Category: Uncategorized

Automated shapes across rows of knitting using slip stitch only reviews several methods to automate/ program short row techniques.
BW repeats may have stripes of color planned and added by simply changing yarns in the yarn feeder in the selected areas.
A recent FB comment asked the question as to whether it is possible to automate intarsia using img2track downloads.
Short row “intarsia” can be executed with manual needle selection following charts, or as an easier, viable option using a knit radar and with shapes illustrated on garment pieces, which can follow the simple geometry seen in the hand selection methods, or graduate to the degree of complexity the knitter has the desire and patience to execute.
These posts illustrate simple shapes and the breaking down of more complex ones.
Intarsia without an intarsia carriage on Brother machines.
A bit of holding 5: intarsia and more 2
A bit of holding 4: intarsia and more
A bit of holding 3: shape variations and more
A bit of holding 2 / moving shapes around
A bit of holding 1
The Silver Reed electronic intarsia carriage was briefly available as an accessory for knitting complex intarsia designs with automated needle selection.
The yarn is placed and used in the same manner as when using the Brother intarsia accessory, but note that there are now multiple needle positions and selections for the 4 types of intarsia offered.  I gave away my standard intarsia carriage, so I cannot speak as to whether programming a design, placing yarn in selected needle groups manually as above, and bringing them all to the required position for the carriage to make successful passes would be a viable alternative.
These images are from the E6000 Passap manual, alongside the electronic repeats, which are mirrored on the front face of the knit. The accompanying technique recommendations. Without a Passap, and the discussion being on Brother electronic knitting, starting with duplicating one of the Passap designs and thinking it through, the 32X28 2-color repeat,doubled in height to 32X56, color separated for 2 color DBJ with each color in each design row knitting twice by color inverting every other row and doubling the height once more, to a final repeat of 32X112, mirrored for use on the 930. End needle selection must be canceled.
The pattern setting in the knit carriage is for slip-stitch in both directions.
The first preselection row is from right to left, with color changes every 2 rows using the changer on that side.
There will be a yarn ends to weave in, as at the start of any intarsia design.
Since each color knits for two rows, eyelets are formed at the color changes, familiar in short row designs. If bothersome, the size of the eyelets could be reduced.
The pattern was tested on 32 stitches and for 132 rows. At any point, more colors could be introduced for horizontal striping color changes.  When the yarn traveled across multiple stitches before being picked up, the resulting floats were hooked up every third needle on the same color stitches before knitting the second pass with it.
Central designs are more problematic to execute and, if not very wide, may knit more easily as single FI motifs, with wrapping at the sides, and latching up floats as needed.
Img2track will separate for 3 and 4-color designs.
This first attempt was a failure and reminded me why I love/hate the single bed color changers.
Aside from the changer and sinker plate crankiness, the colors, as they were carried up the left side, would twist around each other, forming floats to the next selected group.
Img2track selects colors based on the number of cells that are represented  for each in the first design row, hence the sequence is blue, yellow, red.
The test image, 32×28, was doubled in height to 32X56. I prefer to start short rows from the right.
Aiming for a similar attempt with the automated version, the image was mirrored horizontally for use on the 930.
The first preselection row is toward the color changer (remember to change the setting when the carriage is on the left to slip stitch).     This was a day when everything that could go wrong had. In addition, color #3, the red, catches on the left side before traveling to knit the narrow center band of color for no reason I could immediately fathom. What the manual has to offer in terms of threading the changer.After the knitter, accessory, and the machine got a break, analyzing the order for color changes made by the img2track, was reminded that selections are made in order of the number of cells representing each color in each design row, making the red, color 3, the selection for the center shape, catching that color on the left.
To eliminate that problem, one solution is to use a second carriage to select and knit the center color, operating from the right side of the machine, using rails and making sure each knit carriage clears the belt before the second carriage operates from the opposite side.
The programmed repeat stays the same.
