v1.0.0-beta.3

1.0.0-beta.3 – 2026-09-11

Added

• Text styling. A caption can carry a contrasting edge, derived from
  the text colour rather than picked (#23), a plate that takes the photo
  out from behind it (#24), one of three system font families (#25),
  an alignment for its lines (#29) and bold, italic, underline and
  strikethrough in any combination (#34). Each applies to the next
  caption and to a selected one alike
• The text controls and the colour picker float next to what they
  change: the overlay while a caption is typed, the selection
  otherwise. The bottom panel keeps the tool picker, the colour default
  and the size slider (#37)
• Lines and arrows snap to 45° steps while Shift or Ctrl is held, and
  the key can be pressed or released mid-drag (#38)
• Oval redactions, inscribed in the same drag as the rectangle (#31)
• Highlights, shadows and a vignette in the adjust panel (#21)
• The saved JPEG carries the source's Exif, XMP, ICC and IPTC blocks,
  with the orientation reset, the size corrected and the embedded
  thumbnail removed. Only when the editor was handed bytes: a URL gives
  it no source to copy from (#22)
• Before opening an image larger than the browser will paint, the
  editor says what it would shrink it to and asks. ExportResult
  gains downscaled for hosts that want to warn again before
  overwriting (#20)
• The edit history opens from the back arrow, its entries are radios
  that announce the step they land on, and a long list scrolls (#15,
  #16)

Fixed

• A redaction hid the source pixels, not what was under it: annotations
  drawn below it and the adjustments applied to the image stayed
  readable in the export. It now samples the layer as drawn (#30)
• The embedded camera thumbnail was unlinked but its pixels stayed in
  the file and could be carved back out, the uncropped, unredacted
  frame included. Every strip of it is zeroed before the link is cut
  (#35)
• A state stored by an older release, missing the adjustments added
  since, rendered black: an undefined value ran the filter and left NaN
  in the node. reset and initialState now fill in what is missing
  (#36)
• A picture shown smaller than half its size is drawn from a halved
  pyramid instead of being sampled in one pass, so fine grain and thin
  lines match what an <img> shows. Export and thumbnails still scale
  in one pass (#33)
• The text overlay opened 16px left and 56px above the click, and the
  selection toolbar sat off-centre by the same amount: both were placed
  in stage coordinates but positioned inside the padded viewport (#37)

Changed

• Saving defaults to the format the image arrived in rather than PNG,
  so a photo is saved as a photo. Hosts asking for a format still get
  it (#22)
• The text controls read like the rest of the panel and share one row
  (#26, #32)

Still missing

• No translations yet: the Transifex resource behind l10n/ is not set
  up, so every string falls back to English
• Annotations cannot be created from the keyboard alone, and the
  editor's shortcuts are bound to the window rather than to itself
• Export and preset thumbnails still downscale in one pass, so a saved
  copy bound by maxSize is coarser than the same picture on screen
• The eraser, multi-select, per-annotation opacity, freehand redaction
  and exact-pixel crop are not here yet (#2)