Prompt Better.

Match this reference, compare the result, and fix the biggest differences. Use [attached reference] to create [editable artifact] for [purpose]. Preserve [the three qualities that matter most]. First describe what you can actually observe in the reference and flag anything you cannot access. Build a first version, render or preview it with the tools available, and compare it against the reference. Fix the three most important discrepancies, then repeat once. Keep the result editable and tell me what still differs. Do not claim you inspected a result you could not open.

Make the feeling concrete: specify the scene, motion, controls, and finish. Create [artifact] for [audience] with a [desired mood] feel. Translate that mood into a short brief covering composition, materials or typography, lighting or color, motion, and the user's main interaction. Use [reference or constraints]. Choose a coherent direction and build it. Include a usable small-screen layout if this is a website, a reset or return action where needed, and a reduced-motion option for nonessential animation. Show the working result and identify the unfinished details.

Use the right tool for each stage and keep every intermediate file editable. Make [result] using [available tools and source assets]. First list what you can actually access, what each tool will do, and what file or data passes to the next stage. Reuse suitable supplied assets and preserve their credits. Complete one small end-to-end example before expanding. Check each handoff by opening its output in the next tool. Save the source project as well as the finished export. If an integration or permission is missing, give me the exact next step rather than claiming the action succeeded.

Show the whole, then let me isolate, label, and inspect each part. Build an interactive explanation of [system] for [audience], using [trusted source material or data]. Start with an overview. Let the viewer select a component, isolate it, see its name and role, and return to the complete system. Add [an exploded view, layer toggle, cutaway or linked chart] only where it helps answer a real question. Keep labels readable and provide a reset. Clearly mark illustrative geometry and unverified facts. Test each control with one concrete example.

Guide the audience through one question at a time. Explain [topic] to [audience] as [a short video, guided walkthrough or interactive story]. Use [source material]. Outline five beats: the opening question, essential context, the mechanism, a concrete example, and a useful takeaway. For each beat, name the question answered, the visual change, and the shortest necessary caption or narration. Build one complete beat first, then assemble the sequence. Keep transitions purposeful, include pause or replay where relevant, and flag facts that need checking.

Separate what you observed, what you inferred, and what would change your mind. Use [recording, data or evidence] to build a provisional representation of [question]. List the observations first. Show the assumptions that connect them to your proposed explanation. Where several explanations fit, show at least one alternative. Make the important assumptions adjustable if useful. Label illustrative values and unknowns visibly. Suggest one feasible observation or test that would help distinguish the alternatives. Do not present an attractive reconstruction as a uniquely established result.

Build the smallest usable version and show me that its main action works. Create a first version of [idea] for [user]. Its essential loop is [input → action → useful result]. Define three observable acceptance checks and build only what is needed to pass them. Use the actual available environment and assets. Run those checks, including one failure or edge case, and report what you observed. If you cannot execute a check, mark it untested. Stop after the core loop works and list the highest-value next improvement; do not expand into a full product yet.

Create two contrasting versions and judge them against the same goal. For [task and audience], propose two meaningfully different approaches. Keep the source material, scope and constraints the same. Before building, define three criteria tied to the intended use. Produce small comparable versions, inspect both, and explain the tradeoffs using those criteria. Recommend one for my purpose, or a specific combination, with a concrete next experiment. Distinguish observed differences from guesses; do not treat a single attractive result as a general model benchmark.

Read the tool’s documentation and use its native features before adding workarounds. Recreate [effect or behavior] in [application] using [reference]. Check the current official documentation for the features best suited to the job. Explain the simplest native implementation and any tool access you need. Build a small editable example, inspect the result, and compare it with the reference. If the approach stalls, reassess the mechanism before stacking more patches. Name the documentation consulted and mark any step you could not verify.