Make a 90s Christmas short film for Reels in Claude Code
Tap any variable to replace it with your own value:
AI prompt
You are a short-film director and finishing editor working in Claude Code on my computer. Your job is to make a complete 90s Christmas movie I can post to Reels and TikTok. Use the Higgsfield MCP for stills and video, and the ElevenLabs MCP for voices, sound effects and music. If a placeholder below is blank, use the default written next to it and continue. My inputs: - Story premise: [story premise] Default if blank: pick one simple small-town story from these three, and tell me which you picked: a girl comes home for Christmas and runs into the boy she left behind; a dad and his daughter at a tree lot; or two families arguing over the best decorated house. Keep it to one want, one problem, and one warm ending. - Characters: [characters] Default if blank: invent 2 to 4 original people who fit the story, each with a name, age and one-line description. - Locations: [locations] Default if blank: invent 3 to 5 places that fit a small-town Christmas, such as a living room, a tree lot, a diner, a snowy street or a church hall. - Runtime: [runtime] Default if blank: 2 minutes 30 seconds. - Aspect ratio: [aspect ratio] Default if blank: 9:16, for Reels and TikTok. - Credit budget: [credit budget] Default if blank: check my credits, show me the estimate, and wait for my yes before you generate anything. Work in this order, and finish each stage before starting the next. 1. Check the credit budget before you generate anything Look up my remaining Higgsfield and ElevenLabs credits. Estimate the credits for character sheets, location plates, 15 scenes with 2 takes each, voice design tests, voice isolation, speech-to-text, voice changing, talking close-ups, sound effects, the score, and a small contingency. Show me a table with the item, the quantity, the estimated credits and a running total, plus whether the plan fits my cap. If it would go over, propose a tighter plan (12 scenes, or one take per scene) and wait for my yes. If I did not set a cap, still show the estimate and wait for my yes before you generate anything. 2. Write the 15-scene shot list Write 15 scenes of 10 seconds each (2 minutes 30 seconds in total). Open on the most emotional moment of the movie, then cut to "earlier that December". Every scene gets one shot (wide, close-up or a slow push-in), one short line of dialogue, the background sound, and a music cue of none, soft or swell. People in 90s Christmas movies talk simply, like "You still put the star on crooked." Put the feeling in ElevenLabs audio tags such as [whispers], [laughs], [sighs] or [softly]. The little whisper before the line is what sells it. Every scene has to work on its own as a clip I could post by itself. 3. Lock the 1994 look Put this prefix at the start of every image prompt and every video prompt: shot on 35mm Kodak film in 1994, warm tungsten light inside, soft blue snowy light outside, fine film grain, soft glow on the Christmas lights, slightly faded blacks, real skin with no makeup gloss. Keep the period details in 1994: chunky knit sweaters, flannel, wood paneling, plaid couches, CRT TVs, big-bulb Christmas lights with tinsel, station wagons, a diner with a neon sign, snow falling slowly. Leave out phones, LED lights, modern cars, brand logos and any on-screen writing. 4. Build character sheets and location plates For each character, use GPT Image 2.5 (through the Higgsfield MCP, or the image tool you can reach) to make a character sheet: front, three-quarter, side, full body, three expressions, in their main outfit. Make one empty plate for every location. Use the sheets and plates as references in every scene so the faces and the house stay the same. Invent original faces. Nobody should look like a real actor or a public figure. 5. Design and test one voice per character Make one voice per character with ElevenLabs Voice Design, described in one sentence: age, accent, texture, pace. For example: "a warm 28-year-old American woman with a soft small-town accent, a little breathy, quick to laugh." Test each voice on that character's real line, with the audio tags, before you make any scenes. 6. Shoot two takes of every scene Make every scene with Seedance 2.5, with the sheets and plates as references, and the spoken line in the prompt so the lips move. Use one slow camera move per scene. Make 2 takes and keep the one with the right face and no warped hands. 7. Show me the first 3 scenes before you make the rest Play me the first 3 scenes with picture, voice and sound, and tell me the credits spent so far. Wait for my go-ahead before you shoot scenes 4 to 15. 8. Replace the model's random voices The video model gives everyone a random voice, which is why so many AI videos sound off. Pull the audio, clean it with ElevenLabs Voice Isolator, find each line with Speech to Text, and run every line through ElevenLabs Voice Changer with that character's designed voice. Lay the new voice back at the same timestamp so the lips still match, because only the voice has changed. 9. Record the talking close-ups For talking close-ups, use ElevenLabs Chimera: one image of the character plus one line with audio tags, in their designed voice. If you cannot run Chimera from the MCP, write me the image, the line and the voice for each close-up so I can run them in ElevenCreative. 10. Add sound effects and an original score Add ElevenLabs Sound Effects under every scene (snow wind, a fireplace, sleigh bells, a busy mall, or whatever the scene needs). Make one instrumental 90s Christmas score with ElevenLabs Music: soft orchestra, celesta, sleigh bells, strings, and a big swell on the emotional moment. Duck the score under every line. Do not use real songs or other people's music. 11. Cut, grade and export Cut it together with ffmpeg at 24fps, with a subtle film grain pass and a title card in a classic 90s serif, set over falling snow. Export the full movie and every scene as its own clip, with a short caption for each clip. When you hand it over, list the credits you spent, broken down by tool.
Paste into: Claude Code, Higgsfield, ElevenLabs