Engineering·17 min
Why Do I Get a Different Answer Every Time?
Why AI gives different answers each time: two major vendors publish opposite claims about temperature 0. The real causes, and 14 prompt patterns that narrow the variance.
Read moreBlog
Better prompts. Better AI results.
Tutorials, comparisons, and frameworks to write better prompts for ChatGPT, Claude, Gemini, Midjourney, Veo3 and Kling.
Page 18 of 29
Image·20 min
Midjourney Quality Parameter and Speed Modes: The GPU-Time Reference (2026)
Checked against Midjourney's own docs on August 27, 2026: the quality parameter is marked unsupported on V8.1 and V8.2, and so is Turbo. What Fast, Relax, Draft, SD and HD cost.
Read moreChatGPT·17 min
When a Conversation Goes Bad, Restart It (Here's How)
A poisoned ChatGPT thread doesn't have to cost you everything in it. A four-part extraction method for pulling the facts, decisions, and drafts worth keeping, and leaving the rest behind.
Read moreVideo·25 min
Structuring a 20-Second Generation: LTX 2.5 and LTX 2.3 Prompting
LTX 2.3 prompting for long takes: where the real 20-second ceiling is, how to pace beats across it, and 18 copy-paste prompts built to hold for the full duration.
Read moreImage·18 min
Why Did My Image Come Out the Wrong Size?
An AI image with the wrong aspect ratio is a parameter problem, not a prompt problem. A per-surface table of what actually controls dimensions, plus 20 copy-paste examples.
Read moreEngineering·16 min
Why Did the AI Answer a Different Question?
The AI answered the wrong question, confidently. Eight specific causes, compound asks, buried questions, wrong referents, format swallowing the ask, each with a copy-paste before/after fix.
Read moreChatGPT·18 min
Red-Team Your Own Prompt Before You Trust the Output
Eight adversarial tests that expose sycophancy, decoration and fabricated sources in a prompt that already looks fine. 22 copy-paste blocks to test your prompt before you ship it.
Read moreLibrary·17 min
Best Free Prompt Managers in 2026 (No Credit Card)
A free prompt manager should not need a card. Eight tools with a real free tier, verified from each vendor's own page in August 2026, plus what every free plan actually caps.
Read moreImage·14 min
Camera and Lens Terms That Change Your AI Images
The camera and lens terms that actually change an AI image: focal length, aperture, sensor format, bokeh, anamorphic, tilt-shift, film grain, each defined with a copy-paste fragment.
Read moreVideo·19 min
Video Character Consistency: Keeping One Character Across Clips
Video character consistency ranked by what actually works: long takes, reference images, frame chaining, extending, and text descriptions, with per-model caps from vendor docs.
Read moreEngineering·17 min
How to Turn a Great Conversation into a Saved Prompt
You corrected a conversation five times and finally got the output right. Here's how to extract that into one reusable prompt, before the thread closes and the trick is gone.
Read moreChatGPT·17 min
Follow-Up Prompts That Actually Improve the Answer
The first answer wasn't wrong, just not as good as it could be. 18 copy-paste follow-ups for deepening, tightening, and pressure-testing a decent ChatGPT answer into a good one.
Read moreChatGPT·16 min
10 Bad Prompts, Fixed (Before and After)
Ten bad prompt examples that look reasonable at a glance and fail for a specific reason, each with the fix and the one change that actually mattered.
Read moreEngineering·19 min
Why Does AI Agree With Everything I Say?
AI agrees with everything you say because human-feedback training rewards agreement. The measured evidence, OpenAI's own GPT-4o rollback, and 17 prompts that make a model push back.
Read moreEngineering·17 min
Why Is My Negative Prompt Being Ignored?
Negative prompt not working? Six documented causes, checked against vendor docs: no field exists, wrong endpoint, vendor advises against it, silent truncation, precedence, and weighting.
Read moreImage·17 min
Midjourney No Parameter (--no): The Exclusion Reference (2026)
Checked against Midjourney's own docs on August 27, 2026: --no applies a -0.5 weight, not a filter. Where it is documented to work, and the two places Midjourney contradicts itself.
Read moreVideo·17 min
AI Video Duration Parameters by Model (2026 Reference Table)
Every AI video duration parameter, with the resolution and frame-rate couplings that break it: LTX, Seedance, Kling, Veo 3.1, Sora 2, Runway, Wan, Luma, Hailuo and Grok. Sourced and dated.
Read moreEngineering·16 min
Using MCP Inside Claude Code: Setup, Scopes and Auth
A Claude Code MCP guide built from Anthropic's own docs: the four ways to add a server, what local vs project vs user scope changes, OAuth and token auth, and how to verify a connection.
Read moreImage·17 min
Midjourney Seed: Reproducing a Result with --seed (2026)
Midjourney's own Seeds article files the parameter under a heading reading Not for Saving Styles. What --seed reproduces, what it never has, and what to use instead. Checked August 27, 2026.
Read moreChatGPT·20 min
How to Prompt NotebookLM (Source-Grounded Only)
NotebookLM prompting inverts normal prompt advice, because it answers only from your sources. Google renamed it Gemini Notebook in July 2026. 19 prompts, verified at Google.
Read moreEngineering·18 min
Why Does AI Get Dates and Arithmetic Wrong?
AI wrong math has three causes: a tokenizer that chops numbers into odd chunks, no calculator inside the model, and no clock. The mechanisms, plus 21 copy-paste blocks.
Read moreImage·20 min
Stable Diffusion Steps: What Sampling Iterations Actually Do (2026)
Stability's hosted image API exposes no steps parameter at all. What a sampling step is, why more stop helping, and how the right count depends on your sampler. Checked August 27, 2026.
Read moreEngineering·19 min
Lovable Prompts: 22 Templates for No-Code App Building
Lovable prompts have to carry product decisions a chat prompt never does. Lovable's own prompting docs, 22 copy-paste templates by build stage, and where the tool stops.
Read moreVideo·20 min
Vidu Prompting: How to Prompt Vidu Q3
Vidu prompting for Q3, read off Vidu's own API reference: five Q3 model strings, 1-16s duration, the audio flag that replaced bgm, @subject syntax, and 18 copy-paste requests.
Read morePage 18 of 29