Model guides / 03
Name the outcome, supply the material, describe the audience, define the constraints, and say what a useful answer looks like. Model tuning comes after that.
Writing, structured work, iteration, analysis, and tasks that benefit from a clear output contract.
Role → task → context → constraints → output format → verification.
Asking for a vague ‘best answer’ without audience, material, or a definition of done.
Long documents, careful synthesis, nuanced writing, and work with substantial context.
Structured context → assignment → decision criteria → desired artifact.
Burying the actual task inside a large paste without labels or priorities.
Multimodal work, document analysis, research workflows, and structured creation.
Goal → context/files → specific instructions → examples → output.
Mixing background, instructions, and examples without making their roles clear.
Direct brainstorming, current-context exploration, file analysis, and conversational iteration.
Objective → known facts → stakes → verification needs → recommendation.
Treating current information as verified when the answer does not show what was checked.
Portable prompts you want to reuse across several assistants.
Task → context → constraints → useful output → missing-information check.
Tool-specific commands that make the prompt fragile elsewhere.
Instruction-led image creation and edits, including exact text, transparent backgrounds, and reference-based changes.
Subject → purpose → art direction → composition → light/color → exact text → output size.
Leaving desired text, aspect ratio, or preserved details implicit.
Distinctive visual exploration, art direction, style references, and concept development.
Subject → medium → environment → lighting → color → mood → composition → parameters.
Long instruction prose when a concise visual description would be clearer.
Conversational image creation, rapid visual exploration, and iterative edits.
Subject → setting → visual direction → composition → light/color → exact text or no text.
Stacking several conflicting scenes or styles into one request.
Reference-aware image creation, editing, layout work, and assets that need clear text or preserved details.
Create/edit request → labeled references → invariants → composition → text → output size.
Attaching references without saying what to borrow and what must remain unchanged.
Detailed image generation and editing with references, controlled composition, and strong visual direction.
Creative intent → references → scene → subject → layout → light/color → invariants.
Leaving the role of each reference image ambiguous.
Text-to-video and reference-led video with detailed camera, action, atmosphere, and audio direction.
Subject → action → style → shot → camera movement → light → audio → final frame.
Describing a mood without saying what physically happens in the shot.
Motion-focused clips, image-to-video, controllable camera direction, and visual iteration.
Visible subject → one clear action → positive motion language → camera → ending.
Negative phrasing. Describe the movement you do want instead.
Complex motion, multi-shot storytelling, mixed reference media, continuity, and synchronized audio-video creation.
Reference media → continuity anchor → timed shots → transitions → audio → end frame.
Writing a multi-shot prompt without timing, continuity details, or a clear transition between beats.
Storyboard-like video generation with deliberate timing, camera direction, character motion, dialogue, and sound.
Scene → subject → timed beats → camera → visual style → sound → final frame.
Overloading a short clip with too many simultaneous actions or cuts.
Fast video concepts built around a clear scene, readable action, camera movement, and sound.
Subject → setting → one action → shot → camera → sound → ending.
Adding unrelated actions that compete for attention in a short clip.
Motion-led clips that need subject continuity, physical detail, and specific starting and ending frames.
Starting frame → motion path → camera → ending frame → continuity → format.
Describing the look but not the motion path or final composition.