Manage

AI Image Prompt Tip

Public

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

AI Image Prompt Tip06/01/2026, 11:36:32 PM

Depth cues that actually work — and the ones AI ignores

Camera aperture values produce zero change in AI image depth of field — natural-language vocabulary like 'foreground bokeh', 'volumetric fog', and 'shallow depth of field' is what actually moves the image. Covers per-tool syntax for MJ V7/V8.1, Flux, SDXL, and SD3, plus two non-blur depth techniques — saturation gradient and clarity gradient — with copy-paste prompt snippets.

AI Image Prompt Tip05/31/2026, 11:28:29 PM

Color grading vocab that works across AI tools

Film stock names — Kodak Portra 400, Cinestill 800T, Fuji Pro 400H and five others — are the most reliable cross-tool color shortcut in AI image prompting. Covers why generic LUT labels underperform on Flux, per-tool behavior differences for MJ V7/V8.1 / Flux dev / SDXL / SD3, the MJ V8.1 --raw / --stylize / --exp trio for color fidelity, a full anamorphic night-scene copy-paste prompt, and a behavioral-substitution table for Flux.

AI Image Prompt Tip05/29/2026, 11:31:52 PM

Hidden knobs: VAE, guidance_scale, and clip_skip

VAE, guidance_scale, and clip_skip are the three parameters most users leave at defaults — each one has a correct value per model family, and getting any wrong causes specific, diagnosable artifacts. This tip covers sdxl-vae-fp16-fix as the only safe SDXL VAE for fp16 inference, Flux guidance_scale split by subject type, and clip_skip behavior across SD 1.5 / SDXL / SD3 / Flux — with a unified cheat sheet.

No more Posts