Local Image Generator

Generate images locally inside Obsidian — no cloud, no external app, no API key. Choose between two models: SD-Turbo (fast, runs in-process on your own GPU via WebGPU/onnxruntime-web) or FLUX.2 klein 4B (higher quality, multiple aspect ratios, via a local mflux child process — see Requirements).

What it does

• Open the generator from the ribbon icon or the Open generator command.
• Pick a model from the dropdown. SD-Turbo is fixed at 512×512; FLUX.2 shows a size dropdown with 7 curated aspect ratios (square, portrait, landscape).
• Type a prompt, adjust steps/seed if you like, and press Generate. Click a style chip (Sumi-e, Watercolor, Photo, Oil — edit or add your own in settings) to append its look to the prompt; click again to remove it.
• Generate uses the seed from the field (it never rerolls) and greys itself out once the current prompt/model/seed/steps/size exactly match your last result — regenerating without changing anything would just reproduce the same image. Reroll rolls a fresh seed and generates a new variation regardless; use the dice icon to reroll the seed by hand without generating.
• Switch to the History tab to see your past generations as full recipes (prompt · seed · steps · size · time) — group them by prompt, click one to load its recipe back into Generate, delete single entries, or clear all.
• Create saves the image as a new attachment. By default that's all it does (it also opens the image) — set the Create button dropdown in settings to Image + note to have it also create a note with the generation's prompt, seed, steps, size and date in its frontmatter and the image embedded, and open that note instead. Insert always just saves the image and embeds it at your cursor in the current note.

There is no img2img, no LoRA/ControlNet, and no negative prompt/CFG scale — both models are guidance-distilled and don't support them meaningfully.

The interface is available in English and German, switching automatically to match Obsidian's own language setting — no separate language option to set.

Requirements

SD-Turbo (always available):

• Obsidian desktop only (isDesktopOnly: true — this plugin does not run on Obsidian Mobile).
• WebGPU with shader-f16 support. This generally works on macOS and Windows with a reasonably modern GPU; on Linux it depends on your graphics driver. If your system doesn't qualify, the plugin's status area tells you so instead of generating.
• ~2.5 GB of disk space for the model download (see below).
• Roughly 4–7 GB of free memory while an image is generating. If generation fails, try closing other apps first.

FLUX.2 klein 4B (optional — only if you want the second model):

• macOS with mflux installed yourself (uv tool install mflux) — the plugin detects and runs it, but never installs or downloads it for you.
• ~8 GB of disk space for the FLUX weights (separate download, see Settings).

Install & first steps

1. Install and enable the plugin.
2. Open Settings → Local Image Generator, section Model, and click Download under SD-Turbo (~2.5 GB). This is a one-time download; nothing happens automatically. (Optional: install mflux and download the FLUX.2 weights in the same section if you want the second model.)
3. Once the download finishes, open the generator (ribbon icon or the Open generator command), pick a model, enter a prompt, and press Generate.
4. Use Create to save the image as a new attachment and open it, or Insert to save it and embed it at your cursor. Set the Create button dropdown to Image + note in settings first if you also want a note with the generation's details.

You can re-download or delete a model's files at any time from the same settings page.

Settings

Settings → Local Image Generator is grouped into collapsible sections:

• Model — one section covering both models: download the SD-Turbo model files, and (below it) find/point at your mflux install, choose where FLUX weights are stored, and download them.
• Output — the image folder (leave empty to use Obsidian's attachment folder, with autocomplete for existing folders), the note folder used when Create makes a note (leave empty to put the note next to the image), the Create button dropdown (Image only or Image + note), and the starting value of the steps slider.
• Styles — the same style presets shown as chips under the prompt field. Edit a preset's label or prompt text, delete it, or add a new one.
• Danger zone — delete the downloaded SD-Turbo model files.

How network and storage are used

This section is here so you know exactly what the plugin does over the network and what it writes to disk — nothing beyond what's listed here happens.

Network — only on explicit user action ("Download" in settings). The plugin fetches five files, once, from Hugging Face:

Total download: ~2.5 GB. This download only happens when you click Download in settings — the plugin makes no network requests at startup, during editing, or at any other time. Files that are already present are skipped, and each file's integrity is checked against its expected size. After the download, image generation itself is fully offline: your prompts never leave your machine, and no image data is ever sent anywhere.

Storage. Downloaded model files are written to the browser's Cache API inside Obsidian's Electron profile — not into your vault, and not synced by Obsidian Sync or any vault sync tool. You can delete the cached model files at any time via Settings → Local Image Generator → Delete model. Generated images, by contrast, are saved as normal attachments inside your vault, exactly like any image you'd add yourself.

The WebAssembly runtime (onnxruntime-web) is bundled inline with the plugin; no plugin code is loaded from the network at runtime.

FLUX.2 klein 4B (optional second model)

• Runs via mflux, a local CLI tool you install yourself (uv tool install mflux). The plugin never downloads or executes code on its own — it only detects and runs the tool you installed, as a local child process.
• Model weights (8 GB, black-forest-labs/FLUX.2-klein-4B) are downloaded from Hugging Face only after you explicitly start the download in the settings. They are stored in the Hugging Face cache (`/.cache/huggingface` or a folder you choose), outside your vault, and are shared with other Hugging Face tools.
• No telemetry, no other network use.

Privacy

• No telemetry. The plugin does not collect, transmit, or phone home any usage data, prompts, or images.
• No network access other than the one-time, user-initiated model downloads described above. Once a model is downloaded, generation runs entirely on your device, offline.

Model & licenses

• Plugin code: AGPL-3.0-or-later (see LICENSE).
• SD-Turbo weights: stabilityai/sd-turbo — subject to Stability AI's model license. Please review the terms on the model card before using generated images, especially for commercial purposes.
• SD-Turbo ONNX export used by this plugin: schmuell/sd-turbo-ort-web, the reference export used by Microsoft's WebGPU demo for onnxruntime-web.
• SD-Turbo inference engine: onnxruntime-web.
• FLUX.2 klein 4B weights: black-forest-labs/FLUX.2-klein-4B — Apache 2.0. Please review the model card before using generated images, especially for commercial purposes.
• FLUX.2 inference tool: mflux, installed and run entirely by/on your machine — not distributed with this plugin.

This plugin only distributes code; all model weights are downloaded directly from Hugging Face by the user, on demand.

Roadmap

Two models are supported today (SD-Turbo and FLUX.2 klein 4B, both text-to-image only). Planned next: image editing with reference images from your vault for FLUX.2 (mflux-generate-flux2-edit already supports this upstream), and a small provider API so other community plugins can request locally-generated images without duplicating this infrastructure. Additional backends (e.g. ComfyUI/DrawThings) remain out of scope — this plugin only runs models locally, either in-process or via a local CLI tool you install yourself.

License

AGPL-3.0-or-later — see LICENSE. Model weights have their own license terms (see Model & licenses above).