Scaled ECommerce Content Creation with AI and Computer Vision
Challenges
Manual product editing made it difficult to create hundreds of colour variations at scale. Fashion products required separate flat-lay photoshoots while maintaining garment-level accuracy. Campaign video and audio production was time-consuming and slowed content delivery.
Outcome
Product image processing time reduced from ~10 working days to 1–2 days. The workflow delivered a 5x improvement in processing speed for images. AI-enabled video and audio generation also accelerated campaign content production.
Solution
AI Powered eCommerce Content Generation
Challenges
Solution
Technology Stack
Outcomes
For a large retail e-commerce business, creating product content at scale involved several teams and several manual steps. Products had to be photographed, edited, checked for colour and texture accuracy, approved and then prepared for the website. Fashion products required separate model shoots and flat-lay photography. Marketing ad campaigns added another layer of video and audio production. The client wanted to reduce this effort by leveraging AI without lowering the visual standards expected from a major retail brand. Cloudaeon worked closely with the client’s marketing and imagery teams to build an AI-led content generation approach covering three areas: furnishing colour variations, fashion flat lays and ad campaign video and audio. The objective was to automate repetitive content creation, produce assets at scale and keep the product details accurate enough for customer-facing use.
Challenges
Creating hundreds of product colour variations
The furnishing product portfolio included bedsheets, pillows, blankets and other products, with around 420 to 450 products across the range. Each product required multiple colour variations. Previously, an editing team had to recolour these images manually for the website. This process was time consuming and required effort. The challenge was not simply changing the colour. The new image still had to retain the original fabric texture, folds, lighting, contrast and sharpness. Poor recolouring could easily flatten the texture or make the product look unnaturally dark.
Creating flat lay shots needed another photoshoot
Fashion product imagery required a different workflow. The client already had model photographs, but separate flat lay photography was still required. Each garment had to be arranged on a clean background, photographed again, edited and approved. Here, accuracy went beyond colour. The generated garment also had to retain its construction. Buttons, zips, cuffs, proportions, fabric texture and other small details all mattered.
Producing Ad Campaign videos and audio faster
Marketing campaigns also involved significant content production. For campaigns such as Father's Day or Christmas, the marketing team had to turn a campaign idea and script into finished video and audio assets. The eCommerce giant wanted to use generative AI to shorten this production process.
Root Cause Analysis
Cloudaeon did not jump directly to an AI model and start generating images. We first examined what made the existing manual process work. For bedding, the real challenge was not colour replacement. It was changing the colour without changing everything around it. The fashion products had a different root problem. Removing a model from an image was not enough. The system had to recreate the garment as a standalone product image while preserving its actual construction. This analysis made one thing clear. The three content-generation requirements could not be solved using one generic AI approach. Cloudaeon selected the right combination of computer vision and generative AI for each requirement.
Solution
Cloudaeon built an AI-led content generation approach covering product imagery and ad campaign content. For furnishing products, we created a computer vision pipeline that segmented the product and recoloured it while preserving texture, lighting and contrast. For fashion products, we used generative AI models to convert model photographs into clean product flat lays. For marketing campaigns, AI was used to generate both audio and video assets from campaign scripts.
How We Delivered
Segmenting the product before recolouring: Cloudaeon used the Segment Anything Model through Segmind to isolate the product and generate a mask. This recoloured only the target product rather than manipulating the complete image.
Using reference colours from the client: The client provided the target colour references. Once the product was masked, computer vision techniques applied the required colour to the selected area. Rather than treating recolouring as a simple colour replacement exercise, the pipeline was designed to preserve the product's original photographic characteristics.
Preserving texture and lighting using LAB processing: The recolouring process used the LAB colour space. This allowed the team to separate colour information from other visual characteristics within the image. Rather than manipulating every part of the image, the pipeline changed the required colour component while retaining information associated with lighting, texture and contrast. This helped protect details such as folds, fabric texture and sharpness.
Comparing outputs against manually edited references: AI-generated or computer-generated outputs were not accepted automatically. Cloudaeon and the organisation’s marketing teams compared generated outputs against these references for:
Colour accuracy
Fabric texture
Lighting
Product folds
Sharpness
Overall visual quality
The outputs went through a feedback cycle with the client’s marketing and imagery teams. This helped ensure that automation did not come at the expense of the standards required for an e-commerce website.
Generating fashion flat lays from model images: For fashion, Cloudaeon used the model photographs already available from the client as inputs. AI image-generation models were then used to create standalone flat-lay images of the garments on a clean background. This removed the need to repeat a complete flat-lay photography process for every product.
Generating campaign video and audio through AI: The same content-generation initiative also extended to marketing ad campaigns. AI was used to generate both audio and video assets. This extended the automation beyond static eCommerce imagery into campaign content creation.
Technology Stack
Segmind
Segment Anything Model
Computer vision libraries and techniques
LAB colour processing
Generative AI image models
Nano Banana
AI video generation models
AI audio generation
Outcome
The bedding pipeline processed around 100 images in one to two days, including review. Manually, the same volume would take about 10 working days. This delivered a 5x improvement in processing speed.
The campaign-content workflow created similar efficiencies by allowing scripts to move into AI-generated audio and video production before entering the client’s approval process. The result was not automation for the sake of automation. It was a faster content-production process that still respected the quality standards required for customer-facing retail imagery.
Conclusion
For a large eCommerce business, product content generation is a challenge. Hundreds of products, multiple colours, additional photography, editing and campaign requirements can consume a large amount of time before a customer ever sees the finished asset.
If manual photography, editing and content production are limiting how quickly your e-commerce catalogue can scale, talk to Cloudaeon about engineering an AI content-generation workflow around your existing process.