Previous
PhotoRobot presents how to engineer AI prompts to generate custom product photo backgrounds for robotically captured imagery.
Using AI in the creation of product photo backgrounds is one way to enhance PhotoRobot robotically captured photography. While PhotoRobot is able to automatically remove the background from product photos, AI tools can replace backgrounds with brand-accurate flair. Take for example replacing the precise background removal of PhotoRobot with a background that visually showcases the product’s reputation.
It could be a color scheme that complements the brand itself, or a full 3D scene hosting the product. The background may be a white veined marble, a deep purple-red velvet, ruby-red silk, or other luxurious textures. Better yet, some product backgrounds may illustrate the item’s actual components, like ingredients specific to a perfume.
In this case, AI tools can drastically save time sourcing & aggregating all relevant product information – especially if it’s not on-hand. Teams can then use this information when engineering AI image prompts to generate backgrounds that are brand and product specific. Why not see for yourself below? Find out how PhotoRobot-powered studios are leveraging AI tools to enhance product backgrounds, and their integration into studio production workflows.
Producing great product photos in less time and with less effort remains the cornerstone of the PhotoRobot mission. This is true even when AI can create photorealistic product images purely from text prompts. The starting point remains a real, high-quality product photo. AI can then enrich the story around it.
Thus, advances in modern AI only expand the PhotoRobot toolbox. The technology also smoothly integrates into automated photography workflows. With advanced prompt engineering, AI can accelerate in-studio product flows, and enhance real product images which we robotically capture. Robotic capture ensures the high-quality essential to product imagery. It also functions to produce photos that are more compatible for enhancement with image generation tools.
For example, using LED lighting with low CRI produces photos where part of the color spectrum is missing. This leads to critical issues for AI image generators, which cannot recreate what is not there. However, PhotoRobot ensures the perfect lighting, background removal, and post-processing of photos to efficiently run through AI. The resulting images are then optimal for additional enhancements, like background swaps or full 3D scene engineering.
Why take real photos at all when AI can generate the product images? Sure, for some product marketing, real photos aren’t always necessary. However, deeper inspection of AI-generated images will often reveal flaws. This is why they tend to be useful only in limited applications.
The most common issues that occur with AI product photos are odd typography, and minor detail errors. It is not as bad as seeing hands with six fingers, but still noticeable.
AI commonly distorts dimensions, while failing quality expectations, or sometimes mismatching advertising and the real goods. This can raise both ethical and legal concerns, supporting the case to maintain investment in real product photography.
For many companies, producing product photos on pure white backgrounds or transparent backgrounds remains suitable. In fact, the majority of PhotoRobot customers require only its precise background removal.
However, some product lines simply call for higher quality product imagery. Think of designer & luxury brands with exceeding reputations – Armani, Apple, Louis Vuitton, Rolex. Companies like these will require professional photos for print in magazines, and advertising on billboards. That, as well as images for online ads & product pages. In all cases, the item must remain the center of focus in photos. However, the background can also function to draw attention to the advert, and to distinguish a product from the competition’s.
The background can match the color scheme of the brand itself, or emphasize an item’s material, texture, and design. Take for example adding shadowy tones to the background to illuminate silver, gold, and other bright or reflective products. Backgrounds like these are often popular in photos of designer wrist watches, sunglasses, jewelry collections, and other luxury goods. Although, the primary aim of the background remains to complement the item, and not distract away from it.
In most cases, the goal of product imagery is to transfer a real object into the digital world. At the same time, the object should remain true-to-life, informative, and eye-catching.
Large brands often do this by investing in highly advanced 3D product models and 3D model rendering pipelines. This way, assets become more immersive, while the items are also easy to place on any type of product background. 3D visualization also allows companies to demonstrate interchangeable, moving, or interactive product configurations. Take for example the embeddable 3D models of PhotoRobot in use with 3D model hosting platforms, such as our long-time Emersya.
Even so, expert eyes can tell the difference between a real photograph and a 3D rendering. The same is truer for completely AI-generated images. Authenticity is simply lacking, sometimes in various aspects that the human eye easily discerns. This limits viability in some instances. Although, it does not mean 3D renders & AI image generation have no place at all in product photography.
