SAMPLE¶
How the model samples, for the whole request.
| Control | Range (default) | What it does |
|---|---|---|
| TEMP | 0.1–3 (1) | randomness: lower is safer and more repetitive, higher is more adventurous |
| TOP-P | 0.05–1 (off) | sample only from the most likely choices that add up to this probability |
| TOP-K | 0–500 (off) | sample only from the K most likely choices |
| ESC | 1–3 (1) | raise the temperature on each retry after a rejected result |
| TAKES | 1–16 (1) | how many versions to generate in one request (see Takes) |
| PROB | 5–100 % (ALL) | regenerate only this share of the painted bars, picked at random (the seed picks the same bars again); the rest stay as they are. The roll dims the painted bars that were left out. AR tracks always generate all their bars |
| NOVELTY | on / off | reject a result that copies the context |
| SILENCE | on / off | reject a result with empty bars |
| SHUFFLE | on / off | vary the order in which the targets are generated |
A rejected result is retried, up to ATTEMPTS times (see Engine).
Seed¶
The seed screen shows which random seed the next generation uses.
| RANDOM | a new seed every time (the default) |
| Click the screen | type a seed: digits only, 1 to 2147483647. Enter sets it; an empty field or 0 goes back to random. |
| DICE | pick a random seed and keep it |
| LOCK | keep the seed you just heard: the last run's, or with takes, the selected take's. Click again for random. |
The label above the number says where the seed came from: LAST RUN, TAKE 3, TYPED or DICE. A fixed seed with the same settings and context gives the same result again.