Skills-only ChatGPT plugin

Flow Story Maker

Story-to-video prompt packs for Google Flow / Veo 3.1. A rough idea in, a continuity-safe shooting package out.

What it does

The run goes in order, one step feeding the next:

01 · interviewA short two-form interview — setup first, production answers second. Nothing to configure; it works inside the conversation.

02 · overviewA text-only story overview checkpoint: the arc, the setup, and the clip plan in a few scannable lines. Approve it or say what to change.

03 · clip mapA timestamped clip map — one line per clip with its beat, so you know what the run costs before a single prompt is written.

04 · imageOne reference image for the whole run — character, look, and style anchor. Never a batch.

05 · promptsA copy-ready prompt per clip, each carrying its timestamp header, its audio direction, and a binding to the reference image's look so separately generated clips read as one video.

Two specialised paths

riddleTagalog horror-riddle stories. The riddle is spoken by a supernatural Teller and never appears on screen; the answer is sealed and never hinted at. The story always shows a visible consequence of the riddle going unanswered.

mukbangMukbang ASMR packages from a food photo. Silent of speech — no talking, no narration, no voiceover; the soundtrack is ASMR food and eating sounds only.

Change one thing, keep the rest

.revise changes a single item without re-planning the run. Swap the riddle text and you get back only the prompts that quote it; change the character or art style and the affected clip prompts plus the reference image are re-emitted. Everything untouched stays locked.

What it does not do

Writes prompts and plans only — it does not generate video.
Does not touch your Flow account or upload anything on your behalf.
No account, no sign-in. It runs inside your own ChatGPT conversation.
Collects nothing. See the privacy policy.

How to get it

Flow Story Maker is a skills-only ChatGPT plugin, packaged for the public ChatGPT plugin directory. It is not listed yet — the submission is being prepared, so there is nothing to install today, and this page will carry the install link the moment the listing goes live.

Everything it runs on is already public. The compiled rulebook is at github.com/jovylle/googleflow-stories: paste it into a ChatGPT project and it works the same way. To start a run, say what you want to make — for example horror riddle story, or attach a food photo and say mukbang — or type .start for the full interview.

The shape of the output

A clip-map line reads like this (quoted from the rulebook):

Clip 2 of 3, 0:08–0:16, thinking beat: the victims hesitate while the Teller waits in silence

Each clip prompt then carries the same information in shooting form:

[timestamp header]   Clip N of M + nominal block span
[look binding]       master context: character, style, palette, lighting
[handoff]            previous clip's ending state -> this clip's opening state
[scene]              subject, action, camera
[audio]              locked voice / silence + language