This particular design has some jumps in the number of needle sequences that form long floats. They can be managed as one would in FI, i.e. The resulting test swatch.  Devising a manual color separation to work cleanly with a fixed 1,2,3 sequence, the full repeat was cropped to 32X48 pixels in height, with each color in each design row still occurring X3, and the result was mirrored for execution on the 930.  Using ArahPaint, the process shown charted in progress: The double height rendition, second from the right, with the white color locked, can have all other colors filled with black, as seen in the charted repeat on the far right, finally programmed as FI in img2track.
Missing out on the prompts for color changes is not a loss. It is easy to see which colors should knit next, with #1, red, starting on the far right, followed by the dark color, 2, for the center shapes, and then 3, the yellow/orange. The BW color-separated file PNG, 32X96, and the proof of concept swatch. A convoluted approach: DAK Method C, if you have access to a screen grab of the template can convert it manually to a BW PNG, which may work with img2track, the double-length switch will need to be used in Japanese knitting machines. The DAK jacquard choices. If the goal is to use the technique in large knit pieces, then design repeats that tile properly can be used.
Encountering limitations to automation: a 20X20 repeat drawn using the pencil and straight line tool, checked in repeat, 60X60, filling in diagonal lines, isolating a new 20X20 repeat, doubled in height to 20X40, color inverted for use on the 930, and finally color separated, 20X80.   When using a repeating occurrence of the width of groups of each color in programming the final piece, there are some related problems: as selections are made and repeated across a row of knitting, the color being carried skips the non-selected needles, forming floats. Using the end needle selection on the carriage to catch yarns when needed at the side edges will throw off the repeats, as seen in the photo. End needles in work will also knit an extra stitch on either side of the non-selected needles, changing the intended groups of 8.  Returning to hand selecting needles, it becomes easier to add colors as well; there is no need to mirror the chart. The knitting sequences are numbered, and stitches are brought into hold after several rows of base knitting/ waste yarn.  To maintain straight side edges, it is best to chart the repeat in full. The original sketch can represent single rows of knitting, but short rows will be executed with each color making 2 identical passes, resulting in the familiar eyelets, and elongation of the design.
Color changes can occur frequently, limited by patience, yarn availability, and the knitter’s willingness to deal with lots of yarn ends.
The initial charted repeat is 40X40 pixels in width and height, and will tile correctly if larger pieces are planned or if longer pieces in a fixed width are the goal.  My swatch was knit with randomly grabbed yarns of my stash, in multiple thicknesses, not the best choice for the final pieces, resulting in some distortion along the diagonal lines.
All stripes are 8 stitches in width.
Most of the movements are needle one out on the left, one needle in on the right, except when small starting or ending triangles at the top or at the bottom of the diagonals are present.
When the top of the planned test is reached, stop on the right side when the last row of 8 stitches occurs for that color is reached, bring the group out to hold, make a free pass to the left, bring the next selection of needles to work, and continue.
Check stitch counts before returning to the left to pick up the next color. Errors may not be noticed immediately, i.e., in the purple stripe.
If ends are woven in, try not to pull on them; allow for some ease to keep that last row cleaner, all easier to manage with thicker yarns than those in this swatch. There is an accessory called an intarsia yarn brake.   While it was in my stash to demo in teaching days, I much preferred bobbins for personal use. Some knitters, as with all other accessories, find it useful. If possible, check reviews and take one for a test drive before making any purchase.
These are suggestions from published sources on how to deal with stitches in holding position, often found in conjunction with shaping shoulder slopes. Wrapping needles adjacent to shaping reduces the size of the resulting eyelets where the same row is repeated twice. The Brother Techniques Book  is a wonderful resource; their suggestions on partial knitting are on pp. 44-47, their illustration for managing those floats when moving by multiple needles, possible on both sides of the piece, harder to bring to the front if the pattern is automated and programmed for use with the slip stitch setting. The angles and size of the shapes can be varied, straightforward to program if only two colors are used in the intarsia concept, but there is also the possibility of changing colors every 2 rows to create horizontal stripes that interact with each other, forming secondary striped interacting shapes.
The next rabbit hole will be exploring short rows, automated when possible, while making stripes intersect in shapes that go round, a return to “doilies”.

