Skip to content

v1.1.0

Choose a tag to compare

@Drommedhar Drommedhar released this 05 Jun 09:09
· 13 commits to main since this release

Added

  • Generative AI (experimental) — a new opt-in image-generation tier driven by a local ComfyUI
    runtime that Sable manages and runs in the background. Turn it on in Preferences ▸ Machine
    Learning
    ; nothing is installed until you do. On enabling, Sable reuses an existing same-OS ComfyUI
    install if it finds one (models are referenced in place, never copied), otherwise it installs its own
    PyTorch + ComfyUI into Sable's folder (a multi-gigabyte download).
    • Generative Fill (AI ▸ Generative Fill…) — fill the current selection from a text prompt.
    • Generate Image (AI ▸ Generate Image…) — text-to-image.
    • A generative panel with a model preset, compatible LoRAs, prompt and negative prompt, steps,
      CFG scale, denoise, seed (with a Random button), and a CPU-offload option to fit large models.
  • Reuse models from an existing ComfyUI installAI ▸ Models gains a Model sources section
    to add a ComfyUI (or compatible) folder; its checkpoints and LoRAs are referenced in place, never
    copied, plus a Check sidecar / Python env action to verify the runtime.
  • Tidier update notes — the "update available" dialog now groups the release notes into
    Added / Changed / Fixed tabs with a collapsible entry per version (and still combines every
    version you've skipped into one changelog).

Changed

  • Full interface translation — every remaining piece of on-screen text now follows the selected
    language, not just the menus and panels. This covers tool names and their tips, the status bar
    (cursor position, selection size, document info, zoom), all dialog titles and messages (open/save/
    export, AI tasks, update checks, crash recovery), the command palette, the model manager, the
    settings pages, the New Document presets, and default layer names. German translations are included
    for all of it; switch languages in Preferences ▸ User Interface.

Fixed

  • Stamped out the last of the app-wide swallowed clicks — the canvas render loop now runs at input
    priority and prefers any non-blocking display mode (Mailbox → Immediate → relaxed v-sync), so buttons
    and menus no longer occasionally flash pressed without actually firing.
  • Window minimise/maximise/close buttons fire reliably — the title-bar caption buttons no longer
    ignore clicks (including at the very top edge and corner). They now act the instant you click them.
  • Colour panel tabs (Colour / Gradients / Swatches / Histogram) switch on the first click — a slightly
    imperfect or near-edge click no longer does nothing; the whole tab is clickable and responds immediately.

Full Changelog: v1.0.5...v1.1.0