a pet cannot float over furniture - the upstream skill names this exact failure; a pet has no hands to gesture with; a pet's camera is near the floor, not at eye level. - `drone` - Invisible drone. Camera at variable altitude. Good for: outdoor landscape, city, battlefield, world route, canyon. Cannot: a drone has no walking bob - this is the single most common POV contradiction; an invisible drone shows no hands, sleeves or paws; a drone cannot pass through walls or terrain; the drone is the camera, so it can never appear in its own shot. - `vacuum` - Robot vacuum or small robot. Camera at very low, floor gliding. Good for: modern living room, apartment floor, under furniture. Cannot: a robot vacuum cannot jump, climb stairs or fly; a robot vacuum cannot get onto the sofa or a table; a floor robot's camera is not at eye level. - `bird` - Bird, insect or spirit. Camera at floating, unusual heights. Good for: open landscape, city skyline, fantasy world, mythic route. Cannot: a bird or spirit has no walking bob; a bird has no propellers; a bird has no wheels. - `vehicle` - Passenger, driver or mounted camera. Camera at fixed to the vehicle. Good for: car, train, boat, aircraft, racing line. Cannot: a camera mounted in a vehicle does not walk; a vehicle camera follows the vehicle's path, not a free flight path; a vehicle cannot pass through walls. - `cctv` - CCTV, flashlight holder or child. Camera at low or fixed, limited view. Good for: horror, mystery, security, night interior. Cannot: this POV does not fly; the whole point of this POV is hesitant, limited movement. ## The two route modes - `stops` - Numbered stop markers. Use when: close character interaction, interiors, crowded scenes, exact target counts. Requires: a numbered first frame; one reference image per target; a stop order 1..N. - `redline` - Red-line path control. Use when: world maps, fantasy journeys, city flythroughs, racing lines, canyon routes. Requires: an aerial route-control image; 4-6 geography segments; a drawn continuous route. Numbered stops are safer for close character work and interiors: an image model asked for a continuous red path through furniture and people usually draws a broken one, and the line then survives into the video. Red-line control is for world-scale routes, where the geometry of the path matters more than any single character. ## Target count by length - 5-8s: 1-3 targets - one clear approach, few actions. - 8-12s: 3-4 targets - short interactions, one final reveal. - 12-15s: 4-5 targets - ideal for sequential stops. - 15-25s: 5-8 targets - shorter beats and group transitions. - 25s and over: 8-24 targets - split into groups or zones. Above 6 targets, per-person dialogue is dropped for group beats, gestures and glances. Red-line routes segment by geography instead, 4-6 segments. ## The five lanes ### `pov` - POV and route plan Decide who or what the camera is before writing a word of prompt. FPV does not mean a human at eye level, and the identity you pick is what decides which movements the shot is allowed to contain. Sections: `Summary`, `POV Choice`, `Stop Plan`, `Movement Rules`, `Next Step`. Fields: `scene`*, `seconds`*, `pov`, `targets`, `intent`, `aspect`, `language`. ### `stops` - Numbered-stop prompt The prompt most interior and character scenes need. Numbered markers on the first frame give the camera its order, and the prompt defines the movement between them - which is safer than a drawn line that crosses the furniture. Sections: `Summary`, `Route Order`, `Timeline`, `Video Prompt`, `Checks`, `Next Step`. Fields: `scene`*, `pov`*, `targets`*, `seconds`*, `beats`, `draft`, `aspect`, `language`. ### `assets` - Image pack prompt One batch prompt that asks the image model for the whole pack at once, with the count stated explicitly. Ask vaguely for 'several images' and you get one contact sheet, which is useless as a first frame. Sections: `Summary`, `Image Plan`, `Asset Prompt`, `Count Check`, `Next Step`. Fields: `scene`*, `targets`*, `cleanframe`, `characters`, `style`, `aspect`, `language`. ### `route` - Red-line world route The other mode. For world maps, canyon runs, racing lines and city flythroughs, a drawn red route controls the camera geometry - and the prompt has to say, in as many words, that the line itself must never appear in the film. Sections: `Summary`, `Route Segments`, `Route Image Prompt`, `Video Prompt`, `Next Step`. Fields: `world`*, `pov`*, `seconds`*, `segments`, `draft`, `aspect`, `language`. ### `audit` - Prompt audit Paste an FPV prompt you already have. Every finding names the rule it breaks, and the lane returns a repaired prompt rather than a list of complaints. Sections: `Summary`, `Verdict`, `Findings`, `Repaired Prompt`, `Next Step`. Fields: `prompt`*, `pov`*, `mode`, `seconds`, `targets`, `focus`. ## What this app does not do It writes plans and prompts. It does not generate video or images, it calls no generation API on your behalf, and it flies nothing. The prompts are yours to paste into Seedance, Kling, Runway, Veo or anything else.