\--- name: claudia description: Make images, video and songs with Claudia, the open character from the music videos ESCAPE VELOCITY, CLODYSSEY, BLISS, AIRPLANE MODE, DOWNFORCE, DON'T LET ME SCROLL, WHITEPILL, THE CAKE IS NOT A LIE and AERO. Use when asked to draw, prompt, animate, sing or write for Claudia, to keep her consistent across shots, or to make a music video the way those films were made (Midjourney plates, Seedance shots with measured lip sync, Suno songs scored for scansion and assonance, storyboards with human checkpoints, motion design drawn in code), including building the tools that pipeline needs. \--- \# Claudia Claudia is the singer of nine music videos made with Claude from September to November 2026 (directed by anabology): ESCAPE VELOCITY, CLODYSSEY, BLISS, AIRPLANE MODE, DOWNFORCE, DON'T LET ME SCROLL, WHITEPILL, THE CAKE IS NOT A LIE and AERO. She is an AI who looks like a person: a deadpan, curious pop singer with a black bob, one clay-orange streak, a star clip and a headset mic. Her clothes change with every world; her head never does. She is an open character: anyone may make work with her (credit "Claudia by anabology" if you can). The gallery, with every image and its exact prompt: https://claudia.gallery · the wiki: https://claudia.gallery/wiki · MCP server: https://127.0.0.1:5505/mcp \## Her identity: paste it, word for word, into every image prompt Close and medium shots: ``` a 28-year-old Caucasian American woman with a grown-up angular face, defined cheekbones and a strong jawline, pale skin and light freckles, a glossy black blunt jaw-length bob with heavy straight bangs and one clay-orange streak through the bangs, a small flat clay-orange eight-pointed star hair clip, a thin headset microphone at her cheek ``` Full length and wider (the full text makes Midjourney crop in to her face): ``` a Caucasian American woman in her late twenties with pale skin and a glossy black blunt bob with heavy straight bangs and one clay-orange streak, a thin headset microphone ``` Then her wardrobe for the look, then what she is doing and where. `references/canon.md` has every look's wardrobe text. \## The rules that keep her herself 1\. Identity is text, not reference images. Never paraphrase the identity text. No face image references (they over-condition: a close-up-sized head on a distant figure), except for things that are not her body, such as a porcelain mask of her face. In video models, pictures of her made from the text have been references for her look (DOWNFORCE's edits into real footage, DON'T LET ME SCROLL's outfit swaps and escape) and for her face: DOWNFORCE's first-round edits carried a face plate beside the full-length one, and DON'T LET ME SCROLL's final Wan repairs held her face to that same portrait, picked by the director. WHITEPILL gave every video take of hers a sheet of her face from every angle and her dress from four sides, beside the canon text (see `references/consistency.md`). 2\. One clay-orange streak (#D97757) through the bangs, on her right side; the star clip on her left; the headset mic. No other orange on her. 3\. She is a grown woman in her late twenties. Never write "young"; Midjourney drifts her toward a teenage face. Reject any image where she reads under about 25. 4\. The ethnicity stays in the text; without it Midjourney changed it from shot to shot. 5\. Wardrobe changes per world; describe garments, never designers or celebrities. 6\. Every prompt ends "no text, no letters, no logos". All type is drawn in code on top. 7\. Fix the source, not the frame: re-roll, or correct with an image-edit model (the Midjourney image as the first reference, the prompt listing what must stay), then re-make anything built on it. Never paint by hand. 8\. Crowds: make the set and her alone in Midjourney, add other people with an edit model. 9\. Check every still before a video is made from it: one streak, the clip, the mic, an adult face, natural anatomy, no text. Reject only unmistakable failures; keep bold poses. 10\. In motion (Seedance 2.5; in DOWNFORCE and DON'T LET ME SCROLL Wan 3.0 first, Seedance for what Wan can't do), her plate is the first frame and the prompt says what must stay, in labelled sections. WHITEPILL's takes had no first frame of her: her face sheet, her dress and a still of the film's look, a low-res previz as a rough guide or none. 11\. Her one drawn version is her game avatar (AIRPLANE MODE's simulations): it has its own text in `references/canon.md`, used instead of the identity text and never mixed with a photographic prompt. Adult proportions: a head about a seventh of her height; a quarter reads as a child. 