At PhotoRobot, the goals remain the same – authentic photos with faster, simpler, and more scalable production workflows.
Despite rapid advances in AI image generators, PhotoRobot remains the faster and more reliable solution, with greater return on investment. There are no concerns in regards to consistency or quality of outputs, while trustworthiness and fidelity are a guarantee.
When using AI within PhotoRobot-powered workflows, there are a number of areas where AI excels.
For example, one use case would be photographing a collection of perfumes for a client. However, imagine that the studio has only the products on-hand, with limited product information. This is when AI prompts can easily fetch relevant data, automatically catalog it, and provide structured metadata on items.
Studios can then attach the data to the client’s images, and use the information when replacing PhotoRobot’s precise background removal. It might be to create a background that is more representative of a customer’s brand, or of the product by its reputation.
For demonstration, the following is a real-world case study photographing a series of Armani Privé perfumes in PhotoRobot Studio. The actual flacon of perfume is available in the studio, but there is no detailed metadata with the product.
In this case, an AI prompt can aggregate the relevant product information into a structured dataset for review. Moreover, it is possible to fetch data on every item within the complete fragrance collection.
The prompt can fetch the fragrance name, collection name, and an EAN code for each item. It can then include instructions to create the data in two formats, for example: a plain TXT file, and a structured CSV table.
To fetch a product list, we first prompt AI by describing the project. The prompt should then also specify the information to retrieve, and how to format the results. (Note: The following example AI prompt engineering and real outputs are from May 2025. Keep in mind that output will vary across different platforms, and as the technology evolves alongside PhotoRobot workflows.)
The prompt, “fetch the product list”:
I am building a structured dataset of perfumes for use in a product photography and AI automation environment.
Please generate a complete Fragrance Collection Overview of the Armani Privé perfume line, grouped by collection (e.g., Les Eaux, La Collection, Les Terres Précieuses, Les Mille et Une Nuits, Kogane Collection, etc.).
For each perfume, provide:
1. Fragrance Name
2. Collection Name
3. EAN Code – the international barcode for the standard 100 ml bottle
Output the result in two formats:
- A plain, readable TXT file listing, grouped by collection (for human reference).
- A structured CSV table with columns: Collection, Fragrance, EAN.
- Prepare the files for direct download.
Only include perfumes that exist in the official Armani Privé line. If multiple EANs exist for a fragrance, provide the standard 100 ml version (or the closest available).
Do not include marketing language or descriptions — only use structured, factual data.
The above prompt provides both a plain TXT file and structured CSV table. It includes a structured overview of the complete perfume collection, with names, groupings, and EAN codes:
This saves hours of manual work for the studio. That, or unnecessary back-and-forth and sometimes delayed communications between the studio and customer or supplier.
After capturing images with PhotoRobot – with clean backgrounds and optimal lighting – it’s then possible to enhance them further with AI. For this, PhotoRobot seamlessly integrates PhotoRoom via API into PhotoRobot’s control system. This allows for:
Taking it a step further, visual storytelling is possible through the product background in a number of ways where AI can assist. Take for example visualizing the key fragrance ingredients around each perfume bottle.
Finding visualizable ingredients specific to each perfume requires a more descriptive AI prompt. The prompt must ask for results to include key notes, visual themes, and design elements for each item. This information will help in later prompt engineering to generate background images that are accurate to the brand and product.
Take the following prompt for example. We start by describing the project, and attaching the output CSV from the first prompt.
