Tuner (see TUNER.md): a draggable, collapsible dev panel rendering every
group registered via defineTunables() as slider + number-input rows.
- Live: physics/light/material/halo values are read per frame; formation
values trigger a debounced re-sample
- Values persist to localStorage; modified labels tint accent; Copy puts a
JSON snapshot on the clipboard for baking back into defaults; Reset clears
- Shown in dev builds or with ?tuner=1; drag position and section collapse
state persist
- Policy (documented): every new tweakable constant registers with the tuner
by default
- Hero groups: Formation, Physics, Cursor & Light, Material, Halo
Formation: the hex lattice read too uniform. Balls are now placed by
multi-pass dart throwing over the glyph mask — big balls claim spots first,
smaller ones settle into leftover gaps, random depth within the slab, 3D
overlap rejection via spatial hash — a faked settled-pile impression, not a
physics solve. Stroke-margin test keeps big balls inside the letterforms so
the type stays crisp.
- Ball sizes now vary organically: smooth spatial value noise drifts size
clusters across the word, white noise roughens per ball; the neighbor cap
keeps every draw collision-safe
- Radial halo billboard tracks the cursor BEHIND the text plane, so balls
silhouette against the glow (additive in dark, soft accent wash in light)
- Cursor point light switched to physical inverse-square falloff (decay 2):
balls near the cursor are genuinely brighter; the light dims to 30% when
the pointer leaves so the resting state stays moody
- Dark mode is an eclipse rig: near-black lacquer balls (clearcoat physical
material), almost no front light, cool key + counter rim from behind, and
no environment map — scene.environment floods instanced materials at full
strength regardless of envMapIntensity, which was washing the whole word
to gray; light mode keeps the procedural room reflections
The hero brand mark is now the signature interactive moment: 'artcraft'
rasterized per-glyph (manual advances so no ligature shaping), sampled onto a
jittered close-packed lattice 2-3 layers deep, and rendered as instanced 3D
spheres under a perspective camera fitted so world units equal CSS pixels at
the text plane.
- Cursor is both a point light and a push force (radial + swirl, quadratic
falloff, toward-viewer pop); balls scatter, then rejoin on home springs
- Per-ball radii capped at half the nearest-neighbor gap, plus a 3D
spatial-hash collision pass, so spheres never clip through each other
- Mount plays an assemble: balls fly in from a scattered shell
- Idle: light wanders a slow Lissajous path; balls breathe subtly
- Fallbacks: static display-type wordmark for SSR/crawlers, reduced motion,
and WebGL failure; theme-aware colors via shared token watcher (extracted
to lib/theme-colors, now also used by the blocking/render comparator)
- Tagline 'Controllable AI for artists.' remains the crawlable h1
Standalone Next 16 app at frontend/apps/artcraft-website-next (excluded from
the npm workspace like artcraft-dashboard; React 19 + R3F 9 + GSAP + Lenis +
Tailwind 4). Branched from design/prototype0; the Marathon-style navbar is
ported to next/link with theme-aware tokens plus a light/dark toggle.
- Design system: hairline-grid 'instrument panel' language; light (blueprint
paper) + dark (viewport terminal) themes via CSS tokens, system-pref default
with persisted override applied pre-paint
- Landing: hero, capability ticker, feature grid (real product footage),
ownership, made-with (click-to-load YouTube), CTA, footer — all crawlable
SSR HTML; GSAP reveals are progressive enhancement only
- 3D hero: interactive blocking↔render comparator — wireframe R3F set clipped
over the product reel by a cursor-driven scan line; degrades to plain video
on reduced-motion/coarse-pointer/WebGL failure
- SEO: metadata + OG, sitemap.ts, robots.ts, SoftwareApplication JSON-LD
## Fixes
- Slow paste in prompt box: `execCommand("insertText")` is quadratic per
line in Chromium - now inserts the clipboard as a single text node
(linear)
- Model selector submenu not scrolling in focus mode: the modal's scroll
lock swallows wheel/touch on body-portaled panels - flyout now stops
propagation (same pattern as mention dropdown)
- Cap the always-expanded deck panel at container width so cards wrap
into rows; scroll past 3 rows instead of squeezing the editor
- Drop shrink-0 on the expanded deck root (min-w-0 max-w-full) so it
shrinks inside the webapp's flex-row wrapper
- Collapsed fan badge shows total ref count instead of +hidden
MentionTextarea's root wrapper lacked min-h-0, so as a flex child of the
fullscreen modal's bounded column it kept full content height and got
clipped by overflow-hidden instead of letting the editor scroll.
Removes Font Awesome entirely
### Changes
- All ~1,000 icon call sites now import Lucide components directly from
`lucide-react` (`<PlayIcon className="..." />`); no FA packages, tokens,
or `fa*` identifiers remain anywhere.
- New `@storyteller/icons` lib holds the pieces Lucide doesn't ship:
`DynamicIcon` for runtime-chosen icons (`icon={loading ?
LoaderCircleIcon : SaveIcon}`), 11 brand logos (Discord, TikTok, ...) as
inlined CC BY 4.0 paths, and HD/SD quality badges drawn in Lucide's
stroke style.
