Higgsfield prompts and templates: 15 starter prompts organized by business job

Most prompt pages for Higgsfield, the AI image and video platform, are written for film fans. This one is organized by the jobs a small business or agency actually has: a product photo, a social ad, a job-site scene, a property exterior. I wrote the templates as starting points. Treat each as a draft to test on your own product, and expect to change it. The structure comes from published prompting advice, including Higgsfield's own guides on references and consistency, and from common sense about what an AI image model can and cannot hold.

This page may contain affiliate links from which Blur Studio earns commissions.

How a good prompt is built: the brief formula

Think of a prompt as a brief to a photographer. Put these parts in this order, in short plain sentences:

1. Subject: what is in the frame.
2. Key detail: the one feature that must stay true, such as a label, a color or a shape.
3. Action: what happens. One action only.
4. Setting: where it is.
5. Camera: one named move or angle, such as a slow push in, an orbit, or an eye-level shot. Words like "cinematic" tell the model very little.
6. Light and mood: soft window light, warm evening, clean studio.
7. Constraints: what must not change or appear.

Two habits matter more than any wording trick. First, do one job per prompt. Make the still image first, then animate the winner. Keep identity changes out of video prompts. Second, change one thing at a time when you try again, so you know what helped.

Before you start: five things to decide first

• Aspect ratio. This is the shape of the frame. Pick 16:9 (wide), 9:16 (tall, for phones) or 1:1 (square) before you write. The framing changes with it. To reshape a photo you already have, see resizing one photo for story, feed and banner.
• Reference. Higgsfield's own consistency guide (27 July 2026) says a text description gives the model almost nothing concrete to hold for a place, and only a rough description for a face, and that a reference image works better. Attach your real product photo or location image when you can.
• Cheapest first. Images cost far less than video. Higgsfield's pricing table lists image models from a fraction of a credit up to about 4 credits an image, and video clips from about 1 credit up to about 110 credits, depending on model, length and resolution. A credit is the unit Higgsfield charges. The exact cost shows on the Generate button before you confirm. See Higgsfield pricing.
• Text. Do not ask the model to draw a phone number, price or legal line. Broken letters are the most common thing you cannot repair. Add text afterward in Canva (a design tool) or your video editor.
• Who and what is in it. Do not name real people or real brands. Higgsfield and its model providers can block copyrighted characters and brands, and you are responsible for rights and consent.

Prompt length limits differ by tool and model. I could not confirm a figure on a Higgsfield page, so let the box on screen tell you.

The guard line to add to almost everything

Paste this at the end of product, property and trade prompts:

Do not change the product shape, logo, colors or label. Do not add any text, numbers or watermarks. Do not show real people or real brands. One subject only.

The reason is simple. Even Higgsfield's own account of building a website says its AI agent invented a spec sheet that read as real. Treat every generated image as an illustration. Never use one as proof of a size, condition, finish or result.

15 starter prompts by business job (stills)

Replace the words in [brackets]. Keep the rest short.

Product

1. Product hero shot. A [product] on a [surface, such as pale oak table], front three-quarter view, [key detail, such as the matte black cap and white label] clearly visible. Soft window light from the left, clean [color] background, shallow depth of field (a softly blurred background). 4:5. [Guard line]
2. Lifestyle still, no people. A [product] on a [setting, such as a bathroom shelf next to a folded towel]. Morning light, calm mood, natural colors. Leave empty space on the right for a caption. [Guard line]
3. Before and after pair. Generate two separate stills from the same reference, never one image with a split. The same [room or item] with [old condition]. Same camera position, same light. Then repeat with [new condition]. Label the result an illustration.

Ads and social

4. Social ad background. A [simple scene tied to the offer, such as a sunny patio table] with an empty area across the top fifth and the bottom fifth of the frame. 9:16. No text, no logos. [mood] Add your headline and button afterward.
5. Seasonal sale backdrop. An abstract [season] backdrop in [two brand colors], soft shapes, plenty of empty space in the centre. 1:1. No text.
6. Testimonial-style scene, no real person. A [kitchen, workshop or office] with a [object related to the service] in the foreground and warm light, as if someone has just stepped away. No people. Do not invent a customer, a quote or a result.
7. Thumbnail or cover. A bold close-up of [subject], high contrast, clean background in [color], large empty area on the left for a title.

