Collages and stronger voice cloning.
Added
- Image collage combines selected image nodes into one new PNG.
- TokenRouter Seedance can use asset-based references for images, audio, and video.
- Provider-scoped asset IDs let one image remember separate assets per provider.
- ElevenLabs and MiniMax can clone voices from upstream audio.
- ElevenLabs controls add stability, similarity, style, and speed options.
- Generation parameters support range popovers and number inputs.
Changed
- GPT Image 2 uses clearer aspect-ratio and image-size handling.
- Reference thumbnails can show asset IDs from multiple providers.
Fixed
- GPT Image 2 handles more image proportions reliably.
Faster large canvases with lazy media.
Added
- Lazy media loads offscreen images and videos only when they come into view.
- Thumbnail cache creates lightweight previews for large images.
- Performance settings let you rebuild or clear thumbnail caches.
Changed
- Canvas refreshes run after relevant edits instead of constant background scans.
- Reference thumbnails and audio durations reuse lighter cached work.
Fixed
- Large media canvases open with less CPU and memory pressure.
Gemini multimodal refs for text.
Added
- APIMart can route GPT-5.5 text generation.
- Gemini 3.5 Flash is available through Google Gemini and TokenRouter.
- Gemini text generation can use upstream video, audio, and PDF references.
- Agents can inspect upstream audio and PDF references.
Fixed
- Gemini text errors show clearer details.
Clearer APIMart task errors.
Fixed
- APIMart task failures show readable error details.
APIMart support for Nano Banana.
Added
- Nano Banana Pro and Nano Banana 2 can use APIMart as an image provider.
Unified Qwen voices and voice design.
Added
- Qwen Voice brings built-in voices, voice references, and designed voices into one model.
- Voice design lets an upstream text node describe the voice while the current node stays the spoken script.
- CosyVoice can create designed voices from prompt text.
Changed
- Voice sources are clearer: Built-in, Voice ref, and Design.
- Qwen Voice replaces the separate Qwen audio entries.
- Built-in Qwen voices still use the searchable voice picker.
Fixed
- Agent audio generation keeps voice reference and design choices.
- Designed voices keep canvas files smaller.
Voice picker and cloning for TTS.
Added
- TTS voice picker lets you search, filter, preview, and choose voices.
- Voice cloning can reuse a cloned voice from the same reference audio.
- DashScope audio models add Qwen and CosyVoice generation options.
Changed
- Audio model names are shorter in the generation bar.
- Temporary uploads use the new Bragi temp service.
Fixed
- Voice choices work more reliably across fal-backed audio models.
- General bug fixes.
Gemini can read video references.
Added
- Gemini text generation can use upstream videos as references.
Changed
- Video references are uploaded through Bragi Relay before Gemini reads them.
Fixed
- Unsupported text providers now explain when video references are unavailable.
Safer imports and refreshed OpenAI models.
Changed
- MCP server starts disabled for new installs.
- Bragi package import uses a standard file picker.
- OpenAI text models now include GPT-5.5 and GPT-5.5 Pro.
Fixed
- OpenAI Pro text generation routes through the right API.
- Agent edge deletion reports a clear error if an edge remains.
- The language prompt guides users to switch Obsidian language manually.
- General bug fixes.
License update and toolbar polish.
Changed
- License changed to Mozilla Public License 2.0.
- Edge, group, and text-node toolbars have a cleaner Bragi layout.
Fixed
- Remove label now lives in the More menu for labeled edges.
- Group background controls use the Bragi toolbar icon style.
- General bug fixes.
Safer Bragi packages and downloads.
Changed
- Bragi packages now use a single-file format with embedded assets.
- Imported package assets are written only to your vault assets folder.
- Export and media download flows behave more like normal browser downloads.
- Release downloads are easier to verify.
Settings import and broader TokenRouter support.
Added
- Import settings lets you bring in a Bragi Canvas settings file after validation and confirmation.
- Seedance supports reference-video generation through BytePlus.
- TokenRouter now covers image, video, and text generation from one provider.
- HappyHorse video models are available through TokenRouter.
Fixed
- HappyHorse image-to-video works with reference images.
- General bug fixes.
Restored canvas tools and cleaner docs.
Changed
- Canvas toolbar tools are back for text, media, edges, and groups.
- The More menu now holds Focus, Set color, and Delete across more selections.
- Install guidance better matches the current community release.
Fixed
- Split grid, 360° viewer, Mark, media downloads, and connected duplication are available again.
- General bug fixes.
Release packaging for manual installs.
Fixed
- Release downloads include the expected plugin files.
Cleaner wording across settings and notices.
Fixed
- Settings and notices use clearer wording.
Cleaner plugin package for review.
Grok models for every generation mode.
Added
- xAI is available as a provider for text, images, video, and speech.
- Grok text models add Grok 4.3 and Grok 4 Fast.