 

Bubbled textures can be created in a variety of ways, such as with hand techniques, i.e., ruching, or automated, commonly with slip stitch designs. Machine and yarn tolerance are considerations when having multiple stitches and rows skipped, at times, an impressive number of times.
A reminder of the differences between slip and tuck stitch formations.
In both, stitches in non-selected needles grow in length until they are knit again. In slip stitch, needles are skipped, forming floats (bottom knit/purl illustrations), while in tuck stitch (top illustrations), loops are built up in the same needle hooks as the held stitches, making for a less forgiving repeating needle no-selection choices. This repeat, with tuck happening for four rows, began with a 6X24 PNG. Loaded into ArahPaint, it was drawn in a brick repeat, with a final PNG of 24X48, which in turn needs to be color-reversed to be used for tuck stitch. Punch card machine users need to punch all the black cells. 
The swatch was test-knit on 53 stitches, using a 2/8 wool, and measures 7.75 inches in width. Using the same yarn, color, and tension, wondering about making the texture more pronounced, I tested the same repeat with manually selected FI patterning, which produces a narrower width than stocking stitch. A programmable PNG, 6X48charted for punch card knitting.  The swatch measures approximately half an inch less in width than the first sample.  My stash of knittable elastic is extremely limited. Continuing to explore making small changes and exploring their effect, here the same design repeat was used, with elastic as one of the 2 yarns in the FI bands. The repeat is changed to standard as opposed to brick configuration, with textures lining up vertically. The width of the swatch, knit on the same # of stitches as the last sample, measures 6 inches in width as opposed to the non-elastic FI version’s 7.5, with a loss of .75 inches in comparable knit pattern height. The main yarn was switched to 2/20 wool, and the pattern was divided by 4 rows knit using only the elastic, with a return to the brick configuration. The work on the machine has a very different appearance when off. The swatch is nearly identical to the last in width, but the shapes collapse, shrinking the overall height. A few changes in the number of stitches and rows represented by white pixels for stitches that will be skipped, beginning with a brick arrangement, 24X80. The yarn used is a 2/8 wool; the swatch was knit in only one color to check yarn tolerance for stitches being held in needle hooks until they are knit off again, and with the intent of using it as the ground when adding another color to the mix. The test was on 60 stitches, and for 160 rows of pattern, measuring 6 inches in height. The second swatch began with the 24X40 design in simple vertical repeat,  also tested on 60 stitches and for 160 rows of pattern, measuring 7.25 inches in height.
The dark-colored ground knits all stitches in all rows, where there is a pre-selection of every needle.
The yellow yarn is 4 ply, thicker than the dark, a 2 ply.
Secondary shapes formed by the textured gathers recognized by the eye of the beholder are part of this type of experimentation, not often predictable, and at times a fun surprise.
A range of added textures, some created using the thread lace setting, can be found in the blog posts:
Knitting with “unusual” fibers/ elastic 3 5/22
Knitting with “unusual” fibers/ elastic 2 8/20
Knitting with “unusual” fibers/ elastic 1 1/20
Balanced designs are not the best to use with the thread lace setting, a sample that began with a 24X24 PNG with added pixels to break up some long potential floats. The elastic is threaded through feeder A, the “bubble” yarn in feeder B, illustrated aside some of the work on the machine, appearing flat and stretched out. The elastic will shrink back when the finished swatch is steamed. The blue yarn is a 2/20 wool; the elastic is from an unmarked cone in my stash, a tad too thick, with a mind of its own, knit on 50 stitches.
End needle selection is canceled. If needles are selected in a pattern, they need to be pushed back to the B position so both wool and elastic will knit together.
The finished width after steaming is 4.5 inches.
Whether the shapes are recognizable and wearability are open to question. A more successful swatch, returning to the family, aiming for larger-scale shapes; the yarn is 2/20 wool, the elastic is a 280S cotton wrapped, the starting brick repeat is 24X52. Keeping in mind the respective functions of the black and white pixels, the image is color-inverted for use with the thread lace setting.
Side edges with non-selected needles will knit both yarns together from clean vertical columns, i.e., in this 56X104 repeat. The actual swatch was knit on fewer stitches on one side of 0 than the other, explaining the different side edges. If any needles were selected, they were pushed back to B manually before the next carriage pass. After some light steaming, the result is a bubbly texture that measures only 3 inches in width.

