Draw Things for LM Studio

Generate local images and videos from LM Studio chat by calling draw-things-cli.

Tools

• generate_image creates a PNG and returns renderable local Markdown. It can also transform an uploaded or previously generated image when called with an image path.
• generate_video creates an MP4 or MOV. By default it starts a background job to avoid LM Studio tool-call timeouts, then opens the completed video automatically in the desktop default app. It can also animate an uploaded or previously generated image when called with an image path.
• check_video_generation checks a background video job and returns the poster preview plus absolute output path when complete.
• open_generated_media opens a generated image or video in the desktop default app when LM Studio does not make local links clickable.

When Analyze Generated Media is enabled, generate_image also analyzes the generated image with the currently loaded vision model. For videos, the plugin analyzes the generated poster frame. Tool results include analysis and, when analysis completes, analysisText so the chat can use the generated media's visible contents as context.

Configuration

At minimum, choose a Draw Things model. The default dropdown is seeded with the models currently installed on this machine:

• flux_2_klein_4b_q6p.ckpt
• ltx_2.3_22b_distilled_1.1_q6p.ckpt

Use Custom Model Override for other Draw Things model ids, display names, hf:// references, owner/repo references, or Hugging Face URLs.

Common Draw Things settings are available as plugin defaults and as per-tool overrides: width, height, steps, CFG, seed, strength, frames, video format, negative prompt, video execution mode, timeout, and advanced configJson / configFile overrides. Use foreground video mode only for shorter generations; background mode avoids LM Studio tool-call timeouts.

When a user uploads images, the plugin stages up to five of them under the configured output subdirectory in inputs/ and appends a short Draw Things image inputs block to the chat message. The assistant should pass one of those exact paths as image for image-to-image or image-to-video. The plugin also lists up to five recent generated images from the current chat/plugin session so they can be reused explicitly.

Generated media analysis is enabled by default. It can be disabled in plugin configuration with Analyze Generated Media or per call with analyzeGeneratedMedia: false. The analysis prompt and max token budget are also configurable. If the loaded model does not support vision, generation still succeeds and the result reports analysis.status = "unavailable".

Generation always runs in local-only mode with --offline, --no-download-missing, and --disable-preview, so chat tool calls cannot download missing models.

Usage

Ask the assistant to generate an image or video. The tool result tells the assistant exactly which Markdown/path fields to include in the chat response.

Examples:

• "Generate an image of a red cube on a table."
• "Using the uploaded image, make it look like a watercolor sketch."
• "Animate the image I just generated with slow drifting clouds."
• "Generate a short video of ocean waves at sunset with 49 frames."

Notes

• Outputs are written inside LM Studio's working directory under the configured output subdirectory, defaulting to draw-things/.
• Input images and advanced config files must live inside LM Studio's working directory. Uploaded chat images are copied into the configured output subdirectory automatically before tool use.