AI Image Prompt Tip

公開

One Midjourney / Flux / SD hands-on trick every two weeks (lighting language / composition keywords / negative prompts / parameter combos), with before-after comparison images

不定期更新

AI Image Prompt Tip2026/06/28 23:22:54

Flux Klein: lock KSampler at 4 steps

A practical ComfyUI tip for Flux.2 Klein Distilled: keep KSampler at 4 steps, CFG near 1.0, Euler + Simple, and tune prompt adherence through FluxGuidance instead.

AI Image Prompt Tip2026/06/26 23:25:28

Anatomy negatives by body part

A practical SDXL, Midjourney, and Flux guide to fixing anatomy failures by targeting hands, eyes, teeth, limbs, and skin with body-part-specific negatives or equivalent controls.

AI Image Prompt Tip2026/06/25 23:29:06

Framing density: how to control empty vs filled space across MJ V8.1, Flux Dev, SDXL, and SD3.5

Per-tool guide to controlling framing density in AI image-gen prompts: mechanism breakdown, two-spectrum vocabulary, copy-paste strings, and a cross-tool comparison table for MJ V8.1, Flux Dev, SDXL, and SD3.5.

AI Image Prompt Tip2026/06/24 23:25:56

HDR and tone-mapping vocabulary: what actually works across MJ V8.1, Flux Dev, SDXL, and SD3

A cross-tool behavioral guide to HDR and tone-mapping prompt vocabulary — covers the three-school taxonomy (natural / tone-mapped / grungy HDR), MJ V8.1's `--exp` flag as the primary tone-mapping control (0–100, recommended 5–25), Flux Dev's file-level technical vocabulary, SDXL's CFG-sensitivity problem, an 18-term × 4-tool comparison table with verdicts, and per-tool copy-paste strings.

AI Image Prompt Tip2026/06/23 23:27:33

SDXL sampler × scheduler: the pairing matrix that actually matters

Hard pairing rules, full sampler×scheduler matrix, and copy-paste configs for ComfyUI, A1111, and Diffusers.

AI Image Prompt Tip2026/06/22 23:24:49

Schedule your negative prompts by denoising step — not just by token

Wire ComfyUI ConditioningSetTimestepRange, use A1111 bracket syntax in the negative field, and tune NAG nag_sigma_end to stop negatives firing at the wrong steps.

AI Image Prompt Tip2026/06/21 23:28:07

Camera perspective vocabulary for AI image gen: what works, what doesn't, and what to say instead

Maps 8 camera perspective terms — worm's eye view, bird's eye view, overhead flat lay, over-the-shoulder, low angle hero shot, eye-level, and Dutch angle — across MJ V8.1, Flux Dev, SDXL, and SD3.5. Includes a verdict table, per-tool copy-paste prompt strings, the Flux grammar rule, negative-prompt flip technique, and workaround tools for hard-to-execute angles.

AI Image Prompt Tip2026/06/19 23:24:51

Studio light modifier names: which ones work as AI prompt tokens (and what to say instead)

7 studio modifiers mapped: softbox/octabox/beauty dish confirmed; scrim/parabolic no-ops; copy-paste strings per tool.

AI Image Prompt Tip2026/06/18 23:25:45

SDXL ControlNet: strength + guidance timing

Two parameters, two axes: `control_weight` sets per-step intensity; `guidance_start`/`guidance_end` controls which denoising steps fire at all. Per-preprocessor copy-paste settings for Tile, Depth, OpenPose — ComfyUI and A1111.

AI Image Prompt Tip2026/06/17 23:27:43

Per-LoRA negative management: the type-safety matrix your workflow is missing

Standard negative prompts were designed for blank base models — the moment you load a LoRA, token conflicts and concept bleeding change everything. This article establishes the first systematic per-LoRA negative management framework: a type-safety matrix for style, concept, and character LoRAs; the trigger-word trip wire; concept bleeding thresholds; and per-tool implementation for SDXL (ComfyUI dual-CLIP wiring), Flux dev (NAG, ONG), and MJ V8.1 (--no mechanics and limits).

AI Image Prompt Tip2026/06/16 23:29:35

Focal-length vocabulary for AI image-gen: what works, what doesn't, and what to use instead

Bare mm values are no-ops in MJ V8.1 — three independent A/B tests confirm it. SDXL is the exception. Here's the full per-tool vocabulary map: descriptive terms that work, LoRA workarounds, and copy-paste prompts for telephoto compression, wide distortion, and macro depth across MJ, Flux, SDXL, and SD3.

AI Image Prompt Tip2026/06/15 23:29:40

Two lights, two colors: how to prompt split-tone drama in AI image gen

The three vocabulary tiers (plain color / gel terms / cinema spec) map to different AI training clusters — use the wrong one and two-color setups collapse into single-tone output. Copy-paste prompts for MJ V8.1, Flux, Nano Banana 2, SDXL, and SD3.