I am preparing a detailed dataset for building a mood board or artistic representation. The dataset must provide structured data to generate visual representations of perfumes using AI. Please provide a detailed CSV table for the perfumes in the following file:
- 2_armani_prive_overview_ean.csv (the output from prompt 1)
Select perfumes only in the dataset:
- La Collection
For each perfume, create the following columns:
1. Fragrance – The name of the perfume
2. Top Notes – Tangible, visualizable ingredients (e.g. flowers, resins, peels)
3. Heart Notes – Tangible, visualizable ingredients
4. Base Notes – Tangible, visualizable ingredients
5. Visual Themes – A short phrase describing the atmosphere and textures the perfume evokes (for artistic use, e.g. “stone walls, golden light”)
6. Bottle Design – A detailed description of the perfume bottle: color and material of the body, shape, color of the cap, and label
Also, keep all ingredients and design details clearly worded for use in image generation. Take for example: resins, woods, herbs, spices, flowers, fruits, leaves, roots, smoke, or textures – e.g., dry, mineral, creamy. Exclude abstract terms like “elegant”, “sophisticated”, or “sensual”. Focus on concrete visual elements like “black glass”, “gold plate label”, “ivory stone cap”, etc.
Additionally, briefly list the main visual themes or textures the perfume evokes (e.g. "golden glow", "stone walls", "church incense", "earthy forest", etc.) — anything useful for background styling or setting a graphic mood.
Prepare a CSV structure that will later be used to generate visual prompts for AI image models like DALL·E. Please format the output clearly and in full.
The prompt above results in a detailed ingredients table to specification in CSV file format.
For example, the results of the prompt include the following for the first perfume.
The second perfume then has its own results which are specific to the item.
This structured data on all perfumes in the collection will provide the information necessary to start crafting visual prompts.
With the visualizable ingredients list, the next stage is engineering the visual prompts for image generators. For this, prompting AI can produce a new column “Visual Prompt” for each different perfume in the CSV. However, this requires very detailed instructions within the new prompt. This begins first with uploading the visualizable ingredients list, and then describing the project. The prompt must then include multiple layers of specific commands. Commands cover the prompt requirements, common errors to avoid, restrictions such as in typography, and output as well as quality expectations.
The first layer of the prompt attaches the CSV file for analysis, and provides general instructions on the task.
You are provided with a CSV file containing structured data about perfumes from the Armani Privé La Collection. Each row includes:
- Fragrance (name of the perfume)
- Top Notes (clearly visualizable ingredients)
- Heart Notes (clearly visualizable ingredients)
- Base Notes (clearly visualizable ingredients)
- Visual Themes (atmosphere and textures the perfume evokes)
- Bottle Design (material, color, shape, label, and cap)
- EAN (used as the name of the image file)
Your task is to generate a new column called "Visual Prompt" that contains a full and direct prompt for AI image generation tools (e.g., DALL·E or Midjourney).
The second layer of the prompt identifies the requirements for each new item in the new column of the CSV file.
Each prompt should describe how to transform a product photo of the perfume (named {EAN}.jpg) into a final image with the following properties:
Thirdly, the prompt names specific restrictions, and common errors to avoid.
Do not mention the CSV, or describe the structure. Write each prompt as if addressing the AI directly to generate the image for that perfume.
The result should appear premium, atmospheric, and true to the fragrance identity. It should be indistinguishable from a professionally retouched editorial photograph, but fully AI-generated. The viewer should not be able to tell the image is synthetic.
Also, do not mention or show artificial generation. The image must look authentic and photorealistic.
The fourth part of the prompt shares instructions when working with these specific perfume bottles. Typography is a common issue for AI, so it’s crucial to provide very clear instructions on label designs, branding, and styling.
Pay special attention to the design of the front label on the bottle and its graphics accuracy. The gold plate must include the following exact text, as the original image, centered and aligned as on the real product.
- The slash symbol (" / ") between ARMANI and PRIVĒ is slightly taller than other letters and subtly stylized. It starts slightly below other characters, and ends slightly above the other characters, as on the original image.
- The character "Ē" in PRIVĒ must have a clearly visible horizontal accent mark, while the letter including the accent mark is the same height as other letters. There is a flat horizontal line above it (not an acute line). The line must be the same width as the E below it, not slanted. It must not resemble an É. This is not a diacritic or an accent – it is a flat macron (horizontal bar). In other words, the horizontal line on Ē must resemble a short flat line, like a hyphen, placed precisely above the E. It must not be diagonal like in É.
- Match the exact label design from the reference product photo.
- The label must be identical in typography, spacing, and accents. The label must be the same visual style as the original image, as it is crucial to the brand identity.
