Automating with Magnific (Freepik)

Automating product content with Magnific: a practical workflow for web shops

Automating product content with Magnific starts with combining a central product database, an automation tool such as n8n or Make, and clear quality checks. Product data forms the basis for texts and prompts, while Magnific is used to enhance product images, cut them out and generate new ones. By capturing every step in a workflow, web shops can produce consistent content faster without processing every product entirely by hand.

Creating good product content takes a lot of time. Product data has to be collected, texts written, images edited, and every sales channel asks for its own formats and information. With a small range that can often still be organised by hand, but as soon as a web shop holds hundreds or thousands of products, a backlog quickly builds up.

With AI and automation, much of this work can be partly taken over. Magnific can be used as the visual production engine. Think of removing backgrounds, improving resolution, adjusting aspect ratios and generating lifestyle images. The full workflow, however, consists of more than just an image tool. You also need reliable product data, fixed templates, an automation platform and a review process.

Why automating product content is worthwhile for web shops

Automating product content mainly saves time on recurring tasks, such as structuring product data, writing basic descriptions, removing backgrounds and creating channel variants. Automation also helps present products according to a fixed structure. The aim is not to leave every creative decision to AI, but to reduce manual production work.

At many web shops, product content is scattered across supplier files, Excel documents, folders of images, emails and separate systems. As a result, staff have to gather data again and again before they can publish a product. That makes the process slow and increases the risk of inconsistencies.

Automation is especially worthwhile when a web shop regularly adds new products, variants, brands or collections. The same basic steps are then repeated over and over. By placing these steps in a workflow, a member of staff only needs to check exceptions and end results.

Common tasks you can automate are:

• Retrieving product data from a CSV file, PIM or supplier feed.
• Flagging missing or incorrectly filled fields.
• Drafting product titles, descriptions, benefits and meta descriptions.
• Removing image backgrounds.
• Enlarging and enhancing product images.
• Generating lifestyle images and different crops.
• Giving images logical file names and alt texts.
• Exporting content to Shopify, WooCommerce or a marketplace.

Not every step has to be fully automatic straight away. Start with the most time-consuming and predictable tasks. Only then add more complex parts, such as fully generated lifestyle images or automatic publication.

How an automated product content workflow is built

A product content workflow usually consists of five parts: a central data source, an automation tool, a text model, a visual AI tool and a publication channel. The automation tool connects these parts and determines which action is carried out. By also tracking statuses and error messages, it stays visible where each product is in the process.

Within this set-up, Magnific is not the system in which your full product catalogue is managed. It is the environment in which visual tasks are carried out. You can manage the product data itself in, for example, a PIM system, Airtable, Google Sheets, a database or directly in your web shop.

A simple technical set-up could look like this:

• Product source: PIM, Airtable, Google Sheets, CSV or supplier feed.
• Automation: n8n, Make or a custom integration.
• Text generation: a language model for product texts and metadata.
• Image processing: Magnific for upscaling, crops, backgrounds and visual variants.
• File storage: Shopify Files, a CDN, cloud storage or an image library.
• Publication: Shopify, WooCommerce, Bol.com, Amazon or an export file.

The workflow starts when a product gets a certain status, for example “Ready for production”. The automation then collects all the data, carries out the configured tasks and writes the results back to the central product database.

Start with a reliable central product database

An automated workflow is only reliable when the source data is complete and correct. For each product, record at least the SKU, product name, category, properties, benefits, materials, dimensions and source images. Also create fields for the desired output, processing status, error messages and approval. AI must not fill in missing product facts on its own.

Product data is the basis for both the textual and the visual content. If a material, colour, dimension or product part is wrong in the database, that error may be passed on automatically to several sales channels.

That is why it is wise to create a fixed data structure before you start automating. Wherever possible, use separate fields instead of one long text field. A language model works better with structured information, and the final output is also easier to check.

