← Back

Case Study

LIMITLESS: SEE THE IDEA BEFORE YOU MAKE IT

I keep references for the clothes I want to make: a photo of a fit, a print I like, a color I saw on a sleeve. Limitless is the board where those pieces meet the models that can try them out. On an infinite canvas, images, references, palettes and prompts connect to generators, and each result lands back on the board as one more thing to work from.

Every board on this page is live. Move things, rewire them, run the generators; nothing is saved, and Reset puts a board back the way it was.

Limitless logo: a folded ribbon forming an infinity symbol

Platform

Web app

Focus

Garment design and references

Models

Gemini (Nano Banana Pro and 2, 3.1 Pro for vision and text), GPT Image

A working board. Images, a clip, a sound, text in a few faces, and a group of raglan studies.

Try it

  1. 01Drag anything, or pinch and scroll to zoom
  2. 02Select the three raglans and press O to organize them
  3. 03Open the minimap at the bottom and jump across the board
  4. 04Switch the theme from the top right

01_ The drop

Put the artwork where it belongs

The first thing I wanted was to see a print on a garment before committing to it. A generator takes a photo of the piece, a reference lifted from another image, and a palette, and the placement editor lets me draw the region where the artwork should land. The model gets the base, the artwork, and a guide with the region marked, so the print goes on the chest rather than wherever the prompt is loosest.

The reference node carries a note as well as pixels. "Muted and a little worn, like a vintage transfer" rides into the prompt beside the image, which is how the print comes back looking like it was screened onto cotton rather than pasted on top of it.

The tee, the painting as a reference, a palette, and the take the generator produced.

Try it

  1. 01Click the generator and open its placement editor
  2. 02Move or resize the chest region
  3. 03Press Run and watch the print land on the tee

A run takes about a minute. The result replaces the output in place; the earlier take stays in the history.

02_ Colorways

One garment, every color at once

A palette is a set of swatches with a mood. Linked to a generator it becomes part of the prompt, and Colorways runs the generator once per swatch, each result landing beside the last so the whole set can be read in a row. Every take keeps its own pixels, so flipping between them is instant.

The palette can also suggest colors from its own vibe: "warm autumn streetwear, earthy with one loud pop" gets a few candidates that work worn together, named the way a designer would name them.

Four takes from one palette, one per swatch.

Try it

  1. 01Flip through the four takes on the generator
  2. 02Change a swatch on the palette, or ask it to suggest more
  3. 03Run Colorways again and watch the row grow

03_ The read

Ask what the garment is made of

Before reproducing a piece I want to know what it is: the cut, how it sits, what the fabric probably weighs. An analysis node reads a photo the way a buyer's sample note would, with a confidence on each material and the visual evidence for it, and honest about what a photo cannot settle.

Linked to a generator, that read rides into the prompt as facts to keep. "The same piece in cream waffle knit" then keeps the boxy cut and the drop shoulder instead of drifting into a different garment.

The photo, what the vision model read off it, and the reproduction the generator made from the read.

Try it

  1. 01Open the analysis node and read the materials and their confidence
  2. 02Change the prompt on the generator to another fabric or color
  3. 03Run it and compare the cut to the original

04_ Talk

Say it, then let the optimizer write it properly

Most of my prompts start as a sentence I would say out loud. The generator has a mic: describe a change and the words become the prompt, or revise the one already there. Optimize then rewrites it the way the model wants to be told, naming each input by its role so "the reference" and "the tee" can never be confused.

The reference here lends only its shape and color, which is a setting on the link rather than a sentence in the prompt. Borrow the raglan seam and the two-tone split, keep the tee.

A rough prompt, on purpose. The optimizer is the thing to try.

Try it

  1. 01Tap the mic on the generator and describe a change
  2. 02Press Optimize and read the prompt it wrote
  3. 03Run it

The mic needs a browser with speech input; on a phone, type the change instead.

05_ Refine

Correct the take, not the prompt

The first result is rarely the last. Rather than rewriting the prompt and starting over, a take can be refined: point at what is wrong, say what should change, and the correction is applied to that take with the originals alongside for comparison. Each refinement is its own take with a parent, so the chain can be walked back.

The last step is not a model at all. Sharpen runs in the browser, an unsharp mask with a radius that scales with the image, and the result is a take like any other, ready to download.

Three takes: the run, a refinement of it, and the sharpened version of that.

Try it

  1. 01Open the take history on the generator and step through it
  2. 02Click Refine on the output and type a correction
  3. 03Sharpen the result, then download it

06_ Regions

Mark exactly what should change

Some edits are about a part, not a picture. A region drawn on the base can carry a swatch, and the model is told whether the color belongs to the whole part within its seams or only to the fabric inside the rectangle, as a hard-edged block. A region can also carry an instruction of its own, "add a chest pocket here", grounded to the spot rather than left to the prompt.

The parts are identified from the selection, so a rectangle over a sleeve becomes "Left sleeve" in the prompt, and left means image-left, never the wearer's.

Two regions on one base: a part-scoped recolor and a placed instruction.

Try it

  1. 01Open the placement editor on the generator
  2. 02Switch the sleeve recolor between part and area
  3. 03Move the pocket region and run once

07_ Next

Coming next

  • 01AI tech pack generation
  • 023D visualization

The purpose of Limitless is to take an idea and bring it closer to being produced in the physical world. AI tech pack generation is the next step: turning a garment mockup into the details someone needs to start making it, from front and back views to measurements, fit references, colorways, and a bill of materials.

The pack will give a designer and a sample maker a shared starting point for discussing the piece and resolving what still needs to be decided. Details drawn from images will carry their sources and confidence, estimates will stay clearly marked, and manual corrections will be preserved as the design develops.

3D visualization will bring the same idea closer to a physical garment by exploring how its panels, seams, proportions, and fabric work together on a body. The aim is to turn a flat mockup into something you can rotate and inspect, making construction and fit decisions easier to discuss before producing a sample.

Soundtrack04