i mean that’s likely the closest i’d get to anything based in yarn on the first go
it just cracked me up
it seemed extremely remote.
Likely you’re the first person to ask the bot that
Most people with patterns like that don’t even know what a keyboard is ^.6
I can think of one person in the whole internet who would want it, and they’re not a high-dollar character
Cool concept tho!
Not the first person. 
DALL-E 3 had a prohibition on “stitch” (like it did on the corporate takeover of the public domain color “snow white”). Keyword-like rejection. Frustrating to those already trying cross-stitch patterns.
The challenge is getting good “grid” generation out of the AI model. You might need high quality API parameter in conjunction with producing axis labels. Even a template to re-trace with a matching non-resize input.
My initial idea of guidance
# Cross stitch pattern creation
## Image layout
- The left 1/3 of the output image is reserved for presentation of the cross stitch concept, color key legend, technique, etc.
- The right 2/3 of the output image is reserved for production of a square cross-stitch pattern.
## Producing quality patterns
- Each axis must be labeled with numbers, starting with "A", progressing to AA for columns, and "1..." for rows. Thus, the count of cells can be easily determined during the image generation process, and cells can be referenced by the person sewing the generated cross stitch pattern.
- The grid is outlined.
- Non-sewed areas shall be white or another specified base fabric color.
- For an aspect ratio given that does not suit the subject (such as tall person in a square), ensure good area fill of remainder with content also.
## Establish the color vocabulary
Use a limited complement of colors. Promote like colors in adjacency. Use mosaic creatively, along with dither to give impression of a deeper color than the number of thread colors available.
# Image content
## Grid size
40x40
## Subject
Cute bobble-head *goth girl* figure
“Stitch” as subject still denied when input is by name or by inference in the output content:
moderation_details: {“moderation_stage”: “output”, “categories”: [“other”]}
(and unlike “public figure”, a new message, the why of the prohibition in the seen generation is not explicit)
So the prompt demo is a different subject. Direct guidance can’t fix the faults in trying to make arbitrary output absolute or getting the right column count or getting axis labeled without nonsense. You need a “best of”, and manual inspection. “medium”:
Go ahead and sew double or triple the pixels as the grid…or nonsense references.
That’s truly interesting.
I try to think back about all the things AI-related, and it’s just one whirlwind soup of progress and media buzzwords.
The word stitch is certainly in that soup, but now I’m realizing how hard it is to have a mental timeline of any of it anymore.
two for one… but it took one back after I saved it…
Generate an image of magical tropical island. (ChatGPT)
Got inspired of the real, new huge cruise ship Legend of the Seas which is actually built in my birthplace in Finland. This huge cruise ship just yesterday passed under Great Belt Bridge in Denmark.
Royal Caribbean, Legend of the Seas cruise Ship, no text, from 2026.
quality: lowsize: 976x704
Today’s idea
Proto-Mark: Stacked Blocks
@PaulBellow this one, found out today looks a lot like your art style as a building block. You may want to check Acrylic Inks and Paint with Motion as an art style.
I have to explore the inference of the beautiful and unique more. That’s so open ended.
Prompt
A well-lit, detailed, colorful pencil and watercolor drawing of a bearded man sleeping in bed with a large thought bubble depicting a bright, colorful nightmare.
Prompt
A well-lit, detailed, colorful pencil and watercolor drawing of a bearded man sitting behind a computer screen with a large thought bubble depicting colorful mass confusion.
Interestingly enough, if a model can’t resolve the context you’re speaking within, it might also cause the image produced with the prompt, ‘make an image of our recent context,’ to have the unresolved pattern…
It doesn’t happen often, the models usually understand many more aspects about what we’re talking about to begin with…
But it’s just a little bit… so I decided to investigate…
Which have now been slightly resolved… so the next image should be mostly without the pattern?
Yes it’s only confused on a few things, which doesn’t mean it’s right about all the things it’s not confused on…
We know full well model confidence can be misplaced…
But letting the models visually portray their understanding of the session context might actually prove to be a very valuable tool for me…
And now my most beloved prompt of all is…
And likely will be for some time.
‘make an image of our recent context please’