Decide which product fields are mandatory

Which fields are needed differs per product category. For virtually every web shop, the following data is useful:

• SKU or internal product number.
• Product name and brand.
• Product category.
• Short factual product description.
• Main properties.
• Benefits for the user.
• Material, colour and dimensions.
• Package contents.
• Available variants.
• Source images.
• Target audience and situation of use.

Set up the workflow so that it stops when key information is missing. The product can then automatically get the status “Complete data”, including a message stating which fields have not yet been filled in.

Use statuses to manage the production process

A clear status prevents products from being processed twice or published too early. You could work with, for example:

• New.
• Complete data.
• Ready for production.
• Processing.
• Review needed.
• Approved.
• Published.
• Error occurred.

Each status can trigger a next workflow. The status “Ready for production”, for example, starts text and image generation, while “Approved” triggers publication to the web shop.

Decide in advance which content needs to be created

Not every product needs the same amount of content. A simple accessory may be fine with three packshots and a short description, while a complex device also needs images of it in use, dimensions, benefits and comparison images.

So make it clear per product or category which assets need to be produced. Think of:

• A cut-out main image.
• Several detail images.
• A lifestyle image.
• An image with dimensions.
• A listing image with benefits.
• A social media format.
• A short and a long product description.
• SEO title, meta description and alt texts.

This lets the workflow automatically check whether all required parts are present before a product is submitted for approval.

Prevent mistakes with automatic and human quality control

Quality control is a mandatory part of automated product content. Check automatically whether required fields, files and formats are present, and then have a member of staff assess whether the product is shown correctly, visually and in terms of content. With technical products in particular, generative AI can unintentionally change parts, proportions, text or materials.

AI can speed up a lot of production work, but it does not automatically understand which parts of a product are technically essential. A generated lifestyle image can look convincing while a handle, connection, button, seam or included accessory is shown wrongly.

That risk is greater with complex products, such as fitness equipment, machines, electronics, furniture and sanitaryware. With products like these, the original product image must always remain available as a reference and the generated output must be compared carefully.

Build in automatic checks

The workflow can carry out a large number of simple checks on its own. Check, for example:

• Whether every mandatory product field has been filled in.
• Whether at least one usable source image is present.
• Whether images have the required resolution.
• Whether files have the right format.
• Whether the SKU is in the file name.
• Whether all requested assets have been created.
• Whether texts contain no placeholders or error messages.
• Whether prohibited or unproven claims appear.
• Whether the same image has not been added more than once.

When a check fails, the workflow must not continue to publish. Give the product an error status and record which check did not pass.

Organise clear human approval

The final check can be carried out in Airtable, an internal dashboard or a task management system. Have the reviewer assess at least:

• Whether the shape and proportions of the product are right.
• Whether colours and materials are correct.
• Whether logos, labels and text are shown properly.
• Whether no parts have been added or have disappeared.
• Whether the product description contains only correct information.
• Whether the lifestyle image shows realistic scale and placement.
• Whether the content fits the brand style.

Also record who checked the content and when. That keeps the process traceable if a mistake is discovered later.

Build the automation as separate workflows

A reliable automation preferably consists of several small workflows instead of one big process. Use one workflow for preparing the product data, a second for processing images and results, and a third for approval and publication. That makes errors easier to find and lets you rerun parts separately.

A common mistake is building one big automation in which dozens of actions run one after another. When one API request fails, it becomes hard to determine where the problem arose and which parts need to be processed again.

So split the process based on clear responsibilities. A practical division is:

• Workflow 1 (preparation): retrieve the product, check the data and start jobs.
• Workflow 2 (processing): receive results, download and store files.
• Workflow 3 (text content): generate, validate and write back texts.
• Workflow 4 (quality control): check whether all content is complete.
• Workflow 5 (publication): send approved content to sales channels.

For every external job, store the task number, timestamp and associated product ID. That way a result can always be linked to the right product.

Create content for multiple sales channels from one source

A central workflow makes it possible to convert the same verified product information for different channels. Use fixed templates for Shopify, WooCommerce, Bol.com, Amazon, ads and social media. Adjust the length, aspect ratio and density of information per channel, but keep the same product facts and brand look.

