You approved the Create plan ("approve all", "do all"). All four parts are built, tested and deployed. Create is now the studio's first page. Every video shows how it is being made, step by step by its rulebook. Everything we made lands in Generations. Claude plans each brief with the format's rules and the game's DNA, free on the team's subscription, and nothing is spent until you press Generate.
Tested on a copy of the live data (models stood in, so nothing was paid). The brief: "A nurse coming off a night shift stops at the street booth on her way home; she asks if the snake game on the screen is real, the host shows her the cash-out and she tries it on her phone", with Street Booth and Diamond Snake for the UK.
13 decisions, each with its rules. They arrived live (2, then 7, 9, 13). Examples:
Cast & set: "Passer-by becomes the brief's nurse: mid-thirties, plain navy scrubs under a charcoal wool coat, no badges, ID lanyard or health-service logo" (R1, R7).
Shot 3: "The brief asks the host to show her the cash-out on a screen, but a model may not draw the game. So shot 3 is a close-up of him holding his phone out…" (S1, R4, S4).
Lines: "Wait, is Diamond Snake on that screen real?" · "One press, one step, then cash out." · "Eat the diamonds, cash out before it crashes." · "Go on then. I'll have a go."
A picture: "A poster: the gem snake coiled around a glowing cash-out button". Claude kept to the rules: "the CASHOUT button and its amounts are the game's own screen, which a model never draws, so the button is plain gold with no words or numbers".
A whole video's plan took about 6 minutes; a picture's about 2. The page says so.
From Higgsfield lists your Higgsfield plan's own generations: 59 today, 23 clips and 36 pictures, with thumbnails. Each can go into a video or become a format's first example. They are your own, so they are not shown here.
| Planned | Built |
|---|---|
| Its motion presets as a Moves row | A Moves row on every shot: dolly in, dolly out, orbit, whip pan, crash zoom, handheld follow, crane up, tilt reveal, locked off. Higgsfield's command line has no camera presets for the models we run: Seedance 2.5 and Kling 3.0 take words and a start frame, and its presets are 3D-character animations. So the move is said in words that every model reads. |
| Soul ID characters as cast | Listed in Settings, and offered on every person's cast picture once one exists. Your plan has none trained yet. Train one on higgsfield.ai (Soul ID, 5–20 photos) and it appears within two minutes. |
| Your past generations in Generations | Done: the "From Higgsfield" tab (59 today). |
| Before | After | |
|---|---|---|
| Where you start | "What's next", then New video (recipe, inputs) | Create: one brief; Claude plans it with the rulebook |
| Seeing it made | Cards filling in on the storyboard | Being made: every step with its rule, model, price and outputs, live |
| What we made | Scattered across videos | Generations, with how each was made and what to do next |