# I've been working on a different approach to regional prompting in ComfyUI, with paintable regions and more control over blending — retained record

> Retained reddit record from reddit:StableDiffusion.

Canonical: https://highsignal.app/data/records/aa3bba4fa9697074

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:StableDiffusion)  published 2026-10-27 12:58 ingested 2026-10-28 02:37

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

## retained text



```
I've been working on an open-source ComfyUI extension called **Conditioning Layers** and wanted to share it here.

The idea is to give you more control over where different prompts influence an image, especially with irregular shapes and overlapping regions.

It uses editable grayscale **Influence Fields**, which you can paint or generate procedurally. You can assign exclusive regions (FIXED), blend multiple prompts with normalized weights (BLENDED), and inspect their effective contributions before sampling.

The examples show a castle transitioning into an underwater reef and a single tree with four seasons following curved influence fields. Both were generated locally using **Krea 2 Turbo**.

**GitHub:** [https://github.com/SurrealByDesign/conditioning-layers](https://github.com/SurrealByDesign/conditioning-layers)

**Workflows:** [https://github.com/SurrealByDesign/conditioning-layers/tree/main/examples](https://github.com/SurrealByDesign/conditioning-layers/tree/main/examples)

The tested setup is ComfyUI 0.34.6 with Krea 2 Turbo BF16, 1024×1024, 8 steps, CFG 1, Euler/simple. The extension is free under Apache-2.0 and still experimental.

The fields guide conditioning during generation, not the finished pixels, so boundaries aren't guaranteed to be exact.

I'd appreciate any feedback, especially from people already experimenting with regional prompting in ComfyUI.
```



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