nodes/theonetwoone-Multimodal-LLM-ComfyUI-node

Files

T

theonetwoone a853a17ec6 Restructure repo with extension at root

Moves all node files/presets/workflow to repository root so cloning into ComfyUI custom_nodes works without an extra subfolder.

Made-with: Cursor

2026-04-28 18:46:55 +02:00

1.1 KiB

Raw Permalink Blame History

Model dialects quick reference

Use this as a reminder for how to format prompts per generator.

SDXL

Likes structured spec + optional tag stack.
Can use weighting (token:1.2) and sometimes BREAK.
Negatives are supported and useful; keep them short and targeted.
Front-load subject and identity anchors.

FLUX

Prefer structured natural language (subject + action + style + context).
Do NOT rely on SDXL weighting syntax; it’s usually ignored.
Usually avoid negative prompts unless your specific pipeline supports them.
30–80 words is a good sweet spot for complex scenes.

Z-Image

Plain natural language, explicit constraints.
Subject first; then scene/lighting/composition/style/materials.

WAN (video)

Focus on motion + camera movement + environmental motion.
Time markers can help multi-beat motion: [0-3s] ... [3-7s] ...
Negatives: flicker, morphing, jitter, distorted face, watermark/text.

LTX (video)

One flowing paragraph, present tense, 4–8 sentences.
Include shot/camera, scene/lighting/atmosphere, characters (physical details), action sequence, camera movement end-state.