Sandra's Designing Blog

This is an exclusive area of my web site where you can find out more about my knitted designs, what inspires me, how I work and what I like (and don't like) to design. 
 
Click on the arrows on the right to see my other blog posts.
 

 RSS Feed

Category: AI - Artificial Intelligence

  1. AI and virtual garment modelling

    Posted on

    Did you know that AI can take a photo of a garment laid out flat and magically show it on a virtual model? When I found this out, I thought yeh, that would be great for ebay users who are just trying to sell their vintage dresses but would it really work for my knits? Could I really get away without doing a proper photo shoot? It would be a game changer if I could. 

    So I thought I'd do a little researching and a little testing, just to see. 

    Here are some AI images of my very latest design. Can you guess what it is?
     
    Apparently It does this by “reading” the colours, shapes, and textures in your photo, then re-creating them on a model’s body .But AI doesn’t really know what a cardigan, shawl, or jumper is – it’s just guessing based on other pictures it’s seen. That’s why things can go wrong such as:
    • Adding extra fabric to the garment.
    • Intricate stitches turn into plain fabric.
    • Extra frills appear from nowhere.
    • Colours and patterns change or blur.
    • Parts of the garment can vanish.
    • Wrong garment is assumed

    These glitches happen when AI can’t work out how the garment hangs or where details should go, so it “fills in the blanks” with what it thinks should be there.

    I learnt for now, nothing beats a real photo of my actual work.  So here is my latest design, the Sea Kelp shawl.
    Dive into texture and movement with the Sea Kelp Shawl, a dramatic, ocean-inspired wrap that blends fluid shapes and playful stitchwork. Worked from a neat garter stitch tab, the shawl begins as a soft semi-circle before flowing gracefully into a crescent shape. Stripes alternate between single-colour and two-colour brioche, each section rippling with waves, dotted with bobbles, and finished with whimsical frills reminiscent of drifting seaweed. The journey ends in a beautiful brioche picot edge – the perfect flourish for a statement piece. Whether you choose ocean-hued yarns or a bold contrast palette, the Sea Kelp Shawl is a joy to knit and a show-stopping accessory to wear.
    View the Sea Kelp Shawl Pattern
    Want to start learning Brioche?

    and for the techies out there, here is a fuller explanation

    AI image generation tools can take a photo of a garment laid flat – for example, a sweater, shawl, or dress – and create a realistic image of that garment worn by a virtual model. This process uses a technique called image-to-image generation or virtual try-on. The AI analyses the original photo to detect colours, shapes, textures, and patterns, then “projects” those details onto a chosen model pose.

    However, there are challenges. AI doesn’t actually understand what a garment is or how it’s constructed. It’s matching visual patterns, not working from a pattern schematic or knitting chart. This can lead to:

    • Misinterpretation of the garment’s shape – A shawl might be draped incorrectly, a cardigan may appear to be a pullover, or sleeve lengths might change, because the AI guessed wrong about how the fabric should hang.

    • Added or lost details – The AI may invent extra texture, change stitch patterns, add seams or embellishments, or even simplify complex areas into flat colour. This happens because the AI is “filling in gaps” where the original photo didn’t provide enough visual information for the new angle or pose.

    • Colour and pattern shifts – Subtle shades, gradients, or intricate colourwork may get blurred or replaced with more generic patterns if the AI can’t recognise or recreate them convincingly.

    • Incomplete or missing parts – Edges of the garment might disappear, accessories could be removed, or folds may get erased if they don’t match the AI’s training data for similar garments.

    These issues occur because AI generates images by blending the input photo with its own learned patterns from millions of other images. If the garment has unusual shapes, stitches, or construction, the AI’s “best guess” may stray far from reality.

    The more accurate and high-quality the source photo – especially with clear edges, good lighting, and minimal distortion – the better the AI’s results.

    IMPORTANTLY for knitters, crocheters, and other makers, AI still can’t replace the eye and knowledge of someone who truly understands the craft.