To mesh or not to mesh 10: more large eyelet variations provides a review of information in some of my previous posts, and links to information to achieve the combinations shared by others.
The direction of transfers is marked in many of my charts with cyan cells for transfers to the left, and magenta for those to the right.
In these knits, large eyelets are created by side-by-side transfers.
Two empty needles in standard transfer lace may indicate an error, or that the repeat needs to be mirrored, while here, if in testing these patterns, only single stitch eyelets are produced, they indicate the same.
The transfers resulting in side-by-side needles on which loops will form with the next knit carriage pass are circled. The next group of transfer selections will anchor the side-by-side loops.
The repeats mentioned as used on the 930 may need horizontal mirroring, depending on the machine model or the download software.
The 32X24 design produces a vertical column of small eyelets on a field of larger ones.  The eyelet placements are planned this way, but the full repeat, 32X24, needed mirroring horizontally for use on the 930The work on the machine: after the side-by-side transfer in opposite directions is made, there will be a loop on each empty needle hook. The usual second pass with the knit carriage repeats the loop formation, and the first pair of loops is dropped. The choice can be made to keep the dropped loops, forming visual ladders, or to lift them onto the hooks so that when they are secured by the next pair of transfers, the edges of the eyelets form cleaner rounds.
Here, the dropped loops are undisturbed.
A potential troublesome spot is those loops getting hung up on gate pegs, so a light color is useful when first experimenting. As the return to transfers is made, the transferred stitches will anchor down the side-by-side loops as new double transfers are executed.   Vertical columns formed by small eyelets on a large eyelet ground, mirrored for use on 930 to this 32X24. Vertical column formed by large eyelets on a chevron ground, for knit side borders of any size, cast on a larger number of needles than in the width of your programmed lace repeat, here a 24X32 repeat is centered on 32 stitches, as used on the 930. Creating shapes: in Seasonal knits inspired by published repeats 2_hearts, 36X88,  used a combination of large mesh and single transfers to create the heart shape. 

 

