Uploading assets

You do not build a game from a blank canvas. Instead you upload your source material — the rulebook and the art — and an AI pipeline reads it and generates a first-draft table: a table surface with components, zones, seats, and your setup already laid out. That draft is a starting point, never the finished game. Remember the taybl model: the AI is arranging a physical scene, not writing a program, so the more clearly your assets describe what is on the table, the better the draft.

How to edit in the Studio

On a new or draft game you land on the uploader. Drop each file into the bucket that fits, optionally tell the AI a bit more, click Upload Assets, then Build my game.

A rulebook is required to build with AI. Until there is one, Build my game stays disabled. The rules are what the AI builds the game from — without them it has only filenames and pictures to go on, and what comes back is a guess rather than your game. A PDF or a text file is the surest thing to hand it, but a photo or scan counts too: when a file has no text to pull out, taybl reads the pages as pictures instead. That covers the flattened export most design tools produce. Make it legible — if the words are too small, blurred or handwritten to read on screen, the AI will not manage either. Everything else the uploader says is advisory: it warns about a thin upload or an uncut sheet, but lets you proceed anyway.

The no-AI Build it yourself path has no such requirement — start with nothing but art if you like. You will need a rulebook before you can list the game publicly; see Publishing.

The upload buckets

There are five drop targets. You do not have to pre-sort every image perfectly — board art, pieces, tokens, and dice all go into Components together, and the AI sorts them in the next step where you can correct anything it gets wrong.

The distinction that decides it is one component per file versus many components per file — not whether your game is print-and-play. A single card face goes in Cards; a page holding twenty card faces goes in Uncut sheets, because it has to be cut up before any of those cards exist. That applies just as much to a retail game's token punchboard as to a PnP download. Drop a PDF on Components or Cards and taybl moves it to Uncut sheets for you — you will see it change group in the staged list.

When you navigate back to a draft, an Already uploaded list shows the files on the server with thumbnails, so you can see what landed and remove anything that should not be there. Each row carries the one action that applies to it — Cut it up on an image, Extract components on a PDF. You do not set types here; you confirm those on the confirm-your-components step once the build has run.

Rows that could be the rulebook also carry Rulebook, filled in on whichever file players get when they open the rules — from the table's Rules button and from the Read the rules button on your public game page. That is every PDF, plus anything typed as the rules document; press it on another of them to move the answer. This matters when a game ships more than one: taybl reads rules text out of every PDF, so several can be typed as rules, and Rulebook is how you say which one a player should actually open. Marking one un-marks the rest.

The Rulebook bucket takes a PDF, a text file, or an image (PNG, JPEG or WebP). A PDF with real text in it is read directly; anything else is read as pictures. The file players open is a separate, looser question, and you can set one from the Rulebook card on the Details tab after the build.

Slicing a sheet of cards

Games often ship one image holding a whole grid of card faces rather than twenty separate files. You do not have to crop them yourself. Drop the sheet on Uncut sheets and the slicing dialog opens on its own as soon as the upload finishes.

If you filed a sheet somewhere else and only spot it later, you can send it to the slicer from wherever you notice — in Already uploaded press the row's Cut it up button; in the review lists of Build it yourself and on the confirm-your-components step, set its Type to Sheet of cards (uncut), which is not just a label but opens the slicer. Once the sheet is cut it drops out of the list and its cards take its place.

Answering "Not a sheet" is reversible. The slicer's Not a sheet button keeps the file whole and types it as whatever Slice as says, which is the right answer for a file that only looked like a sheet. If it really did need cutting, find it in Already uploaded and press Cut it up — the slicer re-opens, and Slice as remembers what you told it the first time.

A preview draws the proposed cuts over your image so you can see the grid before anything happens. Nothing is cut until you click Slice.

