How to write a stable AI prompt
An 8-part formula — subject, scene, style, light, composition, material, quality, parameters — for prompts you can reuse and edit.
Read guideA structural system for AI prompt writing: 5 long-form guides + 4 real teardowns + 35 deep articles, paired with 6 browser-local tools. No cloud APIs. Your data stays in your browser.
Need a stable, reusable prompt framework — no more reroll roulette.
Want product, poster and IP visuals as reusable templates.
Want to understand prompt structure, not just copy keywords.
See real differences across models, parameters, LoRAs and ControlNets.
An 8-part formula — subject, scene, style, light, composition, material, quality, parameters — for prompts you can reuse and edit.
Read guideThree product-shot skeletons (e-commerce, brand poster, luxury) with material, lighting and negative-space discipline.
Read guideIdentity anchors, appearance anchors, fixed wardrobe, signature props, and LoRA / Midjourney --cref.
Read guideA ceramic-artist portrait broken down by identity, wardrobe, scene, light and lens — with three rewrites.
See teardownA floating green tea can with material, light, composition and negative-space cues, rewritten for three categories.
See teardownA 5-second rain-night still with static camera, single micro motion and explicit stability.
See teardown0, 100, 250, 500, 1000 compared — what each value does and which to pick per scenario.
Read articleFour-dimension comparison covering quality, speed, LoRA ecosystem, prompt friendliness and licensing.
Read articleWeights, trigger words, stacking multiple LoRAs and common pitfalls explained.
Read articleOpenPose, Canny, Depth and Lineart preprocessors with the standard workflow.
Read articleRunway, Pika, Kling, Sora, Seedance, Hailuo and Veo with scenario-based recommendations.
Read articleMidjourney, SDXL, Flux Dev/Schnell and LoRAs — what is allowed and the four risk areas.
Read article