CHATGPT SETUP
=============

You need BOTH files for your model. They do different jobs.

FL2VA   minimax_h3_fl2va_*.safetensors
        Text to video, first frame, first and last frame, last frame.

REF2VA  minimax_h3_ref2va_*.safetensors
        Reference mode: images, video or audio that define a character, place, style,
        camera move, voice or soundtrack.

Pick the pair matching the model you loaded in ComfyUI.


STEPS

1. Create a new custom GPT.

2. Under "Knowledge", click Upload files and add the KNOWLEDGE (upload) file.

3. Open the INSTRUCTIONS (paste) file, select all of it, and paste it into the
   "Instructions" box, replacing whatever is there.


WHY BOTH

The Instructions box only holds 8000 characters and the full rule file is far bigger,
so it has to be uploaded. But an uploaded file is only SEARCHED, not read in full every
time, so a rule reaches the model only when your request happens to match it. The small
file carries the rules that must never be missed, and it tells the GPT to search the big
file for the rest.

The INSTRUCTIONS files are sized to just under 8000 characters. If you edit one, check
it still fits before saving.


CONVERSATION STARTERS

FL2VA                                    REF2VA
Write a video prompt from text only.     Use this image as the character.
Animate this image.                      Use this audio as the voice.
Go from this first frame to this last.   Reuse this song as the soundtrack.
Use this image as the last frame.        Copy the camera moves from this video.


See "README - start here.txt" in the folder above for what these formulas fix.
