v5.6.11

What's new in v5.6.11

Generate

• Fixed a setup failure that could leave the send button permanently disabled. If models were downloaded before ComfyUI's source finished installing, setup could silently fail to put ComfyUI in place — everything looked installed, but the send button stayed off saying to run Setup Requirements. Setup now recovers from this automatically (and repairs existing installs stuck in this state: just run Setup Requirements once).

v5.6.10

What's new in v5.6.10

Downloader

• Danbooru works without an account again. Danbooru had started rejecting the app's requests (HTTP 403 on searches and file downloads). The app now identifies itself honestly to Danbooru, which restores anonymous downloading — no API key needed (2 tags max per search on free/anonymous, per Danbooru's rules).
• Clearer error messages. When a source rejects a request, the log now shows the site's own explanation — e.g. Danbooru's "You cannot search for more than 2 tags at a time" with a hint about the free-account limit — instead of a generic "check the API key".
• Error badge on Activity. If an error is logged while the Activity console is closed, a red dot appears on the Activity button so you know to open it.
• Pexels & Pixabay attribution. The source card now links the provider, and downloads record the photographer/author credit in the .txt sidecar (Pexels videos now get a sidecar with the videographer credit too).

v5.6.9

What's new in v5.6.9

LoRAs

• LTX-2.3 now applies your LoRAs. Selected LoRAs were silently ignored by the LTX-2.3 (22B) video model — they now chain into both sampling stages, so style/motion/character LoRAs actually take effect.
• Krea 2 LoRAs are recognized. The picker identifies Krea 2 LoRAs (both Civitai/ai-toolkit trained files and the official Krea releases like sunsetblur, which ship with no metadata) and files them under the Krea 2 tab with their own badge — no more "base?".

LoRA previews

• Animated thumbnails. LoRAs whose Civitai page only has video previews used to get no thumbnail at all. They now get a poster image — and a looping animated preview in the picker when ffmpeg is available (found automatically on PATH or from an installed Pixal3D).
• Right-click → set thumbnail now accepts animated GIFs too.

v5.6.8

What's new in v5.6.8

• Fixed: sending a video to the AI chat crashed installed builds. The chat's video frame sampler could be the very first thing to touch the VLC library, before the app had located your VLC install — the failed library load brought the app down (crash code 0xC06D007E). It now runs the VLC setup first, same as the video player. If VLC isn't installed at all, the chat shows a clear message instead of crashing.
• Crash minidumps (added in v5.6.7) now write correctly.

v5.6.7

What's new in v5.6.7

• Crash reporting. If the app ever crashes, it now writes a report to %APPDATA%\Clarity TagFlow\crash reports\ — the error, the module that failed, a debugger-ready minidump, and a log of everything the app's native libraries printed. If you hit a crash, sending the newest files from that folder makes it fixable.

v5.6.6

What's new in v5.6.6

Important fix

• Sending a video to the AI chat no longer crashes the installed app. A memory bug in the video frame sampler corrupted the heap on release builds (dev builds masked it); the fix also hardens video posters and playback.

New

• SVG support — .svg files now show up in the gallery and viewer like any other image, always razor-sharp at any zoom (they're vector, so they render at whatever size you view them). They can be AI-tagged, deep-scanned, and backed up too.
• AI chat media viewer — attachment thumbnails in the conversation are bigger, and clicking one opens it large over the app: images in high resolution, and videos actually play (with sound, honoring your loop setting). Close with Esc, the ✕, or a click outside.

v5.6.5

What's new in v5.6.5

AI Chat

• Voice dictation — a mic button in the message box, powered by Whisper running fully on your device. First click downloads the speech model (~148 MB, also in Get Models); then click to record, click again to stop, and your words land in the message box. Multilingual, works completely offline.
• Model picker — a dropdown next to the mic switches between your installed chat models without opening Settings.
• The send button now shows only when there's something to send, in blue.
• The Tools popup got the Apple-style card/switch look and stays open while you toggle; turning Role playing on or off starts a fresh chat.
• Redesigned "Thinking" section: smoother animation, and thoughts display as an elegant quote block.

Look & feel

• The AI orb is brighter and easier to see on every theme, with its classic light-blue/white glow restored.
• The voice-sample recorder shows a live waveform of what it's capturing, Voice-Memos style.

Fixes

• Renaming a caption-length tag no longer clips off the panel edge, oversizes, or overlaps neighbouring tags.
• Fixed a crash when the AI chat panel was squeezed very narrow.

v5.6.3

v5.6.3: multi-select media filters, AI model toggles, glass-light inp…

v5.6.2 — Crash fix & Pixal3D polish

Crash fix and a cleaner Pixal3D.

Fixes

• Fixed a crash when switching from a Text to Image tab (e.g. Krea 2) to Text to Video before any videos had been generated — the app could panic and close. Selections now carry over safely between generator views.

Pixal3D

• Redesigned in the same Apple style as the rest of the app: grouped Generation, Quality, and GLB Extraction cards with toggle switches and tidy rows.
• The panel now scrolls, so nothing is cut off on smaller windows.
• The one-time runtime setup and Hugging Face token moved into a settings gear at the top of the panel.

v5.6.1 — Quality of Life Update (fix)

A small follow-up to the Quality of Life update.

Fixes

• The Tags/Metadata box in Details & Actions (and in the Gallery's detail popup) no longer has its bottom edge and rounded corners clipped flat against the panel boundary — the border now draws cleanly all the way around.