Lab
Prompt parameter experiments
Small prompt constraints change outcomes more than “more adjectives.” Each experiment states a hypothesis, shows copy-ready control and variant prompts, then before/after results and notes.
Experiment 1
Product: soft fill vs directional key light
Hypothesis
Naming a single key-light direction and soft fill reduces flat plastic highlights on product shots.
Control prompt
Studio product photo of 【product】 on a clean table. Bright lighting, commercial look, sharp details.
Variant prompt
Studio product photo of 【product】 on a clean table. Single soft key light from camera-left at 45°, gentle fill from the right, subtle contact shadow under the product, natural material reflections, commercial catalog look.
Results
Takeaway
The variant usually keeps materials readable and shadows grounded. Prefer Pro when packaging text must stay sharp.
Experiment 2
Interior: vague “modern” vs material + camera constraints
Hypothesis
Specifying materials, camera height, and time of day yields more believable staging than “modern interior” alone.
Control prompt
Modern interior living room, beautiful design, high quality photo, realistic.
Variant prompt
Wide interior photo of a living room, camera at eye level. Oak floor, linen sofa, matte white walls, soft afternoon window light from the left, subtle depth of field, no clutter, photorealistic staging.
Results
Takeaway
Control prompts often invent mixed styles; the variant locks a coherent palette and light path.
Experiment 3
Portrait: beauty enhance vs texture-preserving polish
Hypothesis
Explicitly keeping pores and fabric detail prevents over-smoothed plastic skin while still improving clarity.
Control prompt
Enhance this portrait, make the skin perfect and beautiful, high-end beauty retouch, ultra smooth.
Variant prompt
Enhance portrait clarity and lighting. Keep natural skin texture and visible pores. Soften only temporary blemishes. Preserve fabric weave and hair strands. Mild polish—no plastic skin, no waxy highlights.
Results
Takeaway
Use identity-lock language when editing a specific person; Pro helps when face consistency is critical.
FAQ
Are these scientific A/B tests?
They are illustrative lab notes for prompt design—not formal model benchmarks. Re-run with your own subjects for production decisions.
Can I copy both prompts?
Yes. Use the copy buttons, replace 【placeholders】, and compare in the create app.
Where do I go next if results still fail?
Check the failures page for symptom-based fixes, or download the resources template pack for industry starters.