Skip to content

DEAS 0.2.8 — Stop interrupts again, H3 inpaint fix, LLM API key + thinking models, spellcheck everywhere

Choose a tag to compare

View all tags
@darksidewalker darksidewalker released this 31 Oct 22:48
v0.2.8
144d149

DEAS 0.2.8 BETA

Fixes

  • Stop interrupts the running generation again (long videos, many-step images). The scoped interrupt adapter was never installed on master backends because the launch check pinned whole-file hashes of one old ComfyUI commit; it now verifies the queue/interrupt protocol statements. Takes effect after the next backend start.
  • MiniMax H3 image edit / inpaint runs again, with and without a mask (metadata config referenced the removed text encoders → KeyError '2').
  • LLM backend: Ollama requests send think:false, so thinking models (Qwen3.5, gpt-oss) answer instead of LLM response has no valid text.

New

  • API key for OpenAI-compatible LLM servers (Backend → LLM backend), sent as Bearer and forwarded to the DaSiWa node pack as openai_api_key for Prompt Forge.
  • In-app spellcheck on every platform (spellbook, Hunspell-compatible) with Ctrl+. / right-click suggestions; dictionary install works on Windows too.
  • Presets: New preset… overlay; Post-Processing grouped by stage with the action buttons inside their groups; Quick Settings toggles show their state; click a video preview to play/pause.

Full list: docs/CHANGELOG.md.

Build provenance
Built locally and unsigned from source commit 144d1494b537fe6cc7049eccb34bb3cb115bf700 with build-linux.sh --release (rustc 1.97.0, Node 26.12.1; GitHub runners gated). deas.exe is a cargo-xwin cross-build without NSIS installer and is not Windows runtime-accepted; Linux binary glibc floor 2.39, ldd clean.

cdd5f1db550e498df646c1eea8cb9f8f44007e01ca0af8e293a6db9a0cc9ad17  BUILD-MANIFEST.json
d3b79bf15b25ed6fdd3bb4cc584934f5483767727ce8e388e8ae0590cfc95fd1  deas.exe
19c3ab5cabcde32cc0027506aa15951827319db34bd1226e1f78560ab73c817d  deas-linux-amd64
Assets 6
Loading