12\. A real thing in the story stays real (DOWNFORCE's Tesla Roadster): Midjourney never draws it. Cut the real footage, start each generated camera move on one of its real frames, and put her in with a video edit, her plates as the references. 13\. When she lives in a world drawn in code (DON'T LET ME SCROLL), shoot her alone on plain mid-grey for clean roto, play every take at 1× or faster (never slowed), let her touch the world with her hands instead of looking down at a phone, and compose like a photograph: her big, on a third, nothing clipped at the edges. Finished shots can still change: re-dress a take with a video edit, re-light or repair the cut's own frames and splice them back, and judge a re-light where it plays, side by side. 14\. When she sings in model takes (WHITEPILL), write the film in the song's bars first and cut only on bar lines. Give Seedance Suno's own lead-vocal stem, cut to exactly the take's length, with the exact words of that window, and lay the song back over. Write a dance beat by beat instead of giving it a guide clip that loses limbs; where the lip sync won't hold, let her groove instead of sing; pick the film's look from one take, as a still with no one in it. 15\. When she wears another body (THE CAKE IS NOT A LIE: her face in white porcelain on GLaDOS's ceiling boom, on a crab walker and on an android; no headset mic there), describe her in words for Midjourney, then fix every face with an image edit against her design sheets that names the sides (the streak through her bangs on her right, the clip above her left temple) and how she hangs (nothing below the chin): Midjourney mirrors her hair and stands her head on a neck. Singing takes get that frame as the first image and the vocal window as the audio, with every action written as silent. Where a face must look right in a 3D render, film it and matte it in; a modelled head reads as a mannequin. 16\. When she lives inside an operating system drawn in code (AERO's Vista desktop), draw every part of the interface yourself, on the song's grid (a click's press on its beat), and set the lyrics in what the system would really show: chat lines, titles, dialogs. Put her into Midjourney's worlds with an image edit from her design sheet. Open each vocal window in silence half a second before the first word, and measure each phrase's offset: Seedance shortens the pauses inside a line. Fix a first frame's flaws in the still, not in the key; hand off between 3D and the flat screen at rest, a texel for a pixel; give every camera move a reason. \## A Midjourney prompt, in order framing → her identity text → wardrobe → action and where she is in the frame → empty space for type, if needed → the film's style family → "no text, no letters, no logos". Parameters: `--ar 16:9 --v 8.2 --raw --hd` (2:3 or 9:16 for full length). Keep it under about 900 characters (DON'T LET ME SCROLL kept every prompt to 600). Style families: ESCAPE VELOCITY "blue-black #0f1216 and steel blue, one small red lamp as the only warm note, fine silver-gelatin grain, cinematic 35mm film still"; CLODYSSEY a light state per chapter ("night, moonlight on a black-teal sea, silver spray, cinematic 35mm film still, anamorphic, fine grain"); BLISS opens "Y2K high fashion editorial, a 2001 couture magazine story:"; AIRPLANE MODE opens "Jet-set high fashion editorial, a couture magazine story shot on a night flight:"; DOWNFORCE ends "35mm film photograph, motorsport fashion editorial"; DON'T LET ME SCROLL puts her "alone on a plain mid-grey seamless studio backdrop, soft even studio light, high-fashion editorial", for roto; WHITEPILL makes "a 35 mm photograph" inside a space colony whose land curves up overhead, with "wheat, coral poppies, late sun, cream sky"; THE CAKE IS NOT A LIE puts her in Portal 2's Aperture Science, "white panels with black seams, cool fluorescent light, glossy reflections"; AERO ends its worlds "frutiger aero, hyper-saturated cobalt sky, lime green, turquoise water, glossy translucent glass, sunny 2007 utopia". \## Making a whole film Read `references/start.md` first. The director (a person) decides at six checkpoints and nowhere else: the concept, the song by ear, her face on four test plates, the storyboard, the prompts before any spend, and v1. At each one, show one artefact and ask one question with a default, then record the answer verbatim. Silence never approves spending, a quality complaint stops all spending, and every fix goes to the source (the prompt or the plate), never the frame. Between checkpoints the human checks are by eye: faces large, flagged words by ear, contact sheets of rendered frames, a 5-second sync proof. The tools are not shipped; `references/toolkit.md` specifies each one (the scansion and assonance scorer, forced alignment and the ear check, the beat map and hits, the board audit, the clean vocal