Added additional guardrails for multiple tool-calls within an ongoing thinking process.
Updated the USER_GUIDE.md with best practices for providing instructions based on my new tool: playbook.
[0.1.30] - 2026-04-07 Revision 30
Added
Added Google Gemma 4 26B capabilities.
Added an option to embed generation metadata in generated images.
Added Post-Thinking Intercept: tool calls inside a reasoning/thinking block are now detected and executed correctly after the block closes.
Changed
Bumped flatbuffers dependency from 24.12.23 to 25.9.23.
Fixed
Fixed race condition, which could cause attachments not to be imported correctly.
Fixed Re-Generate recovery: when the user regenerates an AI response after attaching an image, the agent now correctly re-sees the attachment instead of silently omitting it.
[0.1.29] - 2026-04-01 Revision 29
Added
Added Qwen3.5 9B capabilities.
Changed
Adjustments to progress display during image generation.
Context-window warning threshold lowered from 32,768 to 16,384 tokens.
Match Swift Int() truncation semantics — use Math.floor instead of Math.round when converting float tensors to uint8 pixels.
nextVariantV counter is preserved when all image turns are deleted, so variant identifiers are never reused across generations.
[0.1.28] - 2026-03-22 Revision 28
Fixed
Custom Config video presets (image2video.custom, text2video.custom) now correctly resolve the active model.
[0.1.27] - 2026-03-21 Revision 27
Added
Added an option to auto-unload idleling agent-model on komplex text2video or video2video tool-calls.
Changed
Raised maximum variants from 3 to 4 (text2image, image2image, edit). Vision Promotion window expanded accordingly.
Video encoding (.mov assembly) is restricted to text2video and image2video modes.
[0.1.26] - 2026-03-17 Revision 26
Added
Added support for FLUX.2 [klein] 9B KV to the Flux model family (flux-2-klein).
Changed
Normalize default models to 6-bit versions.
Raised numFrames maximum from 257 to 641 (25,6 s).
[0.1.25] - 2026-03-15 Revision 25
Fixed
Fixed pre-build issues.
[0.1.24] - 2026-03-15 Revision 24
Changed
Improved plugin runtime stability.
[0.1.23] - 2026-03-15 Revision 23
Changed
Updated USER_GUIDE.md.
[0.1.22] - 2026-03-15 Revision 22
Added
Added support for LTX-2.3 (22B) to the LTX model family (ltx-2.3-distilled, ltx-2.3-dev).
Changed
Improved gRPC-connection logging.
[0.1.21] - 2026-03-14 Revision 21
Added
Added support for LTX-2 video generation.
Added LTX-2 q6p model variants (ltx_2_19b_distilled_q6p.ckpt, ltx_2_19b_dev_q6p.ckpt) as new defaults.
[0.1.20] - 2026-03-13 Revision 20
Fixed
Fixed imageFormat override not being resolved to explicit width/height before canvas normalization in image2image, edit, and image2video modes.
[0.1.19] - 2026-03-13 Revision 19
Fixed
Fixed normalizeInputBuffer distortion bug: input images were improperly scaled, causing stretched/distorted results when the requested format differed from the source (e.g. portrait canvas → landscape output).
[0.1.18] - 2026-03-13 Revision 17, 18
Changed
Changed pre-build process for LM Studio Hub.
[0.1.16] - 2026-03-11 Revision 16
Fixed
Fixed TCD Trailing sampler support.
Changed
Qwen3.5 now uses the correct thinking-mode flag.
Progress messages during generation are more informative.
[0.1.15] - 2026-02-26 Revision 15
Changed
Added support for Qwen3.5 35B (qwen/qwen3-vl-30b) as the new default LLM.
[0.1.14] - 2026-02-06 Revision 14
Changed
Slightly increased steps in edit mode for better prompt adherence with qwen-image.
image2image mode now supports moodboard (multiple reference images) via gRPC transport, same as edit mode.
current conversation tokens now refers to the agent model's context length
[0.1.11] - 2026-01-29 Revision 11
Added
New screenshots to reflect the recent update to LM Studio 0.4.0.
Changed
index_image tool calls no longer echo the model-mapping snapshot JSON back into the agent-model prompt history (token economy), while still passing the snapshot through to the index tool itself.
Fixed
README: Updated the model table to show flux_2_klein_9b_q6p.ckpt for all three modes (instead of just text2image with flux_1). Removed the "—" (not supported) markers for Flux.
[0.1.10] - 2026-01-26 Revision 10
Added
Tool summaries now show the model family label alongside the exact Draw Things model file that was used, so you can quickly understand what re-generates an image.
Added review_image() tool: lets you ask the agent to re-view any earlier created image, variant, or picture from tool-based generation or research.
Changed
Increased TTL for "vision-capability-priming" to 120 minutes.
Updated user documentation to reflect the recent changes.
Fixed
Fixed draw-things-index index_image markdown table: model_display is now preserved through picture materialization.
[0.1.9] - 2026-01-23 Revision 9
Added
Added FLUX.2 [klein] 9B (6-bit) as new default for flux model family.
Fixed
Fixed Agent behavior to avoid generations or tool results being displayed more than just once.
[0.1.8] - 2026-01-22 Revision 8
Changed
Improved source display in draw-things-index results: Chat IDs now shown completely, project names displayed without file extension.
Fixed
Fixed picture indexing across multiple queries: p IDs are now stable per chat (including draw-things-index project:// results) and no longer reset to p1 per query.
[0.1.7] - 2026-01-22 Revision 7
Fixed
Fixed a build issue that excluded necessary files from being uploaded.