Conditioning Layers: Open-Source ComfyUI Extension With Paintable Prompt Influence Fields
SurrealByDesign · reddit · 2026-10-27
The author open-sourced a ComfyUI extension, Conditioning Layers, offering a new take on regional prompting: editable grayscale Influence Fields, painted or procedurally generated, control where each prompt influences the image—especially for irregular shapes and overlapping regions.
Key features:
- Assign exclusive regions (FIXED) or blend multiple prompts with normalized weights (BLENDED)
- Inspect each prompt's effective contribution before sampling
- Examples include a castle transitioning into an underwater reef and one tree showing four seasons along curved influence fields
Tested with ComfyUI 0.34.6 + Krea 2 Turbo BF16, 1024×1024, 8 steps, CFG 1, Euler/simple. Apache-2.0 licensed and still experimental; fields guide conditioning during generation, not finished pixels, so boundaries aren't guaranteed exact.
- GitHub: https://github.com/SurrealByDesign/conditioning-layers
More from Multimodal
- Recreating an Engine-Animation Demo with Kimi K3.1 Pro — iamfakhrealam · 2026-10-27
- AI concept mashup: a giant Crayola-crayon traffic cone on a real street — Delahuntagram · 2026-10-27
- AI concept: Apple Pencil dressed in a Crayola crayon wrapper — Delahuntagram · 2026-10-27
- Voxel-city demo shows how to prompt Ling-3.1-flash for orbit, zoom, hover and focus interactions — Mrs-Grumpy23 · 2026-10-27
- Re-OCR of 2,724-page Encyclopaedia Britannica costs ~$3.40 with Baidu's Unlimited-OCR-2 — vanstriendaniel · 2026-10-27
- AI video TWIN FOREST HOPPING shows startlingly lifelike character consistency — poppy-slashpot · 2026-10-27