Conversion recipes
Jobs, not formats. Each one is a chain of routes this converter really performs, and each step says what it costs.
Every step below is a live route. Nothing here is a plan for a conversion the site cannot do, because the list refuses to build if a step names a route that does not exist. Run them in the converter, one hop at a time.
Menu or price list export to an editable workbook and a printable sheet
A till, supplier portal or booking system exports the menu as CSV. Front of house needs one workbook staff can edit and one page that prints without a spreadsheet app.
Gives the flat export a real worksheet people can sort, filter and hand around.
What it costs: Cell types. A delimited file carries none, so every value is written as text: a price column will not sum and a date column sorts alphabetically until someone retypes it. The first row is taken as the header row, rows whose every field is empty are dropped, and the worksheet is always called Sheet1 rather than named after the file.
Excel workbook to PDF document
Produces the fixed page for the wall or the clipboard.
What it costs: Everything except the cell values. The workbook style table is never read, so fonts, fills, borders, column widths and number formats are gone, and a date or currency cell prints as the raw number the file stores. Formulas are not read either, only the values last saved with them. Charts, images and any tab that is not a worksheet are skipped. The page itself is plain Helvetica in tab-separated columns, with no rules and no grid.
The PDF writer can only draw Latin-1 characters. A pound or degree sign survives; a euro sign, a curly quote or any non-Latin script becomes a question mark, so read the PDF before it goes on the wall.
RAR-packed scan or comic set to one ordered PDF
A set of scanned pages arrives as a CBR, which is a RAR archive. The job is one PDF that opens anywhere and keeps the pages in the right order.
Comic Book RAR to Comic Book ZIP
Moves the pages out of RAR into the open ZIP container the rest of these routes read.
What it costs: Everything in the archive that is not a JPEG or PNG page. An entry survives only when its name ends .jpg, .jpeg or .png and its bytes carry the matching signature, so a ComicInfo.xml, a thumbnail, a WebP page and the folder layout are all dropped. Kept pages are renumbered page-0001 upward, so the original filenames are gone. The image bytes themselves are copied across untouched, so no picture quality is lost.
Comic Book ZIP to PDF document
Binds the ordered page set into a single document that prints and shares as one file.
What it costs: Text. Each page stays a picture, embedded as it is with no re-encoding, so nothing in the PDF is searchable or selectable and no text is recognised. Each page is sized from its own pixel dimensions, so the document declares no paper size and no print resolution, and pages that were different sizes stay different sizes.
Page order comes from the filenames and is compared numerically, so a set named 1, 2, 10 lands in that order rather than 1, 10, 2. A set with no numbers in its names is ordered alphabetically, which may not be the reading order.
Styled ASS or SSA captions to portable SRT and a plain transcript
A video came back with a styled ASS subtitle script. The publishing platform wants SubRip, and the writer wants a transcript to quote from.
Advanced SubStation Alpha subtitles to SubRip subtitles
SubRip is the caption format every player and platform accepts, and it is what the transcript step reads.
What it costs: Every part of a cue except its start time, end time and text. Inline override tags in braces are stripped, and the style, layer, actor, margin and effect fields are read and then discarded, so fonts, colours, on-screen positioning and karaoke timing do not survive. Cues are renumbered from 1 in file order.
SubRip subtitles to Plain text
Gives the writer a clean reading transcript with nothing to strip out by hand.
What it costs: All timing. The file becomes the cue texts in order, separated by blank lines, with no cue numbers and no timestamps, so a line can no longer be located in the video. Line breaks written inside a cue are kept as they were.
An ASS line break is there for on-screen layout rather than for the sentence, and it arrives in the transcript as a real newline. Expect to rejoin some lines by hand.
Phone or camera clip to a shareable MP4 and a silent preview loop
A QuickTime clip off a phone or camera has to go on a page: one MP4 that plays everywhere, plus a small looping preview for a slot that cannot autoplay video with sound.
H.264 video and AAC audio in an MP4 with the index moved to the front is the combination that plays in every browser and starts before the file has finished downloading.
What it costs: All container metadata and every extra stream. Metadata mapping is switched off, so the title, creation date, camera make and model and any location tag are stripped. No stream mapping is set either, so only one video stream and one audio stream are carried and a second audio language or an alternate angle is dropped. Both streams are re-encoded rather than copied, so this is a full generation of quality loss, and the encoder runs at its fastest preset, which spends file size to save time.
MP4 video to GIF image or animation
A GIF loops inline with no player, no autoplay policy and no sound, which is what a preview slot needs.
What it costs: Sound entirely, because a GIF has no audio track. Motion is decimated to 12 frames a second and the picture is resampled toward a 960 pixel wide target, so fast movement stutters and fine detail softens. A GIF frame holds at most 256 colours, so gradients band, and the file is usually far larger on disk than the MP4 it came from.
The preview is built from the delivered MP4 rather than from the original, so what loops on the page is the footage that plays when it is clicked. That is a second generation of loss, which is the right trade for a thumbnail and the wrong one for anything else.
Vector logo or diagram to raster images a page can serve
An SVG logo or chart has to ship as ordinary images: a lossless PNG to keep as the master, and a smaller WebP for the page itself.
Produces the pixel master every later raster step is built from, including the compressed web copy.
What it costs: Resolution independence. The drawing is rendered once at one size and everything in it becomes pixels, so text is no longer selectable or editable and enlarging the result will show it. The size comes from the SVG's own absolute width and height, falling back to its viewBox and then to a square default when it declares neither, and an oversized drawing is scaled down proportionally to fit a maximum edge. The canvas is 8-bit sRGB, so no colour profile carries over.
Cuts the bytes a visitor downloads while keeping the transparency the PNG had.
What it costs: Exact pixels. The browser encoder writes WebP at the chosen quality setting, so the result is lossy and no longer byte for byte the PNG, though the alpha channel is kept and no white matte is painted behind it. Nothing else carries over either: a canvas holds only pixels, so any EXIF, XMP, IPTC or ICC block in the source is already gone by this point.
Keep the PNG. It is the only lossless copy in the pack, so every further size or format should be cut from it rather than from the WebP, which would stack a second lossy pass on the first.
SQLite export to a workbook whose numbers are still numbers
A tool or a phone app exported its data as a .sqlite file. Someone who does not write SQL has to read it, sort it and total a column.
Reads every user table into records while the values still carry their types, which is what lets the next step write real numeric cells.
What it costs: The schema. Column types, primary and foreign keys, indexes, views, triggers and the database's own internal tables are not carried across; only the rows of user tables are. A blob becomes a base64 wrapper object, a 64-bit integer becomes a string so that it cannot lose precision, and a NULL becomes JSON null, which the workbook step writes as an empty cell.
Writes one worksheet per table and keeps numbers and booleans as spreadsheet numbers and booleans, so a column actually sums.
What it costs: Nesting and long names. A value that is not a plain scalar is flattened into its JSON text, so the base64 blob wrapper lands in a cell as a string, and a NaN or an infinity is written as text because a worksheet cannot hold one. A table name too long for a worksheet name is truncated to fit, and two tables that are identical up to that point cannot both be written. Column order is the union of the keys seen across the rows, not the table's own column order.
Go through JSON rather than converting the database straight to a workbook. The direct sqlite to xlsx route writes every cell as text, so a totals row comes back as 0, while this route keeps numeric columns numeric. The trade is that a table with no rows stops the workbook step, because a worksheet needs at least one column, and a database with more tables than the worksheet limit is truncated without a warning.