skills.rest
devlitus
devlitusCOMMUNITY
Explore all skills by this author →
GitHub

best-zimage-prompting

Generates natural-language prompts and ComfyUI sampler recipes for Z-Image-Turbo image generation.

Updated Jul 18, 2026
Search Tags:#prompt engineering#image generation#text-to-image#comfyui#z-image-turbo#diffusion model#hunyuan3d
One-click install
npx skills add https://github.com/devlitus/space-flight---combat-simulator --skill best-zimage-prompting-devlitus
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: best-zimage-prompting
Source: https://github.com/devlitus/space-flight---combat-simulator/tree/main/.codex/skills/best-zimage-prompting
Command: npx skills add https://github.com/devlitus/space-flight---combat-simulator --skill best-zimage-prompting-devlitus

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Z-Image-Turbo behaves differently from SDXL-style models: it ignores negative prompts, requires fixed low step counts, and reads natural language instead of tags. This Skill prevents wasted generations by encoding the correct prompt formula, sampler settings, and troubleshooting rules. ## Core Features & Use Cases - Six-block prompt formula: Structures prompts as subject, action, environment, lighting, single style, and composition with positive-only constraints. - ComfyUI recipe: Prescribes steps 8, CFG 1.0, res_multistep/euler samplers, native resolutions, and VRAM variants (INT8, BF16, FP8, GGUF). - Bilingual text rendering: Rules for legible English and Chinese text inside images using exact quoted strings. - Use Case: Generate a full-body, front-view A-pose character reference on a clean gray background as input for Hunyuan3D image-to-3D conversion. ## Quick Start Use the best-zimage-prompting skill to write a Z-Image-Turbo prompt for a full-body chibi farmer character reference for Hunyuan3D.

Frequently Asked Questions about best-zimage-prompting

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write prompts for Z-Image-Turbo in ComfyUI?▼

Write natural-language phrases of 30-120 words following the six-block order: subject, action, environment, lighting, one style, and composition. Avoid SD-style tags like 'masterpiece' or 'best quality', which produce generic results.

What sampler settings work best for Z-Image-Turbo?▼

Use 8 steps with CFG 1.0 and the res_multistep sampler with simple scheduler, or euler with simple/ddim_uniform. More steps or higher CFG burns the image or adds grid artifacts because the model is distilled to about 8 NFEs.

Does Z-Image-Turbo support negative prompts?▼

No, the negative prompt field is ignored because the model was distilled with zero guidance. All restrictions must go inside the positive prompt as explicit exclusions like 'no watermark' or 'no extra text'.

Can Z-Image-Turbo render text inside images?▼

Yes, it renders legible English and Chinese text simultaneously. Put the exact text in double quotes, specify typography and position separately, and keep it short—labels and headlines, not paragraphs.

Why does Z-Image-Turbo produce plastic-looking faces?▼

The default prior is glossy 'beauty stock' output. Add a real action, candid unposed framing, gaze direction, and texture vocabulary like 'natural skin texture, visible pores' or specific camera and film references.

Which Z-Image-Turbo model variant should I use for low VRAM?▼

For 6-12 GB VRAM use GGUF Q5-Q8 with the ComfyUI-GGUF node; around 8 GB use FP8 e4m3fn for quality or e5m2 for speed. BF16 needs roughly 14-16 GB, and INT8-ConvRot works natively from ComfyUI v0.27.

Related Skills

Explore More Design & Creative Skills→
artokunartokun
715

krea2-txt2img

Build Krea 2 Turbo txt2img workflows in ComfyUI with Qwen3-VL encoding and JSON prompting.

Community
Intermediate
guia-matthieuguia-matthieu
145

image-to-3d-pipeline

Transform 2D images into animated 3D models with AI generation and rigging.

Community
Advanced
huivrotikihuivrotiki

comfyui-runpod-agent

Queue ComfyUI prompts and manage workflows on RunPod via OpenClaw agents.

Community
Advanced
wubo0927wubo0927
48

cinematic-director

Generates cinematic photography-style image prompts and negative prompts from scene descriptions.

Community
Advanced