- Grok Imagine supports image generation, image editing, and a Quality toggle.
- Grok TTS adds five multilingual voices for speech generation.
Changed
- Grok Video supports text prompts, first frames, image references, video extension, and 1080p output.
- Video duration choices adjust to the selected generation mode.
- Grok Imagine tiers are grouped into one model card.
- Use a video node as the source for video extension.
Fixed
- Reference-image generation handles one or multiple images more reliably.
- Model and mode dropdowns keep their selected text visible.
- Luma handles larger reference images more reliably.
More reliable agent text edits.
Fixed
- Agent text edits keep the rest of your canvas in place.
Luma images and better generation defaults.
Added
- Luma Uni-1 is available for text prompts and reference-image generation.
Changed
- Generated files open in their own tab while your canvas stays visible.
Fixed
- GPT Image 2 respects landscape, portrait, and square size choices.
- Luma handles larger reference images more reliably.
- A prompt with one reference image opens the right image generation mode.
Friendlier Test button errors.
Fixed
- The Test button now reports clearer reasons when a key doesn't work.
Test every provider key in one click.
Added
- Added a Test button for provider connections.
English-only for a smoother experience.
Added
- Launches in English for consistent behaviour across Obsidian builds.
New nodes always land on empty space.
Changed
- Generated nodes now auto-place into the nearest free spot.
Clearer generation status, recoverable placeholders.
Added
- Live "Generating…" overlay with a running timer on every placeholder.
- Interrupted generations are flagged when you reopen the canvas.
Changed
- Placeholder visuals rebuilt for a calmer look.
Smarter auto-placement on dense canvases.
Changed
- Generated nodes land closer to the source node.
- Placeholders match the target aspect ratio from the start.
Major upgrade for AI agents.
Added
- Eight new agent tools for tracking jobs, switching canvases, and uploading files.
- Optional access token to secure the local agent server.
- A ready-to-use skill package for any compatible agent.
Split grids, face refs, built-in uploads.
Added
- Split Grid button turns tiled images into individual nodes.
- Face references for Seedance videos via BytePlus.
- Show/hide toggles on every API key field.
- Pasted canvas images go into a dedicated folder.
- Built-in file uploads — no cloud setup needed.
Changed
- Cloud storage settings simplified.
- Provider descriptions trimmed to one sentence each.
Polish across settings and pasted images.
Added
- Show/hide toggles on every API key field.
- Pasted images land in a Bragi folder, keeping your vault tidy.
Zero-config uploads, smoother settings.
Added
- Built-in uploads — no keys required.
- Rewritten Add Model and Add Provider flows.
Changed
- Friendlier messages throughout the plugin.
Rebuilt Providers and Models pages.
Added
- Add providers and models one at a time — only what you use.
Friendlier words everywhere.
Changed
- Notices and dialogs rewritten in plain English.
One home for every generated file.
Changed
- Images, videos, and audio all live in one tidy Bragi folder — move or rename canvases freely.
Added
- One-click migration for upgraders, with an automatic backup.
TokenRouter and 360° panoramas.
Added
- TokenRouter — one key for images and text across many models.
- 360° panorama viewer with mirror and capture.
Lighter nodes, new card menu.
Added
- Quick Export and Import for
.bragi packages in the card menu.
Changed
- Refreshed node look — thinner borders, softer shadows.
Test your storage before using it.
Added
- A Test button for cloud storage settings.
Videos keep generating across restarts.
Added
- Pending video generations resume automatically after an Obsidian restart.
Bring your own Cloudflare R2 storage.
Changed
- Reference uploads use your own cloud bucket.
Multiple agents can connect at once.
Fixed
- Parallel agent sessions no longer disconnect each other.
Finer Seedance duration control.
Changed
- Pick any duration from 4 s to 15 s, or leave it on Auto.
Cleaner image size picker.
Changed
- GPT Image size dropdown shows plain aspect ratios.
BytePlus support for Seedance.
Added
- Seedance 2.0 and 2.0 Fast now support BytePlus.
Audio references and MiniMax.
Added
- Seedance videos accept an audio reference.
- MiniMax direct TTS and music, with Chinese voices.
Agents can trigger generations.
Added
- Agent tools to list models, inspect upstream, and start a generation.
Upstream text feeds your prompts.
Added
- Drag to reorder upstream text references — the order becomes the prompt order.
Claude, GPT-5.4 Pro, and AWS Bedrock.
Added
- New models: Claude Opus 4.7, Sonnet 4.6, Opus 4.6, and GPT-5.4 Pro.
- New providers: Anthropic direct and AWS Bedrock.
Text generations can split into many nodes.
Added
- Ask for N sections and get N connected text nodes.
Share a canvas as one file.
Added
- Export and import
.bragi packages — graph plus every asset in one bundle.