• Click any cell to include or exclude it. Cells that are entirely blank are unticked for you, so a part-filled sheet just works — on a grid taybl found and on one you typed.
• If taybl cannot find the grid it says so, and offers to work it out for you — a closer look that can read a sheet the quick pass could not. You can also just type the rows and columns yourself, which is always the last word. Expect the quick pass to give up on sheets whose pieces touch with no gap between them, or that have artwork running behind the whole page: it finds cards by looking for uninterrupted background between them, and those sheets have none. "Work it out for you" is the one part of slicing that uses AI, so it is not offered in Build it yourself, nor in the Studio Files tab once your game is built — see Cutting sheets without AI.
• A page of rules is not offered a grid at all. When the closer look finds prose rather than pieces — a rulebook page, a components checklist, a setup sheet — it says so and offers Move to Rulebook instead of asking you to type a grid, because there is nothing on the page to cut. That files it where the AI reads it for rules text. This is easy to hit by accident: picking pages by hand in the print-and-play importer will import a rules page as happily as a card sheet if you tick it. If the verdict is wrong, type the rows and columns yourself and cut it as normal — you still have the last word. The same offer appears on the confirm-your-components step for a sheet you never opened the slicer on.
• Fit to the printed area is on by default, and it is what makes a typed grid land. Print-and-play pages usually leave a white margin, and dividing the paper by 11 x 10 puts every cut line slightly off. taybl measures where the printing actually is and divides that instead. Untick it to divide the page edge to edge.
• Nudge the fit corrects that measurement when the cuts still sit a little off. Plus is tighter, minus is wider, and the preview moves as you drag. This moves the cut lines — it is not the same as Trim edges below, which shaves each card after the lines are placed. It pulls all four edges together; for one edge on its own, drag block 1's frame on the page instead — see Moving and resizing a block. Touching either fit control hands the placement back to it.
• Slice as picks what the pieces become: card fronts, card backs, pieces or tokens. Fronts and backs are separate sheets, so slice each one in turn.
• Mirror columns is for a back sheet that came off a duplex printer reversed. Tick it and back #1 lines up with front #1.
• Trim edges shaves a percentage off every card after the cut lines are placed, for sheets printed with bleed or crop marks. Leave it at 0 unless you can see the trim is needed — cards with a deliberate white border look wrong once it is cut away. If the whole grid is offset rather than each card being too big, the control you want is Nudge the fit above.
• Merge identical cells is on by default, and it is worth understanding: you never need one asset per printed copy. A sheet of twenty identical card backs is one card back, and a sheet printing four copies of the same card is one card. taybl treats a component as a template and the number of copies as a quantity, so merging keeps your asset list readable and still deals the right number of cards. Turn it off only if you genuinely want the repeats as separate, separately-editable assets.

After slicing, each card becomes its own asset with its own thumbnail, and the original sheet steps aside — it stays in your storage but is no longer treated as a component, so the AI models the cards and not the page they came on.

Cutting a sheet again

A first cut is often not the right one — the grid read 3 x 2 where the page is 3 x 3, or the trim was too tight. Cut up… on the Files tab re-opens the slicer on a sheet you have already cut, and it opens knowing: it says how many pieces that sheet currently holds, and the green button reads Replace 9 pieces rather than Slice into 9.

Nothing is deleted until the new cut has succeeded. If the cut is refused — you are out of storage, or the grid produced no image — the pieces you already had are exactly where they were.

The one thing to know before you press it: a component using one of the old pieces does not follow it to the new one. The cut that replaces them cannot know which new piece a card was meant to be — that is the whole reason you are re-cutting — so a component pointing at a deleted piece keeps its place on the table and loses its artwork. The slicer counts those for you and says so before you commit; point them at the new pieces in the Layout Editor afterwards.

Deleting the old pieces by hand first still works, and leaves nothing to replace — but you do not have to.

Sheets get a larger upload allowance than ordinary art — see Limits below.

PDFs full of components

Many print-and-play games arrive as a single PDF holding everything — the card sheets, the player aids and the rules. Others split it up: a rules.pdf and a separate cards.pdf, plus a few token sheets. Either way, the files that hold components go on Uncut sheets; a rules-only PDF goes on Rulebook. taybl reads rules text out of every PDF regardless, so a misfiled one still teaches the game. An image works the other way round: only one you filed as the rules is read for them, because a photo of a rulebook page and a photo of the board are the same kind of file and only your answer tells them apart. Either way the bucket you choose decides whether taybl goes looking for components inside it.

A PDF on Rulebook is taken at its word. Nothing reads it for card sheets, no dialog opens on it, and it never turns up in the "your components are still inside this file" warning. Dropping it there is your answer to that question, and being asked it again a second later is just the same question twice. The rulebook you upload from the Details tab is treated the same way.

That same answer is what picks the file behind Read the rules, so a PDF you file here is the one players open. If you filed the wrong one, the Rulebook button on the row in Already uploaded moves it.

Replacing the rulebook from the Details tab swaps which file players open as soon as the upload lands — the new file becomes the answer, and the one it replaced is removed for you. Three cases keep the old file instead of deleting it, and the tab tells you which: pieces in your game were cut or imported out of it (a print-and-play bundle filed as the rulebook); you replaced a PDF or text file with a picture, so the words the AI reads on a rebuild are kept; or the delete itself failed. In every case the new rulebook is already the one in play, and the old file is on the Files tab if you want it gone.

A PDF on Uncut sheets is read the moment the upload finishes — its own text, no guessing and no AI cost — and taybl opens the extract dialog for you, one file at a time. A PDF dropped on Components or Cards moves to Uncut sheets first, so it gets the same treatment.

For those, it offers on any PDF that isn't obviously a rulebook, rather than only on ones that label themselves properly. Most PDFs don't label themselves — a plain export of card sheets says nothing about what it is — and staying quiet on those is how a game gets built with none of its card art and nothing anywhere saying so. A document that is solid prose on every one of several pages is read as a rulebook and left alone; anything shorter or more sparse gets offered.

It errs towards offering on purpose, so a short instruction sheet does get flagged sometimes. When that happens, press This is just rules in the dialog. That answer is remembered against the file: the warning stops naming it, in the uploader and on the review step, and it stays quiet after a reload. The file is not ignored — its rules text is still read when your game is built.

Either verdict is reversible, from the same button. Every PDF row keeps an Extract components button whichever bucket it came from — so a whole print-and-play bundle you dropped on Rulebook is one click from the importer. Open it and press It does hold components.

When the file describes itself

A well-made print-and-play labels every page, so the dialog can show you what it found:

• Pick an edition first. PnP files usually contain the same game more than once: full ink and low ink, or two wordings of the same cards. Importing all of them would create every card twice, so you choose one.
• Tick the sections you want. The deck, the player aids, each tier of goal cards. Everything in your chosen edition starts ticked — untick what you don't need.
• Nothing is created until you press Import components.

Importing brings the pages in; it does not cut them. Every page holding components arrives as an uncut sheet and goes straight into the slicing queue, so you see the proposed cuts drawn over each page and press Slice yourself. A deck's backs sheet is imported once however many sheets print it, and in the slicer you untick the repeats and cut a single back.

When it doesn't

Most PDFs are not labelled — a plain export of card sheets says nothing about itself, and there is nothing to read. Rather than send you away to export the art by hand, taybl renders every page as a thumbnail and lets you pick the pages holding components yourself, then say what they hold. The list is the same one the confirm-your-components step offers — board, cards, pieces, tokens, dice, an uncut sheet, rules, photos — so a print-and-play board has somewhere to go. Working through a run of sheets, the answer carries to the next one, so you set it once rather than per file.

taybl then picks what happens to each page, and says which under the dropdown:

• A page holding many things joins the slicing queue — cards, pieces, tokens. It arrives as an uncut sheet, exactly as an uploaded image sheet would, and you cut it with the preview.
• A page that IS one thing is imported whole — a board, a mat, a background, a photo. It is filed as what you said it is and left alone.

A tick box — More than one per page — overrides it either way, because some boards really do need cutting and some card pages are one picture. The sentence under the dropdown always tells you what pressing Import will actually do.

What starts ticked depends on how long the file is:

• A file of three pages or fewer opens with every page ticked. At that length it is a component sheet, not a rulebook, and a line above the grid tells you it happened. Untick anything that turns out to be rules or a cover.
• Anything longer opens with nothing ticked, because it may be mostly prose. Tick the pages you want, or press Select all and untick the few you don't. Clear unticks everything.

One import takes at most 60 pages, so Select all stops there on a very long document — import the rest in a second pass. If you tick more than 60 by hand, the button says so rather than failing when you press it.

Rendering the thumbnails takes a moment on a long document, which is why it only happens when the label parse comes back with nothing. The rules text from the PDF is used either way.

Why importing never cuts

Cutting takes one look from you first. taybl finds the cut lines by looking for the uninterrupted background between cards — which works on a tidy sheet, and quietly does not on a lot of real print-and-play art: cards printed edge to edge, or an illustration running behind the whole page. When a gutter is invisible, the reading that comes back is not "I cannot tell" but a grid with a line missing, so a 3 x 3 sheet reads as 3 x 2 and every cell in the bottom row holds two cards.

Importing used to cut on that reading without showing it to anyone, and the result was opening the slicer to cut a sheet and finding it already cut, into the wrong pieces. So the import stops at bringing the pages in. Every page holding components is stored whole, named -uncut, and queued for the slicer — where taybl works out the layout, draws the proposed cuts over your page, and waits.

Check the preview, adjust the rows and columns if it read the page wrong, and click Slice. (In Build it yourself, and in the Files tab of a game you have already built, there is no closer look — that step uses AI — so you type the rows and columns instead. See Cutting sheets without AI.)

A page you asked to keep whole is different: your board is filed as a board, carries no -uncut in its name, and never joins the queue.

Sheets that hold more than one grid

A print-and-play page is often not one tidy grid. One A4 routinely carries a row of tall cards, a wider reference card, and a dense field of round chits — three different piece sizes that share no row or column lines. A single rows × columns cannot describe that page: every set of cut lines that fits one block cuts the others wrongly.

Every sheet is cut in at least one block, and the slicer always shows the block editor — a card per block, plus its frame drawn on the page above. An ordinary sheet has one, and that one card is where its rows × columns live. When the closer look sees a mixed page it reports blocks — one per group of same-sized pieces — and you simply get a card each. Every block has its own name, its own rows × columns, its own Slice as, and two cut options:

• Name — what this block’s pieces are called. Type Chance and the block’s crops become Chance 1, Chance 2, and so on; the preview under the box shows the numbering as you type. Leave it blank and the crops fall back to a name worked out from the sheet’s filename, which after a print & play import is something like File 2 Uncut 01.
• Round — cut each piece as a circle rather than a square, leaving the corners transparent. Round chits are usually printed nearly touching, so a square crop carries crescents of the four neighbouring tokens into every piece. The overlay draws round blocks as circles so you can see what will actually be cut.
• Turn — rotate that block’s pieces a quarter, half or three-quarter turn, for a block printed sideways to fit the page. This turns the pixels as they are cut, which is the better fix while you still have the sheet. Art that is already uploaded and bound is turned in the inspector instead — see the face bindings in components.

Slice as is per block. There is no sheet-wide one, because a page holding two character cards and a field of round chits has no single answer — and on a sheet cut in one block, block 1's answer is the page's. This is not just a label: the type decides what each crop becomes, and only a block cut as Card (front) inherits the deck name you gave the page during import. A new block starts as whatever the last one was.

Naming a block here is the earliest you can name a component, and it is worth doing: these are the names the AI matches against the components your rulebook describes, so a block called Chance arrives as the Chance deck the rules talk about rather than an anonymous pile. Naming only sets what the pieces are called — the stored filenames stay as they are, so dropping in replacement artwork later still matches. You can rename any file afterwards from the Files tab or at the confirm-your-components step.

One more consequence: Merge identical cells only ever merges within a single block. Two crops that come out byte-identical in different blocks stay separate, because collapsing them would keep one type and quietly drop the other block's piece.

Blocks are cut in order — the whole first block, then the whole second — so a front sheet and a back sheet laid out the same way still pair up piece for piece, and Mirror columns reverses each row within its own block.

Round cuts do not need the closer look. An ordinary sheet that is all chits has Round and Turn on block 1 beside its rows × columns — you never have to split a page into blocks just to cut circles.

Moving and resizing a block

A proposed block is measured by eye and is often out. Select one — click its row, or its frame on the page — and it gets drag handles: pull a corner to resize two sides at once, an edge to move that one side only, or drag inside the frame to move the whole block. Holding ⇧ on a corner keeps the block’s proportions.

This is the answer when your cuts sit in the wrong place, and it works on an ordinary single-grid sheet exactly as it does on a split one — block 1 has a frame like any other. Where Nudge the fit pulls all four edges in or out together, dragging one edge fixes the common case it cannot express: artwork that runs to the paper on three sides and leaves a margin on the fourth.

Nothing snaps. A block is being fitted to printed artwork, so the only edge that is right is the one the art actually has — a cut that jumped to a tidy fraction of the page would be wrong every time. Cells belonging to the blocks you are not editing fade back, because a page mid-correction usually has several overlapping.

For an exact fit, the selected block also shows its X, Y, W and H in pixels of the sheet, with the sheet’s own size beside them. Measure a block of tokens in any image editor and type the four numbers straight in. They commit when you press Enter or leave the field — not as you type — so a four-digit value can be retyped without the block lurching about, and Escape abandons an edit. A number that would run off the page is brought back onto it rather than cutting nothing.

Add block splits a sheet taybl read as a single grid. Block 1 is already there and keeps everything you set on it — box, rows, columns, name and cut options — and a second block appears in the middle of the page for you to drag into place. Remove drops one, and is not offered on the only block: a sheet is cut in at least one. Adding is always available, so you never have to re-detect a page to correct how it was read.

One consequence worth knowing: taybl works out which slots are empty when it first reads the page, so moving a block or changing its counts discards that opinion for that block. Its empty slots stop being pre-unchecked and you untick them yourself — better than the alternative, which is un-ticking real pieces because the verdict was about a rectangle that has since moved.

Redoing an import

A file can only be imported once while the assets it produced still exist. Delete them and the file is importable again — so an import that came out wrong is undone by deleting what it made, exactly as the message says.

Revising the PDF is a different job, and deleting its pages is the wrong move for it — you would lose every binding those pages have. Replace the file instead: drop the new version on the Files tab under the same name and choose Replace, which clears the earlier import so the file can be imported again. The import dialog then offers “Replace the N pages I imported before” — leave it ticked and each new page swaps into the row its old version held, so components stay bound to their art. See re-importing a revised PDF.

Building before the sheets are cut

If you skip the extraction and press Build my game, taybl asks first. This matters more than it sounds: with the cards still inside the file, generation has no card art to work from, so it invents components out of the rules text instead — you get a game that looks nothing like the one you uploaded. The warning names the files that are still holding components, and offers to open the extractor. You can always build anyway.

A PDF has no Type dropdown on the confirm step, for the same reason: it is not a component, it is the container the components came out of, and once its contents are extracted it steps aside on its own. An uncut image sheet does have one — set to Sheet of cards (uncut) — because that is how you tell taybl to cut it.

Removing an image a zone or card already uses is safe: that art disappears from the table straight away, and the next time you Save in the Layout Editor taybl clears the leftover reference and tells you what it cleared. You never end up with a game you can't save.

Cutting sheets without AI

Everything above — the Uncut sheets bucket, the slicer, the print-and-play page picker, the warning that a file still holds its components — is also in Build it yourself, the no-AI path you pick when you create a game. You do not give up print-and-play to opt out of AI.

The same is true in the Studio Files tab, where you cut sheets after your game is built. Both are places you are working by hand, so neither spends on a model behind your back; the closer look stays in the build flows, where asking taybl to work things out for you is the point.

That works because cutting a sheet was never an AI feature. Reading a print-and-play PDF's structure is a parse of the page labels the file already prints on itself, and finding the cut lines is measurement: taybl looks for uninterrupted background running the full width or height of the page and cuts there. Both happen on our servers, from your file, with no model involved.

One thing is deliberately missing from both. Work it out for you — the closer look offered when the quick pass cannot find a grid — is a model call, so it is not offered in either place, and no page of your file is sent to an AI provider for it. Instead you type the rows and columns yourself, which was always the last word anyway. Be ready to use it: the quick pass gives up on sheets whose cards butt together with no gap, which is a lot of real print-and-play art.

One knock-on worth knowing in the Files tab: Move to Rulebook — the offer that appears when a page turns out to be prose rather than pieces — comes from that same closer look, so it will not appear there either. Change the row's type to Rules document instead, which does the same filing by hand.

The rest is the same. Cut components land in the ordinary type buckets — cards with cards, tokens with tokens — where you can rename them, correct a wrong type, or remove one. Merge identical cells still collapses a sheet of twenty identical backs into one asset with a quantity; where that asset ends up in a deck, the count comes with it, and the row shows a x4 badge either way so you can see what the number was.

And the same guard applies before you build: if a sheet is still uncut, taybl asks before building your table rather than quietly leaving that art out. You can always Build anyway and cut it later.

Limits

Images and PDFs get the larger allowance because a full page of high-resolution cards, or a whole print-and-play bundle, is legitimately several times the size of any single card; they are uploaded one at a time so the bigger ceiling is safe. Anything over its limit is rejected before upload. If you hit the per-game or total storage limit, remove unused assets from the Already Uploaded list — or, once the game is built, from the Files tab — and try again.

Naming your decks on the confirm step

The Component groups section on the confirm step is where you say which cards belong to which deck — "Gameplay Deck", "Goals: A King Tide" — and which back each deck uses. This is not a hint. A deck you name here is built as a component-group exactly as you described it, whatever the AI would otherwise have done, and taybl flags it as an error if the finished game spreads one of your decks across several groups.

Filling a group goes through Add components, which opens a grid of every front not yet spoken for. It shows you the art rather than the filename: cards cut from a sheet come out named Deck_2_r1c3, Deck_2_r1c4, and the picture is the only thing that tells one from the next. Each is a card-shaped tile with the whole face in it — nothing cropped off the corners — and the columns refill to whatever width the panel has. Click a tile to select it, click again to change your mind, then Add puts the lot in the group at once. Rest on a tile for half a second (or tab to it) and a bigger peek opens beside it carrying the full, untruncated file name; double-click a tile to throw it up full screen, and Escape or a click outside puts that away.

If you imported a print-and-play PDF, the sections you ticked are already filled in as decks — one per section of card fronts — so you usually only have to check them rather than build them.

You do not need card backs to name a deck. When no backs were uploaded the back-pairing choices disappear (there is nothing to pick) and you are left with just the grouping, which still gets honoured.

Optional: help the AI

Two optional fields below the buckets bias the generation toward the right kind of game. Neither is required.

What happens while it generates

After you click Build my game, the pipeline runs through a series of steps. You watch a live checklist with a progress bar and an activity ticker; the status survives a reload, and a couple of steps may pause for your input (for example, confirming how the AI classified your component and card images).

When it finishes, the game becomes Ready and opens in the Studio. The generated game is a sandbox table — it never includes a procedures block; players resolve every rule by hand.

If the build says it couldn’t read your rules

taybl tries twice — pulling the text out of the file, then reading the pages as pictures — so this message means neither turned up any words. Almost always the pages are too small, too soft, or too handwritten to make out. Any of these fixes it:

• Re-export the rulebook at a higher resolution.
• Photograph the page again in better light, filling the frame with it.
• Paste the rules into a plain .txt file and upload that instead. It is the surest thing to hand the AI, and it takes a minute.

The file players open is a separate choice and does not have to change with it. Set that one from the Rulebook card on the Details tab after the build, and your original scan can stay exactly as it is.

After the draft

The uploader is a pre-build screen: it disappears once the game is Ready, and the Files tab takes over. Everything above is still available there — adding art, cutting a sheet, importing the pages of a print & play PDF — plus replacing a file in place so every component already using it picks up the new art.

Treat the first draft as exactly that. Open the Studio tour to find your way around, then refine the table on the Layout and Details tabs — or describe changes in plain language and let the AI revise. See refining and regenerating for the refine loop and how to update a game when the schema advances.

Related

• What taybl is
• Studio tour
• The Files tab — uploads after the game is built
• Refining and regenerating