- Global `.lucide { display: inline-block; width/height: 1.2em }` rule
in each app stylesheet: keeps icons font-relative like FA (1.2em
compensates Lucide's 24px-viewBox/20px-grid padding) and undoes Tailwind
preflight's `svg { display: block }`. Explicit `w-*`/`size-*` utilities
still win.
- FA-only props migrated: `spin` → `animate-spin`, FA size strings → em
values.
- Dropped `@fortawesome/*` + `@awesome.me` kit deps, both `.npmrc`
files, and
the FA rule in `shared-vite-config.ts` (now externalizes
`lucide-react`).
- Select multiple items in the library and add tags to all of them at
once
- Tag suggestions pop up as you type, based on your existing tags
- Fixed the selection bar looking squished (buttons no longer wrap onto
two lines)
- Video generation now defaults to Seedance 2.5 (was 2.0)
- Omni Reference is the default video input mode for everyone
- Model pickers on the create pages now group models by family
(Seedance, Veo, Kling, etc.) with a flyout submenu per family, so the
list is no longer one long scroll
- The flyout opens on hover or click, and automatically scrolls the
currently selected model into view
- New setting in **Settings → General** ("Group models by family") to
switch back to the classic flat list. Grouped is the default and works
in both the webapp and the desktop app
- Model rows now use full-color brand icons, while the picker button
keeps its existing look
- Video models show capability tags: **Audio Support**, **Start/End**,
and **Reference**
- Short model lists (under 8 models) stay flat, and mobile keeps the
existing bottom-sheet list
- Fix: videos in the media viewer can now go fullscreen (the expand
button was disabled)
## Summary
New marketing page at /minimax-h3 where visitors can try MiniMax H3 for
free,
right on the page.
- Hero is a working promptbox, locked to MiniMax H3 with a "Free" badge
- Full settings available: aspect ratio, resolution, duration, sound,
start/end frames
- New users sign up in a modal when they hit Generate (Google or email),
then their prompt runs immediately without leaving the page
- Generated videos show up right below the promptbox, with a queue
notice and a CTA to keep creating in the app while they wait
- Example videos from MiniMax's announcement, self-hosted with
thumbnails, only load when played, credited to MiniMax
- Overview, highlights, prompt tips, and FAQ copy making the free offer
clear Website promptbox updated to match the current webapp promptbox
(reference deck, drag & drop, focus mode, same buttons)
- SEO, redirects, and URL aliases (/minimax, /hailuo) wired up
- Characters modal: new eye button on character cards to view the image
full size
- Mention chip menu: tapping the avatar thumbnail also opens the
full-size view
- Card buttons (view/edit/delete) are always visible on touch devices
instead of hover-only
- Mobile: Characters open in a bottom drawer with a 2-column grid,
matching the other settings drawers
- All mobile drawers now have a close (X) button
- All mobile drawers can be pulled down to dismiss - the sheet follows
your thumb and closes from where you let go
- Mobile: mention autocomplete list can now be scrolled - dragging no
longer selects an item, only a tap does
- Fixed taps on autocomplete items sometimes not registering on
Android/iOS
- Deleting the space after a mention now reopens the autocomplete next
to the mention instead of at the top of the screen
- Clicking a mention chip while its menu is open now closes the menu
instead of closing and reopening it (desktop + mobile)
- Added missing display names for Seedance 2.5 Preview and ~25 other
models (Vidu Q3, Veo 3.1 Lite, Seedream 5.0 Pro, new Hunyuan 3D 3.1
variants, Marble 1.x, and more) so the task queue and history show
proper names instead of raw IDs.
- Vidu, Recraft, and SeedEdit models now show their brand icons instead
of the generic one.
## Fixes & improvements
### Model picker
- Removed the "Seedance 2.5 - SOON" placeholder from the video model
picker; it was confusing now that Seedance 2.5 Preview is live.
### "Not all media files could be found" error
- Fixed generations failing when the same media file ended up twice in
one reference list (e.g. added as two references, or the same file
uploaded twice).
- Requests now automatically drop duplicate media references before
sending - applies to image, video, audio, mesh, and splat generation on
both the web app and the desktop app.
### Gallery picker UX
- Media already added to a slot is now greyed out with an "Added" badge
in the library picker, so the same file can't be added twice to the same
field.
- Applies per slot across the create pages (video, image, audio, 3D
object, world) on web and desktop - references, start frame, end frame,
video refs, audio refs, and multi-view slots.
- Reusing the same image across different slots (e.g. start *and* end
frame, or two 3D views) is still allowed.
### What
- Drag files onto the promptbox on the Image, Video, and Audio create
pages to add them as references - same look and behavior as the webapp
- On those pages the promptbox is now the only drop target; dropping
elsewhere does nothing
- Every other page keeps the existing upload-modal behavior
### Notes
- The webapp's drop code doesn't work as-is on desktop - Tauri
intercepts file drags before the page sees them, so desktop needed its
own path
- Both now share one component, so the two apps stay in sync going
forward
- Model limits still apply: each box only accepts the reference types
its model supports, and existing slot/duration caps are unchanged
- **Hand-typed mentions always take the character's exact name** -
typing `@person` for a character named `Person` now converts to
`@Person` as you type, so the tag binds correctly instead of silently
failing at generation time.
