Changelog

HoldOrange platform version history — discover new features and improvements in each release.

Last updated: 2026-11-13

Changelog


Updates in the current codebase (deployment version unconfirmed)

These notes reflect the repository as reviewed on November 13, 2026. Availability depends on your deployed environment.

  • 2D production planning: Select image/video models and aspect ratio, preview and apply a production canvas, then run media generation. Includes reference dependency checks and reuse of successful outputs.
  • Canvas arrangement: Five automatic layout modes and alignment guides, with layout scope based on the selection.
  • HOTV community: Showcase, creators, campaigns, and personal publication management.
  • Creative Studio: Screenplay and marketing script workspaces; see the dedicated guide.

This section does not assign a formal release version or release date to these changes. Previous release records are retained below.


V4.2.0 (May 28, 2026)

Canvas & collaboration

  • 2D canvas collaboration stability — Fixed concurrent-editing issues that previously caused node deselection, prompt fields being overwritten by remote updates, and conflicts when one user edits a node while another moves it. Locally-edited fields are now protected from remote reconciliation.
Note: Field-level realtime collaboration (shared cursors, per-user undo/redo) will arrive in a later release.

V4.1.0 (May 22, 2026)

Canvas

  • Canvas history panel is back — The 2D canvas toolbar now exposes a History panel with two tabs:
  • Generation history — Lists every completed task linked to the current canvas (image / video / audio / music) with creator avatar and name. Each entry supports preview, download, copy URL, insert-as-node-input, and regenerate-with-same-parameters.
  • Version history — Automatic snapshots are taken on every content change (up to 10 versions per canvas). Each snapshot shows node count, edge count and editor, can be previewed full-canvas in read-only mode, and restored in one click (a backup is taken automatically before restore).
  • Available for both personal and organization canvases, gated by project.canvas.view.* / project.canvas.edit.* permissions.

Real-human creation

  • Real-human verification — Identity verification flow for real-human character assets is now available behind a system-administrator switch; the real_human_enabled plan attribute decides per-plan availability. Status and upgrade prompts are surfaced consistently across the character editor, subscription dialog and personal / organization dashboards.

Organization & permissions

  • Organization permission scope refactor — Project / team lifecycle permissions (create, delete, view-all) are unified at the organization scope, while within-project / within-team operations remain at their own scope. Permission semantics are now clearer and remove a class of false-permission and over-grant cases.
  • org_admin now implicitly has full team / project administration. org_member now has CRUD over projects they own by default.
  • A one-time data migration runs on upgrade to clean up misplaced permissions in legacy custom roles. Please notify organization administrators before deploying so they can review role assignments.
  • Hide built-in system roles — Organization administrators can toggle “Hide system roles” under Organization Roles. When enabled, member-invite and role-assignment dialogs only list custom roles; already-assigned system roles are unaffected.

V4.0.2 (May 1, 2026)

New video models

  • LTX-2.3 Pro — Lightricks flagship via fal.ai. Supports 1080p / 1440p (2K) / 2160p (4K), with text-to-video and image-to-video.
  • LTX-2.3 Fast — Speed-optimized economical variant covering 1080p/2K/4K, ideal for rapid iteration and bulk generation.

Task pipeline fixes

  • Resolved task_supplier_attempts rows stuck at pending (added tasks.current_attempt_id so attempt records terminate in lockstep with the task).
  • Fixed FrozenCreditsJob uniqueKey collision (no more 42P07 errors under multi-pod deployments).

V4.0.1 (April 29, 2026)

Billing & Credit System

  • Monthly subscription credits — Monthly/yearly plans now issue credits per service month and clear unused subscription credits at month end. SubscriptionLifecycleJob handles monthly issuance and 72-hour pre-clear email reminders.
  • Resource packs aligned with billing cycle — Credit packs / storage packs / seat packs now expire and renew with the subscription cycle. Credit packs are consumed FIFO by earliest expiry first.
  • Permanent gift credits preserved — Registration credits, trial credits, referral rewards, activity gifts, and SLA compensation are not affected by the monthly clear rule.
  • Payment closure hardening — Standalone Pay backend with dynamic callback whitelist (AllowedCallbackHosts) and HMAC signature verification; reconciliation and Revenue accounting unified.

