MNeMoNiCuZandClaude Opus 5.5
92a58d6774
- Migrate all nodes to the V3 node API; old workflows migrate automatically - Add help pages for every node, opened from a ? icon in the title bar - Minor fixes and tooltip, placeholder and default improvements - Bump version to 3.0.0 Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
1.5 KiB
🎲 Load Random Checkpoint
Loads a checkpoint from a list you define, cycling or shuffling through it as the seed changes. Built for batch runs that should sweep several models.
Inputs
-
checkpoints — One entry per line. Each can be:
- a name, fuzzy-matched against your checkpoint files (
realvis) - a path relative to the checkpoints folder (
SDXL/Realistic/) - an absolute path (
C:/models/foo.safetensors) - a folder, which adds every
.ckpt/.safetensorsinside it
Blank lines are ignored.
- a name, fuzzy-matched against your checkpoint files (
-
seed — Drives the selection. Set its control to increment to walk the list.
-
repeat_count — How many consecutive seeds share one checkpoint. With
3, seeds 0-2 use model A and seeds 3-5 use model B. -
shuffle — Off: the pool is walked so nothing repeats until every entry has been used. On: each pick is independent, so the same model can come up twice in a row.
Outputs
- model / clip / vae — From the selected checkpoint.
- path — Full path of the file that was loaded.
How it works
The pool is built once per unique input list and cached, then ordered using the
seed. Within one repeat_count window the same file is reused rather than
reloaded. Name matching prefers substring hits and falls back to fuzzy
similarity; ties keep every equally-good match in the pool.
Turn on Settings → ⚡MNeMiC Nodes → Load Random Checkpoint → Console Logging to see the pool, the index and the match scores.