A web shop, a marketplace and a social media channel each make different demands on product content. A Shopify product page can contain an extensive HTML description, while a marketplace works with fixed bullet points, limits and mandatory specifications.

So do not create one universal text that is placed everywhere. Let the workflow create different versions from the same source data. The underlying facts stay the same, but the structure is tailored to the channel.

For channel variants, think of:

• A short product title for marketplaces.
• A more extensive SEO title for the web shop.
• A short introduction at the top of the product page.
• A longer HTML description with subheadings.
• Bullet points for Bol.com or Amazon.
• Alt texts for web shop images.
• Square images for product overviews.
• Portrait images for social media.
• Wide images for banners and additional product content.

Use a separate output field for each channel. That prevents a change for one platform from unintentionally overwriting other content.

Process product images in the right order

A consistent image workflow starts with the best available source image. First remove the background, save the cut-out product permanently, and only then improve the resolution or composition. Then use this verified master image for different formats and lifestyle images. That prevents every channel variant from being generated from a different source.

The order in which image edits are carried out affects the end result. If you first apply a creative edit and then try to cut out the product, incorrect details may already have been added.

A practical order is:

• Select the best original product image.
• Check whether the product is fully visible and sharp.
• Remove the background.
• Save the transparent version straight away in permanent storage.
• Improve the resolution with a conservative upscaler.
• Create a square master image with enough white space.
• Generate any lifestyle images from this master image.
• Then create channel-specific crops and formats.

Use a conservative setting for primary product images. Creative enhancement can add new details that are not present in the original product. That is sometimes acceptable for a campaign image, but not for a main image on which a customer bases their buying decision.

Also save generated files straight away in your own image library, web shop environment or CDN. Temporary results from external image processing must not be used directly as the final product image.

Work with templates per product category

Templates keep product content consistent without making every product identical. Define per category which texts, images, benefits, specifications and formats are needed. Also use fixed prompt blocks for product preservation, composition, lighting, brand style and prohibited changes. That makes the output more predictable and easier to check.

A piece of furniture calls for a different presentation from a food product or a piece of fitness equipment. When all products get the same general prompt, the results are often irrelevant or illogical.

So create a content template per product category. In it, you decide which information and images are needed. For a chair, for example, you could work with a cut-out image, a detail image of the fabric, a lifestyle image in an interior and an image with dimensions. For an electronic product, connections, controls and included accessories may be more important.

A visual prompt template can consist of several fixed parts:

• Product preservation: which parts must not be changed?
• Environment: in which setting should the product be placed?
• Composition: how big and where should the product be in the frame?
• Lighting: which light source and mood suit the brand?
• Camera angle: from which angle is the product shown?
• Brand style: which materials, colours and look are wanted?
• Exclusions: which parts, text or accessories must not be added?

Save these prompt parts as separate fields or variables. That lets you change one part without rewriting the full prompt for every product.

Step-by-step plan to automate product content with Magnific

Start with one product category and a limited number of fixed outputs. First get the existing product data in order, then build a workflow for text and images, and finally add quality control and publication. Test the workflow with representative products and record errors. Only when the results are structurally reliable should you expand the automation to more categories and channels.

Step 1: centralise the product data

Gather all the information needed to present a product. Move this data to one fixed database, spreadsheet or PIM environment.

Distinguish between verified source data and generated content. A product dimension from a supplier file, for example, is source data. A commercial description of that dimension is generated content. Never let generated text automatically overwrite the original product data.

Step 2: validate the input before the workflow starts

Check automatically whether all mandatory fields are present and whether image links can be reached. Let the workflow stop when a SKU, main image, material or other key property is missing.

Record clearly what needs to be added. That way a member of staff does not have to investigate why a product has not been processed.

