What changed since the August 25 catalog check: new model routes, Brain entities, result workflows, Designer decks, image tools, and reliability fixes.
Where: Open the feature-specific guides linked below, then use the corresponding Flowgen card or editor panel.
Key ideas
- Model additions and retirement: GLM 5.3 Flash joins language-model selection. Wan 3.0/Prime and Omni 1.1 add video routes; Grok Imagine 2 variants have explicit 1K/2K tiers. Gemini 3.5/3.6/3.7 Flash are retired from selection and updated clients resolve saved selections to Gemini 3.8 Flash without changing the base round tier.
- Reusable inputs and visible results: Brain adds Product, Character, and Place plus a fuller Brand preset. Mini-app questions can select records. Generation history is a contact sheet, and result previews can create detached typed input nodes.
- Manual work remains accessible: The free tier allows up to three owned projects and manual editing of accessible projects. New accounts receive 50 welcome Sparks. AI, cloud rendering, and hosting still check the relevant balance and limits at the point of spend.
Steps
- For an older canvas, reopen it in the updated client and confirm a saved Gemini 3.5/3.6/3.7 Flash selection resolves to 3.8 Flash.
- Review the model-specific guide before changing a generator: references, output sizes, and prices differ by route.
- Use Brain records and record questions when a reusable flow needs selectable brand, character, product, or place inputs.
- Compare generation takes, extract independent winners, and lock cards that should not be accidentally deleted.
- Read the relevant editor guide for new crop, smart-object, channel, motion-preset, selector, Lottie, and carousel controls.
- After an update, inspect saved results, exports, and the current cost quote before running a large batch.
Flowgen fixes
- List iteration resolves each row through local editing operations and writes per-row artifacts instead of overwriting one shared output.
- Independent groups can run concurrently. Array outputs and multi-input sockets have distinct visual cues; hovering reveals names.
- Transform offers text, named fields, and list merging without silently replacing custom steps.
- Extract Frame gains a player and frame stepping. Video Matte preserves source FPS; Video Depth nodes can bake their own clips.
- All pages from editor-backed nodes are available downstream, including video scenes; video annotation edits and references survive the handoff.
Designer, editor, and rendering
- Designer presents complete slide builds and uses finished frames for thumbnails and static exports. Plain Share and navigable Presentation are explicitly different.
- Image Editor gains ratio/straighten crop, embedded smart-object operations, layer filtering, component channels, and a histogram; 4K brush/mask and live-brush fixes are included.
- Timeline keyframes use the system clipboard and a shared preset library. Lottie adds fit/alignment; carousels add active/leave actions, viewport start, and hover pause.
- Website layout defaults no longer leak into video/ad documents. HTML ZIP exports and editor page outputs preserve styles.
- Video rendering handles composition completion, precomp identities, image-sequence readiness, and first-frame sets more reliably.
Loading, collaboration, and AI
- Assets and dashboard cards load previews on demand; searches and deep reveals preserve their scope and dragging behavior.
- Collaboration and saving fixes preserve peer updates and stable grouping while reducing idle rendering and repeated reads.
- Targeted website/video repairs retain source reads and avoid unnecessary scene recreation. PDFs and Markdown briefs reach the model through the supported attachment preparation.
- OpenRouter streaming and higher output ceilings support longer responses; actual output is billed, and measured reasoning is not multiplied twice.
Limitations and important notes
- This is a catalog/source release summary, not a claim that every application service was deployed together. Server-side compatibility changes require that service to run the updated code.
- Historical run records keep their original model labels. Updating a runtime selection does not rewrite past results.
Related guides
- GLM 5.3 Flash: multimodal chat, reasoning, and pricing
- Wan 3.0 and Prime: video modes, references, and controls
- Gemini Omni 1.1 Flash: generate, reference, edit, and extend video
- Grok Imagine 2: Standard, Quality, resolution, and image edits
- Brain entity presets: Brand, Product, Character, and Place
- Let a mini-app question choose a Brain record
- Compare generation takes in the version grid
- Extract a result to its own input node
- Lock nodes and groups against deletion or cutting
- Image Editor: crop ratios, straighten, and canvas transforms
- Image Editor: smart objects, embedded contents, and layer filtering
- Image Editor: component channels and the live histogram
- Reuse motion with keyframe copy/paste and the shared preset library
- Build shared class and contextual selectors by clicking
- Designer decks: full slide builds, transitions, and finished-slide exports
- 3D cinematic authoring: lenses, motion presets, camera rails, and reference backdrops
- Animate an image sequence by frame number