A 20-prompt suite for testing any video model
Twenty short prompts, each isolating one known failure axis, so a new model's weaknesses show up in an hour instead of halfway through a client project.
Every guide, comparison and workflow we’ve published on Prompt Engineering.
21 articles
Twenty short prompts, each isolating one known failure axis, so a new model's weaknesses show up in an hour instead of halfway through a client project.
Shadow edge hardness is set by apparent source size, not by adjectives. A size-and-direction vocabulary that outperforms five style words in every prompt.
Some models now plan composition before they render. Prompt patterns that pin zones for posters, covers and thumbnails without dropping into coordinates.
A 12-word prompt and a 120-word one fail differently. Per-class length targets, plus the ordering rule that keeps your one non-negotiable clause alive.
A list of prompts is not a product. The contents template that separates a pack buyers keep from one they refund within the hour.
Negative prompts are a no-op on flow-matching models and leak the negated noun elsewhere. Positive re-specification and the three constraints that replaced it.
Bowed lines, disagreeing vanishing points and drifting facade counts sink generated architecture. The traits that help, and the prompt that fixes interiors.
Four speech providers parse bracketed delivery cues and four take mood as a parameter. The two camps, four incompatible syntaxes, and how to convert a script.
Long-window image models take far more instruction than most prompts supply. A block structure for long prompts, and where extra tokens stop mattering.
Hallucinated stock marks come from training data, not the platform. Prompt changes that suppress them, plus a clean removal pass when they appear.
When one call can plan, generate, edit and post, the prompt stops being the unit of work. A brief — constraints, budget, fallback — replaces it.
One AI icon is a five-minute win. Forty that share the same stroke weight and corner radius is a design-system problem — here's how to actually hold it.
Isometric scenes look clean until one element quietly picks up a vanishing point. Why that happens, and how to hold the projection as a diorama gets busier.
Prose can't tell a model where to put a headline — 'top-left' is a suggestion, not a coordinate. Structured JSON prompts can, and here's when that helps.
Video models predict plausible next frames, not physical forces. The VideoPhy numbers, four failure categories, and what actually reduces them in practice.
An agent's output quality is bounded by how well its tools are described, not by how carefully you phrase the request. Our own definitions as the example.
VEO 3.1 shot prompts, native audio cues, camera language, and fixes. Run the structure on Versely.
Style keywords that actually change AI images: lighting, lens, and medium terms that move pixels, the words that do nothing, and a cheap test protocol.
Negative prompts for AI video and images: when exclusion beats positive phrasing, what modern models still need excluded, and a minimal stack that works.
The advanced video prompt patterns that separate amateur outputs from broadcast-grade AI footage in 2026 - structure, weighted terms, negative prompts, multi-shot chaining.
The prompt patterns that actually work in 2026 for Flux, Midjourney, Ideogram and Imagen — structure, modifiers, negative prompts and the mistakes that produce generic output.