Inference timing metrics are now logged after each generation: [Timing] TTFT: X.XX s | Y.Y tok/sec | N tokens | Total: Z.ZZ s. TTFT (Time to First Token) measures prompt-processing duration; tok/sec measures generation speed. Both values are written to the plugin log and enable direct backend comparison (MLX vs. GGUF) under identical conditions.
Substring alias matching added to the capability registry: model IDs containing qwen36-27b or qwen36-35b now resolve to the corresponding Qwen 3.6 capability entry, enabling derivative / fine-tuned model names to be recognised automatically.
Changed
Removed the undici bodyTimeout / headersTimeout limits (set to 0) so long prompt-processing phases — where the server is computing but has not yet emitted the first streaming token — no longer cause a UND_ERR_BODY_TIMEOUT abort. The OpenAI SDK client timeout is also set to 0 for the same reason.
[0.1.39] - 2026-05-22 Revision 39
Fixed
Progress feedback during the 2nd pass (SeedVR2 refinement) in mode: edit now correctly shows 2nd Pass Step x/y (n%) instead of bare Step x/y.
Audit: render_target.needs_upscaler now reflects the actual zoom-pass decision instead of being recomputed from raw dimensions versus backend limits.
[0.1.38] - 2026-05-20 Revision 38
Changed
Maximum render dimensions increased to 2048 × 2048 px (maxWidth, maxHeight).
Agent model loading now supports any OpenAI-compatible server. When the LM Studio-proprietary /api/v1/models endpoint is unavailable (e.g. Unsloth Studio, RunPod vLLM, remote inference APIs), the plugin falls back to the standard /v1/models endpoint and validates the model against the capability registry (capabilities.ts). Vision gating is enforced via the registry in both paths.
Fixed
Zoom-pass audit: inputs.canvas.original now correctly shows the Pass-1 backend_returned dimensions instead of the pre-Pass-1 source canvas.
Zoom pass (SeedVR2 refinement) now participates correctly in the overlay/Custom Config system. The overlay mode is resolved as zoom instead of falling through to , and the model identifier is passed as (not the raw filename) so no false-alarm error is logged. Custom Configs keyed are respected.
[0.1.37] - 2026-05-15 Revision 37
Changed
Preview size for all MediaTypes (attachments, images, variants, pictures) unified under a single constraint (w + h ≤ 1792 px). One preview file now serves chat display, Vision Promotion, analyse_image, and detect_object consistently.
Vision capability check now falls back to capabilities.ts when the server API does not return capability metadata (e.g. headless/remote servers with a plain /v1/models response). Models registered in capabilities.ts with supportsVision: true are now correctly accepted as agent models in such configurations.
[0.1.35] - 2026-05-11 Revision 35
Changed
Switched default edit mode parameters from Qwen-Image to Flux.
Improved agent instruction for Vision Promotion.
[0.1.34] - 2026-05-07 Revision 34
Added
Added SeedVR2 model family support.
Upscale pipeline extended with a 2-pass upscale process (Pass 1: standard generation; Pass 2: SeedVR2 refinement at target resolution).
Added upscale tool.
Added deployment scenario examples to the setup guide.
Changed
Updated vision instruction to improve accurate and detailed visual description.
i2iProfileUsed audit type extended to include "refine".
normalizeInputBuffer sum-constraint step: aspect ratio was distorted when both axes happened to land on different 64-multiples after independent rounding. Now uses candidate-based selection to best-preserve aspect ratio.
[0.1.33] - 2026-04-27 Revision 33
Added
Added 8-bit S (i8x) model filenames for Z Image Turbo 1.0, Z Image Base 1.0, Qwen Image 2512, Qwen Image Edit 2511, FLUX.2 [klein] 4B, and LTX-2.3 (distilled, dev).
Added support for Qwen3.6 27B (qwen/qwen3.6-27b).
Changed
Custom Configs now accept arbitrary postfixes (text2image.<any>, image2image.<any>, edit.<any>, , ). Previously only the built-in model IDs (, , , , , ) were accepted as postfixes.
[0.1.32] - 2026-04-21 Revision 32
Changed
Added support for Qwen3.6 35B АЗВ (qwen/qwen3.6-35b-a3b) as the new default LLM.
generate_image: When a reference image (canvas) is provided but mode is omitted, the call now fails with an actionable error asking the user to specify image2image, edit, or image2video — instead of silently falling back to text-to-image.
Improved support for external / 3rd-party image generation plugins.
Fixed
Images generated by 3rd-party-plugins in sequence now each get a unique index (i1, i2, i3 …) — the counter no longer resets to i1 on every turn.
[0.1.31] - 2026-04-08 Revision 31
Added
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.
Added support for draw_things_index, an LM Studio Plugin to search through your Draw Things generation history.
Changed
Changed Agent behavior to avoid generations or tool results being displayed more than just once.
Fixed
Small stability and error handling improvements.
[0.1.4] - 2026-01-18 Revision 4
Added
Live progress indicator (gRPC only): Image generation now shows real-time progress (e.g., "Step 12/20 (60%)") in the chat UI.
Explicit hard limit validation for width/height parameters with clear error messages.
Changed
Width/height descriptions now include maximum allowed values.
Dimension validation: Requests exceeding limits are now rejected with actionable error messages instead of silent clamping.
Fixed
HTTP img2img: imageFormat shorthand no longer overrides explicitly provided width/height values (parity with gRPC).
[0.1.3] - 2026-01-17 Revision 3
Changed
Adjusted number of attachments for Vision Promotion to match LM Studio client limit (5 images max).
Updated user documentation.
Fixed
Generation progress now shown consistently for all modes (text-to-image, image-to-image, and edit).
[0.1.2] - 2026-01-16 Revision 2
Changed
Unified preview sizes across all media types for consistent display quality.
[0.1.1] - 2026-01-16 Revision 1
Added
Initial release
Draw Things integration with vision-capable agents
Support for text-to-image, image-to-image, and edit modes
Vision Promotion system for attachments and generated images
Orchestrator-based agent architecture
Tool-based image generation interface
Moodboard images in mode: edit with explicit width/height parameters now preserve their native aspect ratio. Previously, explicit output dimensions caused moodboard inputs to adopt the output AR; only the canvas image should adopt the output AR.