- **Prompt input no longer freezes at a character tag** - the caret
could get stuck inside a chip (e.g. after deleting the space behind it),
blocking all typing.
- **Reference add-menu no longer hides behind the sidebar** - it now
renders above everything instead of getting clipped by the promptbox
layers.
- **Character autocomplete is scrollable in focus mode** - the
fullscreen prompt's scroll lock was blocking it; same fix for the chip
menu's Replace list.
- Fix popovers (model selector etc.) closing instantly when the mention
textarea was focused
- Fix mention autocomplete not clickable in prompt focus mode
- Fix chip "Replace" reopening the autocomplete - now confirms
immediately
- Fix autocomplete appearing at top-left of screen when caret is next to
a leading chip
- Don't open the autocomplete when clicking the caret onto an
already-confirmed chip
- **Fixed: picking an audio reference on mobile opened the photo/video
picker.** iPhones don't have an audio library, so the browser fell back
to Photos - the only thing you could "attach as audio" was a video, and
it silently uploaded as an audio ref. Audio pickers now open the Files
app filtered to real audio files (MP3, WAV, M4A, FLAC…).
- **Non-audio files are now rejected with a clear error toast** instead
of being silently attached (also covers drag & drop and Android's "any
file" picker).
- **Fixed: limit warnings never appeared in the web app.** Errors like
"Total audio duration cannot exceed 15s" from shared components were
firing into a toast system the web app doesn't render. They're now
routed to the app's own toasts, so length/limit errors actually show up.
- Applies to the create video, create audio, and desktop reference deck
flows
### New stuff
- Character `@mentions` in the video promptbox now show as **chips with
a thumbnail**, so you can see exactly which character is attached to the
prompt
- Clicking a chip opens a menu to **Replace** (pick another character),
**Preview** (view their image), or **Remove** the mention
- Chips behave like a single unit - one backspace deletes the whole
mention, and removing/replacing can be undone with Ctrl+Z
- Copying prompt text still gives clean plain text (`@name ...`), and
pasting it back restores the chips
- When two characters share a name, the one you actually picked is now
the one that gets sent
### Fixes
- Mobile: the `@` autocomplete no longer goes off-screen - it stays
inside the viewport, flips below the caret when needed, and taps select
reliably
- Focus mode: the autocomplete no longer pushes the prompt sideways /
causes overflow
- Enter-to-generate now works while the mention editor is active on the
webapp
- Create flow stored `inference_job_token` as the `character_token`, so
mentioning
a new character sent a bogus token - backend drops it and name-matches
the
raw `@name` against stale characters.
- Don't add pending characters to the mention store; they aren't
generatable yet.
- Keep the pending poll running after the modal closes and sync real
tokens into
the store once the character is active.
- Remove unused `addCharacter` store action.
- Same fix in all three copies: webapp, website, shared `promptbox` lib.
Ports the Video Frame Extractor from the desktop app to the webapp at
`/frame-extractor`, with a reworked UI.
### What's new
- New Frame Extractor page: upload a video, pick one from your library,
or drag and drop
- Scrub to any moment and capture the exact frame at the playhead with
one click
- Frame stepping buttons and keyboard shortcuts (arrows to step, Space
to play/pause, C or Enter to capture)
- Burst capture: grab a sequence of frames to find the sharpest one,
with progress and cancel
- Each captured frame can be used as an image reference, animated into a
video, saved to the library, or downloaded
- Save all and clear actions for captured frames
### Integrations
- "Extract frames" action on library videos (selection bar and lightbox)
opens the extractor with that video preloaded
- Sends frames straight into the create-image / create-video reference
deck, same flow as "Send to prompt"
- New sidebar entry (purple NEW badge) and home page card
- Page is public; saving and references prompt sign-up when logged out
- Adds a **Send to prompt** action to the library's multi-select bar
with two
targets: **Create image** and **Create video**
- Selected images are attached as reference images on the target page
without
touching the user's typed prompt or settings; selection is cleared and
the
app navigates to the create page
- Selected videos can be sent to the video prompt as reference videos;
models
without video-reference support (only Seedance has it) get an
explanatory
toast instead of a silent drop
- Media travels via `pendingRefImages` / `pendingRefVideos` store slots
(mirroring the existing `pendingRecreate` pattern) and is merged on the
receiving page once its model list is loaded, so the **real per-model
reference caps** apply - fixes silent no-ops when refs were already
attached
- Merging dedupes by media token, clamps to the model cap, and toasts
partial
adds ("Added 3 of 5 …"), full decks, and unsupported models
- Keyframe-only video models take the first sent image as the start
frame;
reference-capable models switch into reference mode so the deck is
visible
- DRY: shared `galleryItemToRefImage` mapper (reused by the create-image
library picker) and `buildVideoRefsToAdd` (reused by the video-ref
picker,
duration probing + total-duration cap)
- Bar UX: option list adapts to the selection (video-only hides Create
image),
icon-only button on phones to avoid overflowing the bar
BFlat