# How do you keep track of what actually works? — retained record

> Retained reddit record from reddit:comfyui.

Canonical: https://highsignal.app/data/records/d62fc3f70d14214d

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:comfyui)  published 2026-10-27 08:07 ingested 2026-10-28 02:38

[original source ↗](https://www.reddit.com/r/comfyui/comments/1xgmml5/how_do_you_keep_track_of_what_actually_works/)

## retained text



```
I’m new to ComfyUI and I can’t quite put my finger on it, but (local) image gen feels a bit like gambling. You write a prompt, wait a few seconds and get something different every time.

Running it locally, the number of variables just explodes: prompt, base model, LoRAs, LoRA weights, seed, text encoders, CFG, steps, sampler...

How do you make sense of it all? How do you figure out which combinations reliably give you good results?

Specifically curious about:

How you organise and compare outputs visually (grids, XY plots, galleries?)

How you log the settings behind a good result so you can reproduce it

Whether you change one variable at a time or have a smarter workflow



Would love to hear how you all approach this. Thanks :)
```



document fetched 2026-10-28 00:41 · canonical https://reddit.com/r/comfyui/comments/1xgmml5/how_do_you_keep_track_of_what_actually_works