One of the many differences between automated machine-knit transfer lace and hand-knit lace is that the alternate direction of transfers in machine-knit lace occurs after the first is completed and followed by knit rows separating the eyelets, so there is always a slight shift in horizontal lines formed by them.
In hand-knitting or when stitches are manually transferred in both directions on the machine before knitting any additional rows, the appearance is noticeably different to purists and closer to hand-knit.
To truly match some of the transfers, stitches need to be removed from their respective needles, and other stitches are placed on the empty needles before the removed stitch is restored to its place. That is not an automatic function, but an added hand technique.
Changes in aspect ratio are often accompany knit structures and are seen in lace as well, more noticeable when one begins to use representational motifs, as in these recent cases, flowers.
There are so many sources available for designs, one need not necessarily build motifs from scratch, and they can be made more personal with modifications.
After choosing the base mesh, math may be used to change the scale of the superimposed shape.
Using small designs to test differences in results, one way to simply change the result is to multiply the height of the original shape by X3, with or without also doubling its width, thus approaching the 4X6 ratio often attributed to knit stitches other than garter.
The designs are programmed as fair isle in electronic models, which often automatically mirror the design horizontally.
Ways to modify electronic and punch card published transfer lace designs 6/26 returned to a partial image of a card used in 2011 as a mock filet mesh with traditional LC operation from the left side.
For the bse mesh, the LC makes 2 passes for transfers to the left, 4 passes for transfers to the right, each sequence is followed by two knit rows using the knit carriage operating from the right.
With the punchard located, the repeat was rebuilt using Mac Numbers, and a combination of ArahPaint and Gimp to reproduce the 24X162 file as a PNG. A 24X84 segment was, in turn, isolated, drawn in brick repeat, 24X168.  The swatch, tested on 60 stitches, helps to determine whether other DIY arrangements, i.e., widening the repeat or altering the edges of the shapes for use only on electronic machines, should be made.
For some reason, the yarn used, which had posed no problem in recent swatches, was particularly cranky in transfers. The swatch has several dropped stitches.Using a random flower design from a published source, 18X20 multiplied X2 in height and width, 36X40 4X6 brush fill to use as a background on a 60X60 canvas offers 2 choices, one can be to copy and paste the enlarged flower onto the mesh grid with the white color to alpha,  the other to lock the magenta color, and superimpose copy and paste the 60X60 mesh ground file onto it, not anchoring it in place until the preferred position is found.
36X40 The magenta shape is then bucket-filled with white.
Either method will require some cleanup to reach a programmable, final file size. In this case, for a 60X60 BW PNG.
When tested and drawn in repeat, followed by the decision to bring flower shapes closer together, cropping the design to 44X60 with a couple of shifted pixels in the center of the flower to keep the direction of all transfers the same across horizontal eyelet rows, and then tested.  Different distribution of the shapes: an untested brick repeat, 44X120,  and a half drop 88X60.  Changing the aspect ratio of the flower, an 18X20 slightly different design is multiplied in width by 2, in height by 3, to 36X60,   and superimposed on the 60X60 mesh ground with magenta bucket filled with white,  the tested repeat cropped to 48X60, produces a slight change in overall appearance,  adding the change in scale to the list of potential choices in DIY designs, if larger repeats are preferred, this previews the brick version, but the half drop needs work. Early Brother electronic machines, i.e., the 910s, used mylar sheets.
Any programmed design was automatically mirrored horizontally to avoid having images, i.e., text, appear in the wrong orientation on the knit face. When looking at images that accompany files, the direction of the motifs in the program is the mirror of those shared in the accompanying swatch.
The charted image is what appears on the purl side, facing the knitter.
Some download programs will automatically mirror any file, adding to confusion as to whether that is required in some instances.
A way to test whether mirroring is again required in your software/machine model is to program the desired mesh, knit a small swatch, and there should be no side-by-side needle selection.
A simple button selection on electronic models can flip the image horizontally if needed. My preference is to alter the PNG and save it before programming it, to avoid errors in larger pieces, or later returns to the design, since I am not always the best note keeper.
From Brother Electroknit Lace Patterns 3, two larger-scale patterns that are published on a rectangular grid show the mirroring of the design in the final knit and the change in aspect ratio.
Conversions to BW PNGs

60X132120X132
The appearance of the edges of the knit stitch shapes takes on the more familiar look of their edges as seen in my own shared designs. 
60X120
120X120 Ancient magazines are easily found online for browsing and inspiration.
Now tired of flowers, this Knitking magazine published a pattern that included a sailboat design, which caught my eye. Visually, the card offered clues as to a possible problem with the mesh ground, but initially, the only change that was made was to the starting row, as seen on the BW chart on the right.Additional clues are also in the tiled 24X56 PNG, and here are those knit stripes breaking up the ground behind the sails. It is not possible to know whether errata were ever published for the card, or whether the outfit may even have been knit on an electronic model.
If there is a fondness for the design, the background can be adjusted to maintain a consistent mesh while keeping the punch card 24 stitch constraint.
A close-up of the garment’s texture. The new 24X54 mirrored design attempting to duplicate it.  The result allows for a closer look at eyelet placements and further editing based on their distribution if desired. Some of the historically related posts:
To mesh or not to mesh 15/11To mesh or not to mesh 2 5/11

To mesh or not to mesh 3 5/11

To mesh or not to mesh 4 5/11
To mesh or not to mesh 5 7/17 a collection of mesh design repeats To mesh or not to mesh 6: chevrons 6/20

Lace knitting tips, to mesh or not to mesh 7 7/20
To mesh or not to mesh 8: more Numbers meet Gimp 5/21
To mesh or not to mesh 9: more on mock filet design 12/21

Unconventional uses for punchcards 2: thread lace cards for “filet” mesh 8/17

This will be another work-in-progress post, a rough draft.
It is released early and will receive frequent edits and additions.
I have just begun to exchange information with two individuals, with the intended possible outcome resulting in providing Vibe coded scripts for GIMP or free online generators to automate the desired separations.
“Vibe coding” is an approach where an AI generates code from natural language prompts.