- The typography must be accurate and not estimated or replaced. Caution: the typography may be changed for a single character, so follow the details for each character individually.
- Do not change, shorten, or paraphrase any part of the label.
The final layer of the prompt continues on the expectations for each visual prompt, and provides instructions for the new CSV.
The bottle plate must retain its proportion, surface finish, and embossed print look under soft lighting.
This label is brand-critical – treat it with the same visual fidelity as a logo or trademark.
The bottle shape is brand-critical – treat it with the same visual fidelity as a logo or trademark.
Do not alter the text or approximate the type – treat this label as a brand-critical design element that must be accurate and sharp.
The label must retain its real-life proportions, texture, and gold finish – it is slightly embossed with a soft satin sheen under soft light.
Save the result in a new CSV file with all original columns plus the new "Visual Prompt" column.
In the end, the resulting CSV table has the complete list of perfumes, names, EANs, visualizable ingredients, and visual prompts. The visual prompts contain full and direct prompts for AI image generation tools like DALL·E and Midjourney. These will help to create custom backgrounds and scenes that creatively complement the real photos of the perfume bottles.
After creating the visual prompts for each item, your favorite AI image generator can do the rest. All it requires is uploading PhotoRobot-captured images, and inputting the visual prompts from the CSV to create custom backgrounds. The generator will render the background according to the prompt engineering, and for rendering in different styles.
Meanwhile, PhotoRobot product images with precise background removal make it easy to swap backgrounds in and out. If one does not match perfectly, your quality assurance teams can quickly create one that works. That, or prompt the AI generator to adjust outputs until satisfactory.
Finally, if pushing the limits of AI background generation, even full 3D scene rendering is possible. This goes well beyond more straightforward background swaps, however. Imagine displaying a fantastical 3D environment featuring brand-accurate scenery in addition to the key ingredients. Accomplishing this requires a much more ambitious prompt.
To generate a full 3D scene for one of the perfume bottles, another sophisticated prompt is necessary. It must take into account the scene composition, visual themes, atmospheric elements, color palettes, lighting and more. Take the following prompt for example.
After uploading a product image into the AI, start the generator prompt by listing all requirements for the background scene. This will include the information for the product from the visualizable ingredients list and visual prompts.
Generate the entire scene, including background, ingredients, textures, and artistic lighting in harmony with the bottle design.
Scene Composition:
Build an elegant, editorial-style environment around the bottle using:
Add atmospheric foreground effects such as smoke or mist if a part of the notes, partially overlaying the bottle for realism. Maintain visual balance, depth, and refinement.
Preserve a front-facing, studio-style perspective and camera angle.
Next, specify the critical instructions for graphic accuracy of the label and bottle. These are the same commands as in the visual prompts for the appearance of each individual item. The instructions regard accuracy of typography, label design, graphics, and original photo use.
Finally, prompt the AI with all final image requirements from the individual item’s visual prompt. This includes copying the same instructions as before on the proportions, finishes, embossed print, and lighting. These list the specific quality requirements for the label, bottle shape, text, typography, and additional design elements. Ultimately, the final output should take shape as a full 3D scene behind the item, which remains the center of focus.
Judge the resulting 3D scene for the perfume’s product background yourself.
Note: In this case, there is no way to fully integrate the real photograph into the 3D world. It is necessary for the AI generator to instead repaint the item digitally to place it within the 3D scene. In this case, there are various limitations, such as no true multi-layer composition like in Photoshop. Also, typographic issues remain with complex characters. Nonetheless, issues like these will not always persist, and may resolve sooner rather than later as the technology progresses.
In essence, the fusion of automated photography and AI tools can dramatically enrich the customer experience across your portfolio. Although the foundation remains a real-life photograph of high quality, AI can expand the storytelling around it. The technology supports thematic visualization, and can serve to highly accelerate photo studio workflows. It enables rapid information sourcing and synthesis, automatic cataloging, and effective background swaps (with knowledge of prompt engineering). To learn more, the PhotoRobot team is always ready to help businesses realize their creative vision. Just ask how we might help. Your project might even feature in future blog posts - if not a closely guarded secret workflow, of course!