Step 3: generate structured text content

Have a language model convert the product data into predefined fields. Do not ask for one long story, but for structured output, such as:

• Product title.
• Short description.
• Long HTML description.
• Three to five benefits.
• Specifications.
• SEO title.
• Meta description.
• Alt texts.
• Marketplace bullet points.

Explicitly instruct the language model to use only information from the source data. Have missing information come back empty instead of letting the model fill it in itself.

Step 4: process the product images

Send the chosen source image to the right image editing steps. Remove the background, improve the resolution and create a verified master image.

Then generate only the variants that were defined in advance. Start, for example, with one cut-out product image, one lifestyle image and one social media format. A limited number of usable assets is worth more than ten variants that all have to be corrected by hand.

Step 5: check and publish the results

Put all output back into the central product database and change the status to “Review needed”. Have a member of staff assess the texts and images and only change the status after approval.

The publication workflow may only process products with the status “Approved”. Look up the existing product by SKU, fill in the right fields and metafields and add the final images. After publication, record the product ID and publication date.

Common mistakes when automating product content

Many automation projects fail because incomplete product data is linked straight to AI, too many steps are automated at once or human review is skipped entirely. Other mistakes are using temporary image links, unclear prompts, missing error handling and automatically overwriting existing product content.

Automation does not automatically make a poorly organised process better. When product data is supplied inconsistently, the workflow mainly produces inconsistent content faster. So start with the data structure, not with the AI tool.

Another mistake is trying to generate every product image fully automatically straight away. AI struggles in particular with products where exact construction, connections, logos or proportions matter. In that case, use AI for supporting images first and keep original photography or 3D renders as the technical foundation.

Also avoid the following mistakes:

• Using one general prompt for all product categories.
• Letting AI make up product properties.
• Not distinguishing between source data and generated content.
• Publishing all results automatically.
• Not storing task IDs, error messages or statuses.
• Placing temporary image links in the web shop.
• Overwriting existing content without a version or backup.
• Not setting cost limits or maximum quantities.

Start small and measure how many products get through the workflow without corrections. When a particular step regularly causes errors, first adjust the input, prompt or check before you scale up further.

The role of AI, automation and human creativity

AI and automation are especially suited to repeatable production tasks, while people remain responsible for positioning, brand look, factual accuracy and creative choices. The best workflow does not replace the entire content team, but means staff spend less time gathering files, adjusting formats, copying fields and rebuilding similar content.

Product content has not only a functional but also a commercial role. The content must make clear why a product is relevant, how it is used and how it differs from alternatives. These choices cannot be based solely on technical product data.

Human input remains necessary for the brand strategy, visual direction, main selling points and assessment of the final quality. AI can then apply these choices faster to large numbers of products.

For web shops with lots of SKUs, the biggest gain comes when fixed formats, product data and automation are combined. A designer or content specialist then no longer creates every product entirely from scratch, but develops systems and templates with which large parts of the range can be processed consistently.

BrandMonks helps brands and web shops with product content, AI image production, listings, automations and scalable content systems. That means new products can be published faster without losing sight of quality and brand consistency.

Frequently asked questions about automating product content with Magnific

Build a scalable product content process step by step

Automating product content with Magnific takes more than linking an image to an AI tool. A reliable workflow starts with structured product data, fixed templates, clear statuses and a controlled publication process. Magnific can then speed up a large part of the visual production.

Start with one product category and a limited number of outputs. For example, first automate the short product description, the cut-out packshot and one lifestyle image. Then measure how many results are usable straight away, which errors keep coming back and where human correction is still needed.

Only expand the workflow once the foundation works reliably. Then add extra sales channels, languages, image formats and content types. That way you do not end up with a loose collection of AI tools, but with a scalable content system that lets you publish new products faster, more consistently and in a more controlled way.

:::

The technical principles have been checked against the official documentation: Magnific supports creative and conservative upscaling, reference images and asynchronous processing; with background removal, temporary results must be moved to your own storage in good time.

Melle Sikma
Helmsman
melle@brandmonks.nl