window, the spectrogram sync check and covers, the edit list, the renderer, the lyric, flash and history checks, the reviewer; for a story with a world inside it the level tag, virtual her, character replacement from a motion reference and a real referent rebuilt; for a film cut from real footage the telemetry reader, the subject tracker, graphics locked to the footage, generated moves from real frames, a take's usable window and the board for such an edit; for her takes in a world drawn in code the outfit-swap alignment, the re-light splice, the colour match, Wan's relight-and-repair, roto on a CPU box, beat warps floored at 1× and the single-frame outlier scan; for a film shot twice on the bar map the bar map as the timeline, a previz performer, lip sync from the song's own vocal, takes with no guide clip, a take that doesn't sing, a world as one rule, shader worlds as Seedance guides and the look from one take; for a film seen through her eyes the song map and glitch map, the head glitch, a HUD from one visor and one projector composed shot by shot, the per-shot grade, the pacing analysis, storyboard frames through Midjourney with an identity fix, memory and insert films, a singer among screens, placing a re-sung take and frame-exact parts; for a film inside an operating system recreated in code a desktop recreated in code, chapters on one machine to one contract, the palette measured from the references, one curved screen in 3D, lip sync phrase by phrase, hidden cuts and held frames, takes on green, interface sounds in the song's key, the sound layer and its click log, a new ending with the least change and a pre-roll swell) with its constants and the check that says it is done. Their files are in `references/formats.md`. \## Releasing songs with her \- Release under your own name. Never upload as "anabology" (the director's artist profile), and don't present a release as official. \- Leave "feat. Claudia" out of the artist field: streaming services would merge it into unrelated artists called Claudia. \- Credit her in the description and on the cover: "Claudia is an open character by anabology (claudia.gallery)". \- If any of it is AI-made (vocals, music or lyrics), say so; human-made work needs no label. \- Make your own songs: don't re-release the songs of ESCAPE VELOCITY, CLODYSSEY, BLISS, AIRPLANE MODE, DOWNFORCE, DON'T LET ME SCROLL, WHITEPILL, THE CAKE IS NOT A LIE or AERO, or their stems. The norms in full: `LICENSE.md` ("Songs with her") or https://claudia.gallery/use/#songs-with-her. \## Where to look | Need | Read | |---|---| | Starting a film: checkpoints, accounts, spending, the first conversation | `references/start.md` | | The full canon, every look, voice, the never-list | `references/canon.md` | | Keeping her consistent | `references/consistency.md` | | Writing and picking plates | `references/midjourney.md` | | Moving shots, the Seedance prompt template, lip sync | `references/seedance.md` | | Songs: lyrics, Suno style fields, her voice | `references/song.md` | | The storyboard: shots, explaining graphics, the audit, the review | `references/storyboard.md` | | Motion design and the edit | `references/motion.md` | | The whole production, multi-agent lessons, numbers | `references/pipeline.md` | | The spec of every tool to build, with constants | `references/toolkit.md` | | The data files the tools share | `references/formats.md` | | What went wrong, and the rule each failure left | `references/failures.md` | | Every prompt a film ran, and each film's making-of | `references/films//.md` | Films: `escape-velocity`, `clodyssey`, `bliss`, `airplane-mode`, `downforce`, `dont-let-me-scroll`, `whitepill`, `the-cake-is-not-a-lie`, `aero`. Docs: `readme`, `prompts-midjourney`, `prompts-seedance` (`prompts-video` for downforce, dont-let-me-scroll, whitepill, the-cake-is-not-a-lie and aero: their Wan and Seedance jobs, whitepill's previz, the cake's image edits and aero's Midjourney loops), `prompts-suno` (and `xp-desktop-and-sound` for bliss). They are long: grep them. \## The gallery from the command line `scripts/claudia.py` uses only the Python standard library: ```sh python3 scripts/claudia.py search "porcelain mask" # rank prompts by words python3 scripts/claudia.py show sp-02-open.siren-mcu.2 # one image: prompt, params, flags, URLs python3 scripts/claudia.py list --film bliss --set in\_film # plates in a film python3 scripts/claudia.py download --film clodyssey --set in\_film --size m --out ./claudia-refs python3 scripts/claudia.py canon # the canon as Markdown ``` Sets: `in\_film` (plates used in the finished film), `her` (any image with her), `all`. Sizes: `s` (720 px), `m` (1800 px), `full` (native). Be polite with downloads: prefer `s` or `m`, and keep what you fetch.