Video & Audio Capability Upgrades

  • Kling V3 Omni — Integrated kling-v3-omni video model with multimodal text / image / reference inputs.
  • Hailuo Video — Hailuo I2V / T2V / first-and-last-frame generation up to 1080p.
  • GPT-Image-2 — OpenAI gpt-image-2 text-to-image and image-to-image with reference image guidance.
  • Audio & music nodes — Dedicated canvas nodes for audio and music generation, decoupled from video nodes.

Queue & Realtime

  • Queue throughput governance — Three-stage task pipeline (Submit → Resume → Postprocess); task_supplier_attempts tracks every supplier call; permanent-error classifier prevents wasted retries.
  • WebSocket single-channel — Task progress is broadcast via a unified WebSocket task_update event; the frontend no longer polls, reducing backend request load.
  • Multi-angle generation reliability — Improved timeout and retry policy for fal.ai multi-angle image generation.

V4.0.0 (April 6, 2026)

Architecture Refactor: SaaS / DC Separation

  • Top-level directory split — Repository reorganized into saas/, datacenter/, and shared/ so the SaaS and on-prem (DC) editions can release and bill independently.
  • Shared productcore — 87% of common code consolidated into shared/; canvas / task / account / credit core logic is reused by both editions.
  • Standalone Pay backend — Payment capability extracted as a dedicated service shared by SaaS and DC, with HMAC-signed callbacks and a dynamic host whitelist.
  • DC dual-key suppliers — The DC edition supports a dual-key model (platform key + customer-owned key); customer-key tasks are billed at zero credits via CustomerKeyGate + CreditService.EstimateCostForOwner short-circuit.
  • License quota interface — DC quota limits are injected as LicenseQuotaProvider and validated through the unified plan_gate_service.

V3.2.0 (April 1, 2026)

Aliyun Bailian Video Providers

  • Wan 2.7 — New aliyun-wan provider supporting text-to-video and first-frame generation.
  • PixVerse V6 — New aliyun-pixverse provider supporting T2V / I2V / keyframe (KF2V) modes.
  • Vidu Q3 Pro — New aliyun-vidu provider supporting text / image / first-and-last-frame / reference video generation.
  • Kling V3 backup channel — Kling V3 routed via Bailian as a fallback for the direct Kling API.
  • Unified async task contract — DashScope video tasks use X-DashScope-Async: enable with a 30 s initial wait, 5 s polling interval, and 30-minute maximum.

V3.1.0 (March 24, 2026)

Resource Packs & Grant System

  • Resource pack store — 14 standard resource packs across three types: credit pack, storage pack, and seat pack.
  • Batch-tracked credits — New credit_grants table records source, expiry, and consumption per batch; consumption is FIFO by earliest expiry.
  • Batch expiry job — ExpireCreditGrantsJob runs every 30 min to expire stale batches and write expiry transactions.
  • Storage / seat accumulation — Storage and seat packs accumulate into subscription.extra_storage_gb / extra_seats.
  • Order-type expansion — New order types credit_pack / storage_pack / seat_pack; billing details distinguish subscription vs. resource-pack purchases.

V2.5.0 (March 29 2026)

Organization Management

  • Organization Creation & Management — Users can create organizations, invite members, and enable team collaboration
  • Member Management — Email and link invitations, pending invitation management, member role assignment (Owner, Admin, Member)
  • Team Management — Create teams within organizations, assign team members, customize team role permissions
  • Organization Credits — Organization-level credit balance, per-member spending limits, transaction history
  • Organization Subscription — Independent subscription plans for organizations, quota usage monitoring (2D/3D canvases, storage, seats, teams), owner can upgrade/renew
  • Organization Settings — Edit basic info (avatar, name, description), danger zone (transfer ownership, delete organization)

