TalkNova 2.1 · Release notes
Imagine more. Build more. Meet TalkNova 2.1.
Imagine for pictures and clips, seven Top Tier models, a redesigned TalkNova CLI, faster account sync, and improvements to chat, tools, token billing and everyday reliability.
November 15, 2026
Meet Imagine — pictures and clips in one place
A dedicated creative space inside TalkNova, with a simple Picture / Clip switch.
- Nano Banana
- Nano Banana Pro
- Qwen Image 2512
- Flux.1 Dev
- Flux.2 Pro
- Flux.2 Flex
- Flux.2 Dev
- Flux Pro
- Flux Kontext Pro
- Seedance 2.0 Fast
- Seedance 2.0 Ultra Fast
- Seedance 2.0 Fast - Image to Video
- Seedance 2.0 Ultra Fast - Image to Video
- Describe what you want to make, choose a model, and follow generation progress. Model-specific controls show supported sizes, aspect ratios and starting-image options.
- The clip lineup includes Seedance 2.0 Fast and Ultra Fast, plus their Image to Video variants. Availability and image-input support are shown in the live picker.
- Browse compact, newest-first cards with lazy thumbnails. Open the original-quality image or video, zoom and pan pictures, copy its prompt, download the original or confirm deletion from your history.
- Imagine uses the same searchable model picker as chat. Picture and Clip each show their supported choices, with a fresh availability check before generation and synchronized history across your signed-in devices.
- Your media history is private to your account. Saved image history is temporary, so download creations you want to keep.
Seven models, one new Top Tier category
The new Top Tier Models category sits first in the website picker. These are normal, non-batch routes.
- Claude Opus 5.5
- Claude Sonnet 5.5
- Claude Haiku 5.5
- GPT-6 Astra
- GPT-6.1 Sol
- Kimi K3
- Kimi K3.1 Pro
- Claude Opus 5.5 is the website default when available to your account. Your saved favorite stays yours, and unavailable defaults have a usable fallback.
- Use supported thinking levels, function tools, structured replies and web search. Search can return sources; charges apply when a search actually runs.
- The wider catalog keeps the DeepSeek, GLM, Qwen, Llama, Gemma, Muse, MiniMax and Nova families alongside the new models. Names, capabilities and availability come from the same account catalog.
- Options match the selected route: Opus and Sonnet use automatic tool selection, and Astra is currently offered for text rather than image input.
A redesigned TalkNova CLI
The TalkNova coding workspace is now easier to use from your terminal on macOS, Windows and Linux.
- Start with the talknova command. The refreshed home screen, model picker, composer, progress, tool activity and help make longer coding sessions easier to follow.
- Sign in with your TalkNova account in the browser. Your first message waits for sign-in and is sent once after a successful return—no manual API key or base URL setup.
- Use your account’s live models and shared Text allowance while keeping local project tools, approvals and workspaces under your control.
- Install from the Download page and receive updates through the CLI’s signed update verification. The nova command remains available as a compatibility alias.
Your account stays in sync
Move between the website, desktop and Android with less refreshing and fewer stale states.
- Account change notifications refresh saved chats and cloud agent activity through the existing durable sync system. Reconnects catch up with changes made while you were away.
- Model catalogs refresh on reconnect and when you return to the app. A last-good catalog survives temporary network failures, while old availability evidence expires.
- Availability dots now reflect recent completed requests, probes and cooldowns. A stale green result cannot keep masking a later failure, and models can recover automatically.
- Account-switch protections keep late media and sync responses from appearing in another signed-in account.
More reliable chat, coding and API workflows
Fixes reach both TalkNova chat and compatible tools using the public API.
- Improved streaming, function-call arguments and tool-result continuations help multi-step coding and agent requests complete in the correct order.
- The Responses API supports ordinary and streamed generation, function tools, continued tool outputs, structured responses and supported reasoning settings.
- Requests are checked against the selected model’s supported parameters, context and output limits. Unsupported settings get clearer errors instead of an avoidable upstream failure.
- Keyed model listings reflect the key’s access. Rate-limit responses provide retry guidance, HEAD requests finish correctly, and partial generations are not blindly replayed.
- Long reasoning and code replies receive model-specific output budgets, with clearer distinctions between a local limit, a provider failure and an interrupted connection.
Live Preview, with room to build
Build a website in chat, then interact with the real project in the preview beside it.
- Three.js modules, supported import maps, local GLTF/GLB models, textures and project assets now resolve more reliably. Your downloaded project keeps its original files.
- Drag the divider to resize the preview, switch between desktop, tablet and mobile sizes, or use fullscreen without losing the toolbar or restarting the scene.
- Safer follow-up edits preserve untouched files and checkpoints. Ambiguous snippets stay unapplied instead of replacing a working project.
- Clearer script, module, resource and WebGL diagnostics help explain a broken preview. Prepare a repair draft when needed, review it, and send it yourself.
Clearer usage and more accurate token billing
Spend one Text allowance across the text models you choose.
- Plan Text usage remains shared across models; image, video, speech and purchased API balances stay separate. Switching text models does not divide your allowance into separate model budgets.
- Non-streaming token accounting no longer uses the inflated prompt formula. Streaming and ordinary replies use verified provider receipts or the supported tokenizer path.
- Output reservations are not treated as generated tokens. Premium routes require valid token receipts, and unclear accounting is held for reconciliation rather than invented as an exact charge.
- Search fees are based on actual searches. The latest Top Tier search integration charges $0.007 per search plus the model’s token usage.
The creative workspace reaches the apps
The current desktop and Android releases include Imagine and the latest sync and model-health improvements.
- Use Picture and Clip workflows with generation progress, account-bound history, previews and downloads on supported app surfaces.
- Keep building with OpenCrew, project review, snapshots, Git, previews, MCP connections and permission controls in the desktop workspace.
- Android Remote Access connects to an approved paired computer. Nova One provides desktop assistant features with its own required key and device permissions.
- New model catalog entries arrive through your account without a fresh installer for every model addition. Existing favorites and some installed-app defaults remain unchanged.
Speech and files, ready for your workflow
Turn text into audio and keep working with documents and projects in chat.
- Kokoro and Fish S2.1 Pro Free remain the speech lineup, with supported voices, languages and audio controls shown for the selected model.
- Fish adds expressive delivery cues and selectable voices. Voice previews count toward your speech allowance, just like other speech generations.
- Ask AI to create and revise documents in chat, then download supported PDF, Word, Excel, PowerPoint, text and Markdown files.
- File and project workflows retain ownership checks, revision checks, expiry and safe download controls.
Less friction, stronger boundaries
Reliability and privacy improvements run underneath the features you use.
- Fewer duplicate catalog requests and more efficient account notifications reduce background work. Small text chats no longer use large-upload slots.
- Replies, tools and private reasoning are handled separately. Provider keys stay server-side, and model-generated identity claims are not treated as proof of the underlying model.
- Uploads, memory, skills and tool results have clearer untrusted-data boundaries, with authorization and sandbox protections enforced by the application.
- Updates retain readiness checks, active-request draining, encrypted backups and rollback. Existing referral, crypto checkout and Custom API features remain available from their usual places.
Ready to try it?
Open Imagine from the chat sidebar, explore the live model picker, or get TalkNova CLI from Downloads. Refresh the website or reopen your app to load the latest catalog. Features and options vary by model, platform and current availability.