How to write prompts that actually work on Predis

The prompt is where every generation starts — and where most bad outputs originate. If your content looks generic, uses the wrong product, ignores your brand, or feels "off" in a way you can't quite name, the fix is almost always in how you wrote the prompt, not in the AI itself.

This guide teaches you to write prompts that produce usable content on the first try, so you spend fewer credits on regenerations and less time wondering why the output doesn't match what you had in mind.

The one thing most people get wrong

Most prompts fail because they describe what the product is instead of what the content should look and feel like. There's a meaningful difference.

What people type:

"Create a post about my leather handbag"

What the AI hears:

"I know almost nothing about your goals. Here's a generic handbag image with stock text."

What actually works:

"A premium leather tote bag on a clean marble surface, soft natural light from the left, editorial product photography style, minimal background, warm neutral tones. For an Instagram feed post targeting women 28–45."

The second prompt tells the AI six things: the product, the setting, the lighting, the visual style, the color palette, and the audience. The first prompt tells it one. That's why the outputs are different.

The five elements of a strong prompt

Every effective prompt answers five questions. You don't need to answer all five every time — but the more you include, the better the output.

1. What is the product?

Be specific. "My product" means nothing to the AI. "A 250ml matte-black glass bottle of rosemary hair oil" means everything.

If you've uploaded product images to your brand library, the AI will use them — but the text description still matters because it guides how the product is framed, lit, and styled in the output.

2. Where does it live?

Place the product in a setting. A product floating in a void looks like clip art. A product in a context looks like a campaign.

• Skincare: "on a marble bathroom shelf beside a folded white towel, morning light through a frosted window"
• Coffee brand: "on a rustic wooden café table, steam rising, blurred café interior in the background"
• Tech gadget: "on a clean white desk next to a laptop, minimal workspace, soft overhead lighting"
• Fashion: "worn by a woman walking through a sunlit city street, candid street-style photography"

The setting doesn't need to be elaborate — it just needs to exist.

3. What does it look like?

This is where you control the visual style. Think of it as art direction compressed into a sentence.

Lighting keywords that work:

• soft natural light — warm, approachable, lifestyle feel
• studio lighting — clean, professional, product-focused
• golden hour light — warm, emotional, aspirational
• dramatic side lighting — bold, high-contrast, premium
• flat diffused light — even, no shadows, e-commerce clean

Style keywords that work:

• editorial product photography — magazine-quality, composed
• lifestyle photography — product in use, real-world context
• flat lay — overhead shot, organized arrangement
• UGC-style — raw, phone-shot aesthetic, authentic
• high-end campaign — luxury, polished, aspirational
• minimal and clean — white space, breathing room, modern

Color / mood keywords:

• warm neutral tones — beige, cream, tan
• cool minimalist palette — white, grey, ice blue
• vibrant and bold — saturated, high-energy
• muted earth tones — olive, terracotta, sand
• monochrome — single color family, sophisticated

4. Who is it for?

Telling the AI who will see this content shapes the output in subtle but real ways — it affects tone, styling, backdrop choices, and text phrasing.

• for Instagram Reels targeting Gen Z women → more casual, trend-driven, vibrant
• for a LinkedIn carousel aimed at CMOs → more composed, data-oriented, clean
• for a Facebook ad targeting parents of toddlers → warmer, softer, relatable
• for a TikTok ad for fitness enthusiasts → high-energy, dynamic, action-oriented

You don't need to be granular — even a rough audience note improves output.

5. What format is this?

Tell the AI what you're making. An Instagram feed post, a Story, a product-comparison carousel, and a video script all need different treatments.

• Instagram feed post → square composition, bold headline, clear product focus
• Instagram Story / Reel → vertical, text overlays, fast-paced
• Facebook ad → slightly more text-friendly, benefits-led
• LinkedIn post → professional tone, insight-driven
• Avatar video script → conversational, spoken rhythm, 30–60 seconds
• Product showcase carousel → multiple angles or benefits, slide logic

Putting it all together: the prompt formula

Here's a template you can adapt for most use cases:

[Product description] + [Setting/environment] + [Visual style and lighting] + [Mood/color palette] + [Audience and platform]

Example prompts by industry

Skincare / beauty:

"A 30ml glass dropper bottle of hyaluronic acid serum on a marble vanity counter, soft morning light, dewdrops on the bottle surface, clean beauty editorial style, pastel and white tones. Instagram feed ad targeting women 25–40."

E-commerce / fashion:

"A pair of handcrafted tan leather sandals arranged on sun-bleached wooden boards, overhead flat-lay with dried flowers and a linen cloth, warm Mediterranean summer feel, lifestyle product photography. Facebook ad for women 30–50."

Food and beverage:

"A cold-brew coffee in a tall glass with ice, condensation on the glass, sitting on a dark wood bar counter, moody café lighting, warm amber tones, artisan coffeehouse aesthetic. Instagram Story."

SaaS / tech:

"A clean dashboard UI mockup on a MacBook screen, minimal white desk, a small plant to the side, soft diffused overhead light, modern and professional, tech startup visual style. LinkedIn ad for small business owners."

Fitness:

"A shaker bottle of berry-flavored pre-workout powder on a gym bench, chalk dust in the air, dramatic side lighting, high-energy athletic mood, bold saturated colors. TikTok ad for gym-goers 18–30."

Pet products:

"A golden retriever wearing a teal adjustable harness, sitting on green grass in a park, natural daylight, warm and joyful mood, candid lifestyle photography. Instagram Reel for dog owners."

Prompting for video: what's different

When you're generating an avatar or UGC video, the prompt serves a different purpose — it's directing a performance, not composing a photograph. The rules shift.

Write your prompt as a script, not a description

For video, the prompt becomes what the avatar says. This means it needs to sound like spoken language, not marketing copy.

Weak video prompt:

"Create a video about our new protein powder that highlights the benefits and features and explains why customers should choose it over competitors."

Strong video prompt:

"Hey — if you've been looking for a protein powder that actually dissolves in water without clumping, this is it. Twenty-five grams of protein, no artificial sweeteners, and it doesn't taste like chalk. I've tried a lot of them. This one's different. Link in bio."

The second version is a script. It has rhythm, personality, and a clear call to action. The avatar will deliver it naturally because it reads naturally.

Video prompt guidelines

• Write in first person. "I use this every morning" not "Users enjoy this product."
• Keep it under 120 words for a 30–45 second video. Longer scripts produce flatter, more monotone delivery.
• Use short sentences. 8–12 words each. The avatar pauses at periods — use that to create natural rhythm.
• Use contractions. "You'll," "it's," "don't" — not "you will," "it is," "do not." Contractions sound human.
• Spell out numbers and symbols. Write "fifty percent off" not "50% off." Write "twenty-nine dollars a month" not "$29/mo." The voice engine reads text literally.
• Spell brand names phonetically if they're unusual. Write "New-Zah Cream" instead of "NU-XA Cream" if that's how you want it pronounced.
• End with a clear CTA. "Link in bio," "Try it free," "Tap the link below." Don't trail off — finish strong.

What NOT to put in a video prompt

• Stage directions: "The avatar should smile here and then look to the left" — the AI doesn't interpret directions, only speech.
• Formatting instructions: "Make this bold" or "add a text overlay that says…" — the prompt is the script, not a design brief. Use the editor for overlays.
• Multiple speakers: "First the woman says… then the man says…" — each video has one avatar. Generate separate videos for multi-speaker content.
• Timestamps: "At 0:15, show the product" — the AI doesn't process timecodes.

Common prompt mistakes and how to fix them

These are the patterns behind most "the AI didn't give me what I wanted" complaints:

"The output looks generic"

Cause: Your prompt was too vague. "Create a post for my business" gives the AI nothing to work with.

Fix: Add at least three specifics: product description, visual style, and audience. Generic input produces generic output — always.