The goal is to include and consider punchcard machine users, meeting the limitations of stitch width where motifs must fit together within the 24-stitch limit, so individually they must be composed of factors of that number: 1, 2, 3, 4, 6, 8, 12, and 24.
Each card height, to feed properly,  requires a minimum of 36 rows punched for continuous designs. Cards not on rolls are a standard 60 rows in height, multiple cards may be joined together if needed.
Needle selection is centered and fixed on the needle bed.
Added info is found in  Brother KMs: punchcards and their use.

I learned about machine knitting from what others presented during international and local seminars at the height of the craft, or published in magazines and hardbound editions. Stitch Painter with BitKnitter and DAK were at their inception.
Susanna Lewis has always been a personal favorite and a huge influence.
As time went on, I became interested in processing much larger files using whatever tools I became aware of in image processing or paint programs to generate patterns for producing the fabric I wanted to knit.
As a machine knitter, any charts or files I shared often began with numbering on the right and from the bottom up in Excel.
Row numbering on the right was a convention in blank punchcards and published charts; transfer lace passes added numbers and symbols on the left.
An added consideration in charting: while in 2-needle hand knitting, the work is turned over upon completion of each row, so charts are read in turn from the right and then from the left when purl rows are included in the chart. Knitting on the machine is fixed unless there is hand manipulation, ie, with a garter bar, and the knitter’s view is only of the purl side. In automated punched or downloadable patterns, every row is commonly represented, including punched holes with blank squares or black and white pixels.
Electronic models have a double-height button; in punchcard models, the card lock lever can be pushed back to the long triangle position.
I avoid elongation in long pieces, finding it easier to track unraveled rows to return to the proper program row and color when any errors or stitch mishaps occur.
Although the majority of my designing is now using ArahPaint, Gimp became my early go-to as a companion to spreadsheet programs, first Excel, then Mac Numbers.
Presently I have returned to using Gimp 2.10, finding the later 2 updates unpredictable when working with 8-bit images.
There are added baseline considerations in color separations, for example, the initial file must contain an even number of rows.
Experiments are beginning with black and white indexed images, and the goal is to conclude the process with PNGs in that same mode.

Mosaic color separations are different from those for DBJ.
The knit, also known as floatless fair isle to machine knitters, aims to have floats usually no wider than 2 stitches and for only one color, as opposed to 2 with each pass of the carriage in fair isle.
It is executed with color changes every two rows and cam buttons set to the tuck or slip stitch setting in both directions.
Two classic references: hand knitting for 24 and 40 stitch machines. In mosaics and maze conversion, the original file is not rendered double height before being separated. In other knit structures, the original file is commonly doubled in height before the separation, solving by default the even # row requirement.
Aside from the automatic color separation and what comprises the original file input, ie, indexed BW mode and an even number of rows, depending on the manual separation, the actual file on completion of the separation may need to be altered again before knitting, doubled in height.
The last double height can be achieved by altering the separated PNG before downloading it to electronic models. Symbols for that vary depending on machine or model year, in this case, 910 houses and trees, #4. The 950i  uses ships, but the button placements are in the same positions, and perform the same tasks, or the placement of the card lock lever in punchcard machines. I prefer punching or programming repeats that do not require the built-in elongation selections. Returning to specific rows after interruptions in knitting or correcting patterning errors during it are easier to track.
Whether the color separations are performed manually or with automated tools, the prerequisite is that the input file aligns properly when tiled. Often, published designs have added rows that will double when drawn in repeat, solved by cropping rows at the top and at one side, as addressed in other blog posts.

This repeat for a larger design, 38X34, was separated using the custom filter, with the knit-ready result, 38X68, obtained in a single step

A design that combines the mosaic grid work with the straight horizontal design found in mazes began with a 26X26 repeat, followed by rotations and additions of a 26X20 partial repeatresulted in a 66X24 PNG   that, when viewed tiled on the screen, looked OK   the single repeat, separated using the script,66X48, was swatched using the slip stitch setting.