Local service and property

8. Service van and crew scene. A plain white service van parked outside a [style of house], daytime, two figures in work clothes seen from behind carrying a toolbox. No visible faces, no text on the van. Add your real logo in an editor, never in the prompt.
9. Property exterior at twilight. Start from a real photo as the reference. The same house at twilight, warm lit windows, soft blue sky. Keep the roofline, windows, door and garden exactly as in the reference. Check local rules on altered listing photos. See AI virtual staging and listing media.
10. Empty-room mood shot. A bright empty [room type] with [floor type] and large windows, morning light. Keep walls, windows and floor as in the reference. Do not add features the property does not have.

Food and team

11. Restaurant dish. A plated [dish] on a [plate color] plate on a dark wooden table, overhead angle, steam rising, soft side light. Show only the ingredients the dish really contains.
12. Team introduction scene. An empty [workplace] with [object] on a desk, morning light, calm and tidy. Use real staff photos for staff. Do not generate fake team members.
13. Explainer B-roll. Close-up of hands [doing the task, such as sorting paper invoices] on a [desk], gentle push in, soft daylight, 16:9. B-roll means extra footage that sits under the main story.

Retargeting and story

Retargeting means showing ads again to people who already visited your site.

14. Retargeting variant. Take your best still. Same subject and framing as the reference. Change only the background color to [color] and move the subject to the lower left. One change at a time.
15. Story format. The same scene as the reference, recomposed for 9:16 with the subject in the middle third and empty space above and below for captions and buttons.

Starter prompts for short video clips

Animate only a still you already approved. Attach it as the first frame.

• Product reveal (5 seconds): Start from the first frame. Slow push in toward the [product]. Light shifts gently from left to right. Nothing else moves. No text. Keep the product shape and label unchanged.
• Orbit: Start from the first frame. The camera orbits slowly a quarter turn around the [product]. The product stays recognizable for the whole clip.
• Atmosphere only: Start from the first frame. Only the [steam, curtains or leaves] move. The camera is still.

Higgsfield's own command line guide (about its text-based tool for running Higgsfield from a computer) uses the same pattern: an image first, then a short video prompt that starts from the start frame, lists a few timed beats and ends with "No text". Models have different features and settings. Do not paste a prompt between models unchanged. Test each model on one cheap shot first. Models are added and removed over time (Sora, for one, is not in Higgsfield's current model list as of 2 November 2026), so keep prompts that do not lean on one model's quirks. See the Higgsfield and Runway comparison (Runway is another company that makes AI video tools) if you are choosing between video tools.

Camera presets: pick the move, then write the rest

Higgsfield's Camera Controls page (read 2 November 2026) lists dozens of named moves you can pick instead of describing them, such as Static, Dolly In, Dolly Out, 360 Orbit, Overhead, Handheld, Crash Zoom In, Crane Up, Head Tracking and Lazy Susan. A third-party prompt library (Hooked, September 2026) summarizing Higgsfield's camera guide says the preset fixes the type of move, while your words still have to supply the speed, the start and end framing, and what the camera must not do. That matches the brief formula above. When a preset is picked, keep the text short and about the subject.

Business job	Preset to try first (my suggestion, untested)	What to still write
Product reveal	Dolly In or Static	Start wide or tight, end tight on the label, nothing else moves.
Product from all sides	360 Orbit or Lazy Susan	Speed (slow), that the shape and label stay the same.
Flat lay or food from above	Overhead	What enters the frame, and that the camera does not drift.
Phone-style testimonial feel	Handheld	Medium close-up the whole time, no zoom.
Property exterior	Crane Up or Dolly In	Keep roofline, windows and garden as in the reference.

Test the preset on a cheap draft before paying for a final render, as in the three pass method below.

If the clip has someone speaking: how to write the line

Some models can make a person speak a line. The same prompt libraries show a pattern of the scene, then a short quoted line. Use the same pattern, with these business guards:

A presenter at a [kitchen counter], medium shot, still camera, daylight from the left. The presenter holds up [product] and says: "[one sentence you wrote and can prove]". No other camera movement. No text on screen.

• Write the line yourself. A line the model invents can become a claim you cannot support.
• Keep it to one sentence. A short clip cannot carry more.
• Do not script a customer quote, a result or a before and after in an invented person's voice. See the rights section below.
• Where a presenter looks real, say it is AI-generated if the platform or the law asks for it.

How to adapt a prompt to your product in 5 minutes

Field	Your answer
Brand and product
Material and color
Setting
Mood
Camera move (one)
Aspect ratio
Forbidden items

Have your client or manager fill this in. It lets a non-expert approve the brief before any credits are spent. Then ask for 4 variations at most.

The three pass method keeps the bill small:

1. Pass 1: stills only, cheapest setting. Pass if the subject, shape and colors are right. Fail if the logo is warped or hands are wrong.
2. Pass 2: animate the winner, one camera move. Pass if the product stays recognizable for 5 seconds.
3. Pass 3: final quality for one or two keepers.

Higgsfield's own command line guide (19 October 2026) shows why: in its test on one account, one image cost 2 credits, and a 5 second 1080p Seedance 2.5 clip cost 80 credits, or 15 as a draft. That is a vendor example, not a price list. The Generate button shows your real cost. More on this in tips and settings that save credits.

Keeping a person or a place the same across shots

If you need the same room or character in several shots, use a reference, not a longer description. Higgsfield says its Soul ID (a feature that learns one person's face from photos so it stays the same across images) needs 20 or more photos of one person, and its consistency guide says Seedance 2.0 accepts up to 9 reference inputs in one call. It also says to carry your strongest frame forward as the anchor for the next shot. Only use photos you own or have permission to use, and get written consent for any real person. See Soul ID.

Review checklist before you press Generate

• One subject, one action, one camera move.
• Aspect ratio is set.
• Reference is attached.
• "No text" is in the prompt.
• Forbidden items are listed.
• The cost on the Generate button is acceptable.

Keep a prompt log as a simple sheet with these columns: date, tool and model, prompt, references used, credits spent, keeper (yes or no), reason rejected. After about 20 rows you will see which words wasted credits.

What to do when a prompt fails

• Warped logo or label: attach the real label as a reference and shorten the prompt.
• Text looks wrong: remove the text request and add it afterward in an editor.
• Too many things moving: remove actions until only one is left.
• Flagged or blocked: remove any brand, character or public figure names and describe the idea in plain words.
• Still bad after three tries: stop. Change the prompt, not the dice. Rerolling the same prompt burns credits.

Rights, likeness and facts: what a prompt must never ask for

This is information, not legal advice. Check local law and each platform's current rules. Do not ask for a real person's likeness without consent, a competitor's logo, or a result you cannot prove. Higgsfield's terms say you are responsible for having the rights, releases and consents for anything you upload, including any person's likeness or voice, and that you must disclose AI-generated output where the law requires it. Do not assume an output is unique to you.

Who should skip templates and hire a designer

• Packaging, regulated products and anything where a wrong detail could mislead a buyer.
• A brand with strict visual rules and no time to review output.
• Anyone selling a result, such as a before and after for a service, where an AI picture could look like proof.

Free prompt libraries online can be fun. Their quality is unchecked, and paid prompt packs come with no proof of results. I would not pay for one.

How I'd approach this

Start with five prompts for one real product, not fifteen. Run the three passes, keep the log, and stop when you have two images a client would accept. If you are on the fence about the platform itself, read the Higgsfield review first. If you want to start, open a Higgsfield account and spend your first credits on stills only. When a team needs these prompts turned into a shared brief form with an approval step, that is something Blur Studio builds. See how I work with business owners.