Billing & Payments

  • Order Management — Billing page displays full order list with detail view, receipts, invoice requests, repay, and delete canceled orders
  • Multiple Payment Channels — WeChat Pay, Alipay, Stripe, and Bank Transfer support with payment method selection dialog
  • Bank Transfer Flow — Bank info display with one-click copy, order number reminder for transfer notes, "Transfer Submitted" confirmation
  • Invoice Management — Request invoices, manage invoice titles (personal/enterprise), support for regular and VAT special invoices, invoice status tracking
  • Subscription Status Alerts — Prominent warning banners for past due, expired, and near-expiry subscription states

Email Notification System

  • Order Confirmation Email — Automatic notification after successful payment
  • Subscription Activation Email — Confirmation when subscription is activated
  • Security Alert Email — Login alert for new devices with device info and IP address

UI Improvements

  • Tab Style Unification — Consistent tab component variant across the platform
  • Sidebar Badges — Admin panel menu items display count badges (members, teams, pending invitations, etc.)
  • Dialog Component Upgrade — Destructive action confirmations upgraded from Dialog to AlertDialog to prevent accidental operations

V2.3.0 (March 20 2026)

AI Creative Agent Upgrade

  • Unified Agent Mode — removed the three-mode switcher (Chat / Agent / Quick) and consolidated into a single AI Creative Assistant that automatically detects canvas operation intent in chat and routes to Agent execution
  • 7 Atomic Canvas Operation Tools — Agent can directly read and write canvas state via get_canvas_state, add_node, update_node, remove_node, add_edge, remove_edge, and update_node_config
  • ReAct Reasoning Framework — Agent uses the ReAct (Reasoning + Acting) framework to automatically plan execution steps, supporting multi-round tool calls with observation-feedback loops
  • Canvas Capability Declaration — injects 7 node type definitions and connection rules into the LLM System Prompt, ensuring AI correctly constructs canvas operation instructions
  • Preview Before Apply — Agent shows a change preview (Diff) before committing canvas operations; users confirm with one click to apply, keeping everything safe and controlled
  • Quick Actions — added common shortcut commands for one-click script analysis, canvas layout optimization, and other frequent operations
  • Atomic Operations & Concurrency Safety — backend uses SELECT FOR UPDATE for read-modify-write atomic operations, ensuring canvas data consistency during multi-user collaboration
  • WebSocket Real-time Push — each Agent operation pushes canvas changes in real time via task_update; the frontend CanvasEventBus subscribes and auto-syncs

Collaboration & Sharing

  • Real-time chat in 2D canvas collaboration — team members can send messages directly within the canvas editor, with full message history persistence
  • Project sharing — share entire projects with team members or external users, with edit / read-only permission levels
  • Shared projects accessible via a dedicated "Shared with Me" tab

Canvas Enhancements

  • Node label icons — canvas nodes now display contextual icons based on node type and media MIME type (video, image, audio, text) for better visual clarity
  • HOC file import / export — export any 2D or 3D canvas as a .hoc file (HoldOrange Canvas format) and import it to recreate canvases, enabling portable workflow sharing

Batch Operations

  • Batch copy & move for assets, canvases, and projects — select multiple items and copy or move them to a target project in one action
  • Batch action toolbar available in both Assets Library and Canvas grid views

V2.0.0 (March 10, 2026)

3D Canvas

  • All-new 3D Scene Editor: multi-model import (GLB/glTF/FBX), translate / rotate / scale transform controls, snap grid, ground alignment
  • Skeleton rig visualization: automatic rig metadata detection, bone selection & highlighting, pose editing
  • Animation Retargeting Panel (Mixamo): source / target model selector, 22 semantic bone mappings, one-click apply
  • Motion Capture Import Panel: BVH file drag-and-drop upload, parse validation, bone mapping, root motion toggle
  • Cinematic Camera System: depth of field / chromatic aberration / vignette / motion blur / anamorphic post-processing
  • Camera Path Animation: keyframe recording / editing, real-time preview, duration control
  • 3D generation tasks use the unified queue (content:generate) with WebSocket real-time progress
  • 3D scenes support skybox / environment lighting / ground shadow

Visibility & Permissions

  • Canvas and project visibility settings (private / public / link sharing)
  • Public canvases support read-only access; link sharing supports expiration

Real-time Collaboration (2D Canvas)

  • Multi-user simultaneous editing of 2D canvases
  • Real-time cursor position sync, node locking, and conflict indicators

Provider Expansion

  • Integration with additional AI provider channels, expanding model coverage

V1.6.0 (March 4, 2026)

Canvas & Creation

  • 2D / 3D canvas merge — unified management entry
  • Added group image generation support — some image models can now produce multiple variant images in a single generation
  • Canvas image output nodes now display a stacked card visual hint with offset rotated layers behind the main image, reflecting the number of variants
  • Click the group badge to expand and view all variants, positioned above, to the right, and upper-right of the main image
  • Set as Main action lets you swap any variant with the current main image
  • Auto-collapse: expanded group images automatically collapse when the node is deselected by clicking elsewhere on the canvas
  • Assets Library and fullscreen preview support group image browsing and individual download
  • Page transitions now include animation effects for smoother UX

AI Model Integration

  • Integrated Kie.ai Midjourney API with text-to-image and image-to-image support (Relaxed / Fast / Turbo speed modes)
  • Integrated Doubao Seedream 5.0 Lite image generation model
  • Integrated Nano Banana 2 ultra-fast image generation model

Admin & Management

  • Added admin task management panel — view all users' generation tasks, batch operations, and status filtering

Task Monitoring

  • Task list thumbnails now show a group image count badge for quick identification of group generation tasks
  • Task detail page adds a group image gallery section displaying all variants in a grid with click-to-view and individual download

Bug Fixes

  • Fixed incorrect credit consumption calculation for group image generation
  • Fixed cover image (coverUrl) loss when saving image assets
  • Fixed 3D generation panel being too tall and obscuring coordinate axes
  • Fixed 3D canvas left sidebar onboarding description (scene settings → project settings)

V1.5.0 (February 28, 2026)

AI Model Integration

  • Integrated Jimeng (即梦) official channel — supports Jimeng AI Image Generation 4.0 and Video Generation 3.0 Pro
  • Integrated Doubao Seedream 4.5 and Seedance-1.5-pro
  • Integrated OpenRouter — Google Gemini 3 Flash Preview, Gemini 3 Pro Preview, Gemini 3 Pro Image Preview, Gemini 2.5 Flash Image
  • Integrated fal.ai models — Veo 3.1, Veo 3, Sora 2, Sora 2 Pro, Grok Imagine, Kling 3.0 video generation
  • Integrated Kling 3.0 models (kling-v3-omni, kling-video-o1, kling-v3)
  • Multi-angle model using fal.ai official service (qwen-image-edit-2511-multiple-angles)

Canvas & Creation

  • Removed Agent node from canvas, merged into AI Creative Assistant
  • Added first version of 3D Canvas with multi-model 3D editing and transform, plus image/video generation tabs
  • Added canvas onboarding guide system
  • Added camera parameters panel (shot type, camera movement, visual style)

Admin & Management

  • Models and suppliers now have dynamic configuration association; added model-supplier config module and parameter pricing module
  • Model list and execution buttons display credit cost estimates
  • Added popular model statistics (popular multimodal text, image, and video models)
  • Split admin panel into User / Admin / Super Admin roles; added admin statistics dashboard
  • Added general settings group (upload settings, open registration, welcome credits, etc.)
  • Added supplier statistics panel (success rate, failure rate, and other key metrics)

User Experience

  • Redesigned login, registration, and forgot password pages
  • Improved multi-language support (Chinese & English) and Light / Dark mode styling
  • Integrated SMTP email for forgot password / reset password flows
  • Added user documentation system

Architecture & Security

  • Architecture improvements, security hardening (key encryption, CSRF protection)
  • Performance optimization (caching, connection pooling, serialization)
  • Added audit logging system
  • Unified global error system
  • Model calls support smart fallback / retry with automatic supplier switching and circuit breaker
  • AI comprehensive analysis system optimized over multiple iterations, overall score above 8.5
  • Single-device login restriction per account with security alert emails
  • Storage support for Volcengine TOS