Both tuck and slip shorten the end knit along the all knit stitches beside them, which in this case form wide horizontal stripes. Those stripes in turn become distorted, seen here at the top and bottom of the swatch; the degree of 3D effect depends on the ratio of slip or tuck counts to knit stitches.
The repeat adjusted to a slightly different alignment if I were to personally use it in a final piece, 66X26, separated to 66X52The question then becomes whether more is less.
The original 26X26 design, separated to 26X52, shown drawn in repeat X3, 78X52, was test-knit using the tuck setting. The yarns used are Silk City wool crepe purchased as seasonal remnants decades ago. Its fiber content is typically 80% wool and 20% manufactured fiber like rayon or viscose. The surface of the swatch is far more 3D than its slip stitch relative. Returning to a 2015 post on working with online generated mazes, and the selection with the isolated repeat, 14X34rendered in BW the script “mosaic” separation,  14X68color inverted for knitting using either tuck or slip stitch setting. The sample, knit in tuck stitch for 134 rows, measures 9.25 inches in height.

DOUBLE JACQUARD
The abridged view of the same steps in the separation for knitting the original 12X18 row design as DBJ, where each color in each design row knits twice.  

The final file is composed of pairs of identical rows.
The process begins with the original image doubled in height, in this instance,  to 12X36.  Every alternate row is color-inverted, and then doubled in height for knitting to 12X72

12X72The first preselection row is from right to left, toward the color changer in Japanese KM models.
Because each color in each design row knits twice, the design is very elongated, even when knit with birdseye backing. The random color choice was not the best in the related sample. If the goal then becomes to have each color knit only once to reduce aspect ratio distortion, or if other knit stitch structures and combinations are planned, additional processing is required.
For strictly 2-color knitting, with BW starting images, electronic knitting machines automate each pixelated BW row for knitting each color only once per row with the flick of a button, generally labeled KRC.
The starting side for the preselection row is from left to right.
The 2 rows for the first color are split between the top and bottom of the programmed repeat.
The actions performed automatically: In the post ArahPaint in knit design 6: color separation,  this suggested method proved faulty when tested here. The result shows the dramatic difference in scale when each color in each row knits only once, but there are noticeable white stitches that are breaking vertical red lines, not intended and not the result of a programming error. Analyzing the result: the first and last row of the design are unaffected by the separation, and when knitting lengths, that same design row will repeat twice.
This proposed method shifts the first design row to the top, maintaining the 12X36 dimensions, and is then color-inverted, rendering a 12X36 file that is ready to knit. The first preselection row is from right to left toward the color changer.
regrouping, 12X36 with the bottom row shifted to the top, the separation, also 12X36The steps in the separation
1: the original image, 12X18
2: the original doubled in height, 12X36
3: move the bottom row to the top of the repeat, still 12X36
4: as an optional guide, a column was added on the right, increasing the width to 13 pixels, marking pairs of rows to be inverted. The added pixels were originally red; when color inversion was used, they were also converted to blue pixels
5: number 4 with the blue and white pixel column removed, the ready-to-knit repeat, also 12X36. The swatch is knit using birdseye backing, an added technique that helps to maintain the aspect ratio of the original design.
The visible extra white loops on the side are due to yarn feeds in mast issues that have gotten added to my list of “well, I have not run into this before!”  

Duplicating the KRC model, with added changes, but the same result: the hand-selection method to illustrate the color invert steps, from left to right: the 8X8 PNG start,
rendered double height to 8X16,
with the bottom row moved to the top,
the separation illustrated with marks for the placement of the inverted rows in blue,
and the knittable final repeat, 8X16Sample color separated using a Vibe-Coded script to obtain the 8X16 result:
the goal is to knit each color in each design row only once, matching the electronic automated electronic KRC selection, with a critical difference: the preselection row is from the right, followed by color changes every 2 rows. Testing a larger design repeat, 43X44color separated to 43X88knit with birdseye backing, with preselection starting from the right

More to ponder: the method used to have each design row knit twice produces files that can be altered again in double bed work for other stitch structures, or used as they are in single bed slip stitch that knits carrying one color at a time for 2 rows.

Reviewing charts for different separations, including use of more than 2 colors: depending on the published source, sometimes letters are assigned, others numbers, and they sometimes differ in identification sequences.
Continuing with Double bed separations, the longest to execute manually, while considering different stitch structures: