Best Practices for Generative AI and Images
Interesting tips on practices for generative ai that will help you get the roi you want out of your ai tools.
In a previous article where we talked about ‘beware the shiny tools’ to caution against AI tools that can demo extremely well but may struggle with providing an actual ROI. We wanted to take that a step further now and drill into a few key best practices that should be considered when looking at using Generative AI to support design and development.
Key concepts when generating images & videos
- Integration with CAD Tools. Designers must be able to easily leverage images generated in their preferred CAD tool to finalize and then hand off to the Product Developer / Technical Designer (examples: Adobe Illustrator or Photoshop). No system done right should force teams to manually rekey data or have to download/upload files between systems. If you use any Adobe product, Adobe is also doing some great work in the AI space so it must be complementary to Adobe.
- Leverage Historical Data. New designs should be assessed against similar designs produced in the past to see how well did it sell, what was the average margin, were there a high number of returns? Simply identifying similar designs can be a significant time saver to development if it can identify an existing tech pack that can be leveraged to reduce data entry time in PLM.
- Integration with Enterprise Tools. Speaking of PLM, the data in an AI platform should consider where the data starts from and/or needs to flow to. Depending upon your process, the data may start in a Merchandise Planning tool, PLM, or the AI platform. Similarly, once the design in AI is ‘finalized’, that information should then flow to these same tools.
- Support 3D / DPC / e-Commerce Needs. AI to 3D is not mature enough yet to help initiate 3D, but the AI platform should be able to leverage product photographs or 3D renderings to create photorealistic renderings (e.g. avatar/face swap, background manipulation) and automate the process of generating outfits / looks that combine multiple designs onto a single person (e.g. 10 tops + 10 pants = 100 possible outfit combinations).
- Brand Specific Models. You MUST be able to maintain multiple AI models (the “brains” of the AI platform) that understand your brand identity and respect intellectual property concerns. Technically, the system should be able to combine multiple models together as some will be seasonal while others are seasonless (e.g. print designs vs. silhouettes). Some tools do support the ability to have a custom model but buyer beware as those costs are not disclosed up front!
There are many more best practices that we’ve built into the Straight Lines AI platform but felt like the above are the most critical to share. If you have any additional comments, we’d love to hear your thoughts!
