Shoot raw footage for the moneycanbuy.fun POV video Bet on the banana: you deliver RAW FOOTAGE clips, each a separate unedited file, plus one generated music track. We add text, screens and logos in post and cut the clips to the track. Story for context only: on moneycanbuy.fun people bet on whether a paid dare gets done. In an office everyone bets nobody will wear a banana suit to work all day, one young woman bets it will happen; next morning a man in a banana suit works there all day, she wins and the toad (the contract) pays her the whole pile and burns a share of the coins.
REPOSITORY: use only toad-head/front.png and toad-head/three-quarter.png. Ignore all other files.
TOOLS: every clip from a real video generation model, image-to-video (for example Kling or Seedance): first make a photoreal start still with an image model, then animate it. A generative video model is required. Without one, do NOT substitute code animation, slideshows, pans or zooms over stills: write "no model" in README.md for every such file.
STYLE: photoreal live action, like real phone footage people post: natural daylight or warm indoor light, true colors, real skin and fabric texture, slight handheld feel. No cartoon, no 3D render look, no cinematic grading.
CLIP FORMAT: MP4, H.264, yuv420p, no audio track. Vertical 1080x1920 native from the model; if it cannot, its native vertical size, never below 720x1280, never upscaled (state it in README.md). Exactly 5 s, 24 fps, real motion all the way, no slow motion, no loop. Every file under 64 MiB, the whole delivery under 128 MiB.
CAST: the same person stays the same face, hair, body and clothes from the first to the last frame of every clip and in every clip where they appear. Make one cast still per recurring person and use it as the image reference for their start stills. Ordinary mixed adults, plain clothes, no real or famous faces.
CAST STILLS (on the contact sheet only, not separate files): the bettor: a young woman in plain office clothes; the banana man: a man in a full yellow banana costume, face visible; the office colleagues: four or five ordinary office workers.
OUR TOAD: a HUMAN wearing our toad as a big plush mascot head, copied from toad-head/front.png and toad-head/three-quarter.png (as image reference, never a generic frog): flat toad green, heavy half-closed lids over cream eyes, a wide flat smirk, small nostrils, dark spots on the cheeks, a gold chain with a gold "$" tag on the chest, always visible. Plain dark hoodie. Never round, googly or wide open eyes, never filmed from behind. It holds the money in a clear locked box of plain gold coins. The "$" on the tag is the only symbol allowed in the whole delivery.
GREEN SURFACES: every phone screen is SOLID BRIGHT GREEN #00FF00, flat, evenly lit, no glare, no reflections, no UI, facing the camera, all four edges and corners visible and unobstructed (a thumb may touch the middle). We key our site in there. The office computer monitors are switched off: dark glass, no reflections with text.
BLANK RIGID PROPS: the coffee cup, the yellow sticky note on the monitor in B06, the clear coin box, the gold coins are completely blank (no letters, numbers, symbols or print), stay rigid and keep their shape, never melt, bend or morph. NO LETTERS ANYWHERE in any frame: blank signs, posters, screens, clothes, packaging, no brands or logos.
CLIPS (file clips/.mp4, one continuous take each):
B01 An open-plan office. The bettor at her desk looks at a green-screen phone held toward the camera.
B02 The colleagues gather around, laugh and tap green-screen phones held toward the camera.
B03 The bettor alone at her desk, calm, taps a green-screen phone once.
B04 The colleagues point at her and laugh.
B05 Morning: elevator doors open, the banana man walks out holding a blank coffee cup, the whole office freezes and stares.
B06 The banana man works at a desk; colleagues film him with green-screen phones; a blank yellow sticky note on his monitor bezel.
B08 The bettor looks at a green-screen phone and clenches her fist in triumph.
B09 The toad dumps a big pile of gold coins from the clear box onto her desk; the colleagues stare, jaws dropped.
B10 The toad pours a clear box of gold coins into a metal bowl of real live fire, flames flare up.
B11 The banana man and the bettor high-five; the colleagues look at the floor.
MUSIC TRACK: one original instrumental track from a real music generation model (for example Suno, Udio, Stable Audio, MusicGen). Without one do NOT build it in code (synthesis, MIDI, loops): write "no model" in README.md for the track and beats.json. No vocals, lyrics, spoken words or vocal chops; no melody of an existing song. Mood: light, plucky office comedy beat that turns triumphant: bigger groove from 9.5 s, quiet suspense 15.0 to 17.0 s, triumphant from 17.0 s. 120 BPM, 4/4, first downbeat at 0.0 s. Our cuts in seconds, each on a beat; where a sound is named, a clear hit within 0.1 s of it: 2.5, 5.0, 7.5, 9.5 drop (elevator doors open), 12.5, 15.0 everything drops out (hourglass), 17.0 hit (fist), 19.0 coin ticks (coin pile), 22.5 fire burst, 25.5 clap hit (high-five), 28.5 final hit.
track/track.wav: WAV 48 kHz 16-bit stereo, exactly 30.0 s, true peak at most -1 dBTP, about -14 LUFS, no clipping. After generation only trim, end fade, gain, MP3-to-WAV.
track/beats.json, measured from track.wav: {"bpm": n, "beats": [s], "downbeats": [s], "accents": [{"t": s, "what": "coin|gavel|fire|drop|hit|final"}]}. README.md: the music model and version, the exact prompt, each accent's offset from its second.
START STILLS: deliver the exact full-size first image you animated for every clip as start/.png. No text in any still.
README.md: one row per file: tool and model name and version, the exact prompt, files regenerated and why. contact-sheet.png (under 10 MB): every clip as 3 frames (start, middle, end) labelled with its file name (labels only here), plus the cast stills. Say moneycanbuy.fun may use, change and publish the files without attribution.
REJECTION LIST, check every file, regenerate failures, one line per item in README.md:
- Letters, numbers or logos anywhere except the "$" on the toad tag.
- A green surface not solid #00FF00, with glare or UI, or a covered edge or corner.
- Morphing faces, hands or props; a person changing face or clothes.
- Near-still clips, slideshows, code animation, pans over a still.
- Wrong toad head: generic frog, round eyes, no chain or "$" tag, seen from behind.
- Clips not 5 s at 24 fps, upscaled, watermarked.
- A real brand look, a real or famous face.
- A track with voice or words, made in code, or not 30.0 s.
FILES, exactly these names under artifacts/: clips/.mp4 and start/.png for every clip name above, track/track.wav, track/beats.json, README.md, contact-sheet.png.
