I recently rebuilt my side project around one job: turn a picture into a counted cross stitch chart with DMC thread numbers. While building it, I learned that most bad charts are not caused by the converter. They are caused by three settings people leave on the default: stitch width, maximum colors and fabric count.
This post is what I wish I had known before I wrote the first line of code. It applies to any photo-to-pattern tool, not only mine.
What does "width in stitches" actually control?
Width is the single most important number in the whole process. It sets two things at once: how much detail survives, and how long the project will take.
A chart is a grid. If you choose 100 stitches wide, every row of your photo is squeezed into 100 cells, and the height follows the photo's proportions automatically. A face that looks sharp on your phone may end up as a 12-cell blob.
Rough guide I now use:
- 60 to 80 stitches for a small motif: a flower, a mug, an initial.
- 100 to 140 stitches when the picture has a face, a building or a landscape that needs room.
Going wider has a cost. Every extra column adds a full row of stitching to every line. Before you push the slider up, picture yourself filling in that grid by hand.
One more thing: crop first. Empty edges around the subject still cost stitches. If the cat only fills the middle third of the photo, two thirds of your width is spent on carpet.
How many colors should a photo pattern use?
I start at 20 to 24 colors and adjust from there.
Too many colors produce what stitchers call "confetti": single isolated stitches of a slightly different shade scattered everywhere. Each one means a thread change, and the shopping list grows. Too few colors and important shadows merge, so a nose or the edge of a petal disappears.
My process:
- Generate at around 20 colors.
- If the preview looks speckled or the thread list is long, lower the count.
- If a key edge or shadow vanished, raise it a little and check again.
Every retained color is then matched to the nearest entry in a DMC reference table (the one I use has 489 colors) and given a symbol for the chart. This matching step is why a photo with subtle gradients often looks flatter as a chart. The gradient gets snapped to the nearest real thread colors that exist.
What is "clean stray cells" for?
This was a feature I added after looking at my own test charts. Even at a sensible color count, photos produce lone cells: one stitch of a different color surrounded by a neighbour color. They are almost invisible in the finished piece and annoying to count.
The cleanup step replaces those isolated cells with a nearby chart color. It makes the chart calmer and easier to follow.
The trade-off: sometimes a single cell is the point. An eye highlight, a tiny sparkle on a glass, a star in a night sky. If something important disappears after cleanup, turn it off for that chart.
Why does background removal sometimes miss?
The background option in my tool follows similar colors connected to the corners of the image. That works well for an object photographed on plain paper or a plain wall. It is not an AI cutout, and I chose that on purpose: it is predictable and it runs locally.
But it means a busy background will not disappear, and a hole inside the subject (the gap in a wreath or a mug handle) stays filled unless you also clear enclosed areas. That second option can remove matching areas inside the subject too, so always look at the preview after you enable it.
If background matters, the best fix is upstream: take the photo against a plain surface.
What does fabric count change?
Fabric count is stitches per inch. It does not change the chart at all. It only changes how large the finished piece will be.
The math is simple: finished width in inches = stitches wide divided by fabric count.
For a 100-stitch-wide design:
- 11 count: about 9.1 inches
- 14 count: about 7.1 inches
- 16 count: 6.25 inches
- 18 count: about 5.6 inches
Those numbers are the stitched area only. You still need extra fabric around it to hold it in a hoop and to frame it later. I add a generous margin on every side before buying fabric.
14-count Aida is the common starting point for beginners because the holes are easy to see. Higher counts give a smaller, finer result from the same chart.
How reliable are thread estimates?
A useful chart lists, for each color: the symbol, DMC number, name, how many stitches use it, and an estimated number of skeins. My estimates assume two strands and round up.
Even so, treat the estimate as a floor. Starting tails, unpicked mistakes and dye-lot differences all eat thread. For any dominant color, especially a large background area, I buy one extra skein, because matching it later can be hard.
What did building this teach me?
A few product decisions came directly from these problems:
- Convert in the browser. People stitch photos of family, pets and homes. Doing the conversion locally means the photo never leaves their device, and it also makes rebuilding the preview instant, which matters when you are adjusting three sliders.
- Let people rebuild the preview as often as they like. Most of the work is iterating on width and colors. If every attempt cost something, people would stop at the first mediocre chart.
- Show finished size next to the settings. Without it, users choose 18 count for a "fine" look and get a much smaller piece than they pictured.
If you want to try these settings on your own picture, the photo to cross stitch page has the width, color, cleanup and fabric controls in one place, plus the same guidance in more detail. Use photos you took yourself or have permission to use, especially when they show other people.
Quick checklist
- Crop to one main subject.
- Pick width by detail needed: 60 to 80 for small motifs, 100 to 140 for faces and scenes.
- Start at 20 to 24 colors, then lower until confetti goes away.
- Use cleanup, but check that eyes and highlights survived.
- Choose fabric count by the finished size you want, then add margin.
- Buy one extra skein of any dominant color.
I'm the maker of AI Draw Kit.
Top comments (0)