# Minimax workflow for absolute highest quality output? — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/8fc80786f7576270

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-27 15:54 ingested 2026-10-28 02:37

[original source ↗](https://www.reddit.com/r/StableDiffusion/comments/1xgwamv/minimax_workflow_for_absolute_highest_quality/)

## retained text



```
There are a thousand minimax workflow, and while each serve their purpose, I’ve yet to find one that produces the same near-lossless or cinematic quality output I’ve seen others post here. Almost every workflow uses 8 steps and/or Patch Sage Attention, Model Attention Backend, H3 Memory Optimization, etc, which ultimately hurt the output. I have tried 20, 25, 30 steps — and while they’re good, they’re not great. 

I have used LM Studio to build my prompts, which has drastically improved the adherence and flow, but the quality still isn’t there. I have tried resolution ranging from .5 - 1.5, marginally impacts quality but hinders adherence when above 1.1. 

Assuming a clapped out system with a 5090 or renting a pod with bigger specs, is there a key workflow to use that focuses on output quality, speed be damned?
```



document fetched 2026-10-28 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1xgwamv/minimax_workflow_for_absolute_highest_quality