"It used stock images instead of my product"

Cause: The AI defaulted to its image library because it wasn't told to use yours.

Fix: Upload your product images to your brand library first. Then add "use only my uploaded product images, do not use stock or AI-generated product imagery" to your prompt. Be explicit.

"The content doesn't match my brand"

Cause: The AI doesn't know your brand unless you tell it — in words.

Fix: Include your brand's visual identity in the prompt: "Our brand uses navy blue and gold as primary colors, a clean modern aesthetic, and a premium but approachable tone." Better yet, set up your Brand Kit in settings so these are applied automatically.

"The AI ignored my prompt"

Cause: Usually the prompt was too long, contradictory, or buried the important instruction inside a wall of text.

Fix: Put the most important instruction first. If "use my product images" is the critical requirement, lead with it — don't bury it after three sentences of context. The AI weighs the beginning of your prompt more heavily than the end.

"Every output looks the same"

Cause: You're reusing the same prompt (or a very similar one) across generations.

Fix: Vary one element each time — different lighting, different setting, different mood keyword. Even small changes produce noticeably different output. Try: "same product, sunset lighting, outdoor patio setting" vs. "same product, studio lighting, clean white background."

"The avatar said something weird / mispronounced words"

Cause: The script contained symbols, abbreviations, or brand names the voice engine couldn't parse.

Fix: Read your prompt out loud before generating. If a word sounds awkward when you say it, the avatar will struggle with it too. See our avatar video quality guide for detailed pronunciation tips.

Advanced: prompt layering for carousels and multi-post campaigns

If you're generating a series of related posts (a carousel, a campaign, a week of content), use a consistent "base prompt" with one element varied per post.

Base prompt:

"Premium vitamin C serum on a white background, clean beauty editorial style, soft studio light, warm tones."

Variations:

• Post 1: base + "close-up of the dropper with golden serum visible, focus on texture"
• Post 2: base + "bottle placed next to fresh orange slices and green leaves, natural ingredients angle"
• Post 3: base + "before-and-after skin texture comparison, split-screen layout, educational tone"
• Post 4: base + "woman applying the serum to her face, lifestyle photography, morning routine"

This produces a visually cohesive series where each post feels related but distinct — exactly what a real campaign looks like.

The prompt checklist

Before you hit Generate, run through this:

• [ ] Did I describe the product specifically (not just "my product")?
• [ ] Did I include a setting or environment?
• [ ] Did I specify the visual style (editorial, lifestyle, flat lay, UGC)?
• [ ] Did I mention lighting?
• [ ] Did I say who the audience is or what platform it's for?
• [ ] For video: is the script written in spoken language with short sentences?
• [ ] For video: are numbers spelled out and brand names written phonetically?
• [ ] Did I say "use my uploaded product images" if that's what I want?
• [ ] Is the most important instruction at the beginning of the prompt?

Three prompts to try right now

Copy these and adapt them to your product. They're designed to show the range of what's possible:

1. The hero product shot

"A [your product] centered on a [surface material] surface, [lighting type] lighting, [color palette] tones, editorial product photography, premium and aspirational, designed as an Instagram feed ad for [your audience]."

2. The lifestyle-in-context shot

"A person using [your product] in [realistic everyday setting], candid lifestyle photography, warm natural light, [mood — joyful / calm / energetic], relatable and authentic feel, for a Facebook ad targeting [audience]."

3. The 30-second avatar video

"Hey — I've been using [product] for [timeframe] and here's what I noticed. [One specific benefit, stated conversationally]. [Second benefit]. If you've been thinking about trying it, honestly just do it. [CTA — link in bio / tap below / use code X]."

Need help writing your prompt?

If you're stuck or not getting the results you expect, our team can help:

• In-app chat: Click the chat icon in the bottom-right corner of your Predis dashboard
• Email: hello@predis.ai

Share your prompt, the output you got, and what you were hoping for — we'll help you rewrite it.

Updated on: 18/04/2026