CLAUDE SETUP
============

Claude takes ONE file per formula. There is nothing to upload and nothing that can be
left half set up: you paste the file, and that is the whole setup.

  1. Open a new Claude conversation, or make a Claude Project.
  2. Paste the whole file in. In a Project, put it in the project's custom instructions
     instead and every new chat in that project already knows it.
  3. Then describe the video you want, and attach the picture or pictures if the
     formula takes them.


WHICH FILE
----------

THE THREE SINGLE-MODE FORMULAS are the ones to use with the FastH3 workflows. Each does
one job and its rules were written for that job:

  Minimax H3 Text2Video - CLAUDE (paste) v1.txt
      A video from words alone, with its own sound. Nothing attached.

  Minimax H3 FirstFrame2Video - CLAUDE (paste) v1.txt
      A video that STARTS on a picture you attach. One picture.
      Attach it to the message where you ASK, not to the message where you paste
      the formula.

  Minimax H3 FirstAndLastFrame - CLAUDE (paste) v1.txt
      A video that starts on one picture and ENDS on another. Two pictures, the first
      frame first. This one also needs the number of seconds, because the closing
      picture has to land on an exact second. Say it in your request and you will not
      be asked for it.


Match this to the model you actually loaded in ComfyUI.


WHAT YOU GET BACK
-----------------

One code block, in the exact shape MiniMax H3 wants. Copy it into the Prompt node in
the workflow and press Run.

Sometimes there is one short line under the block. That line is for you and not for
the workflow: it is the formula telling you it had to drop some of what you asked for
to fit the seconds, or that your spoken line runs longer than the clip. Read it, and
either accept it or ask again with a longer duration.

The first-and-last-frame one may ask you a question instead, and only ever one: how
many seconds.


THINGS WORTH KNOWING
--------------------

  * ASK FOR WHAT YOU WANT IN PLAIN WORDS. "locked-off camera", "no music", "silent
    clip", "add gentle acoustic guitar" all work and are all obeyed. You do not have
    to learn the format; that is what the formula is for.
  * DO NOT ASK FOR SUBTITLES. The video model speaks every line of text it finds, so a
    subtitle makes it say the line twice. The formulas refuse them for that reason.
  * A LINE TAKES ABOUT THREE WORDS A SECOND. Twenty words needs seven. The formula
    keeps your words exactly as you wrote them and tells you if they will not fit; it
    will never quietly shorten what somebody says.
  * THE CLIP LENGTH COMES FROM THE WORKFLOW, not from the prompt, except in the
    first-and-last-frame mode where the closing picture has to be placed in time.

