[V6.9.6 Major Update] (Released on 2026.09.20): Refactored the large-model dialogue architecture with full support for both the OpenAI Chat Completions and Responses endpoints; added multi-type document understanding for official models (based on model capabilities, via the Responses endpoint), a large-model global configuration center, and a login device & remote-login detection system; refactored the points system (detailed records for all model usage: consumption / refunds / no deduction on API model errors); added in-dialogue file preview (PDF/Word/PPT/Excel/Text/Markdown/HTML...); upgraded super/admin remote-session security; frontend UI updates and optimizations...

🚀 Feat (Refactor / New / Adjustment)

• Refactored the large-model dialogue invocation architecture to automatically route between the Chat Completions / Responses endpoints based on model capabilities, with a built-in multi-level automatic fallback retry chain on failure
• Added support for the official OpenAI Responses endpoint: official series that match the rules (GPT / o-series / Codex) use the official file input channel
• Adjusted non-whitelisted models (Claude / Gemini / DeepSeek, etc.) to consistently use Chat Completions, avoiding failures caused by being mistakenly routed to Responses
• Added multi-type document understanding for large models: official models support uploading and recognizing officially supported types such as PDF / Word / PPT / Excel / Text / Markdown / HTML (based on model capabilities; for non-PDF file extraction please choose a GPT model, routed via the Responses endpoint)
• Added multiple document parsing submission forms: URL concatenation / Base64 inline / file link, automatically choosing the best option and falling back based on file size and token limits
• Added a large-model API streaming parameter setting: controls the stream parameter submitted upstream, enabled by default
• Added the "Large Model Global Configuration" center: consolidates global policies such as endpoint routing rules and whether to charge on failure; supports built-in defaults, regex syntax, and real-time testing against model names; changes take effect immediately on save without a restart
• Refactored the user-end points details system: unified presentation of consumption, failure refunds, and failure exemptions for all model usage
• Added a points details system to the admin backend, consistent with user-end account records, supporting filtering and display by recharge type, change category, points category, time range, etc.
• Added a zero-charge mechanism for failed model calls: 4xx / 5xx errors, timeouts, and empty replies do not deduct points (controlled by a switch), with anti-abuse and infinite-retry protection
• Adjusted points settlement and transaction records for dialogue exceptions: failures / interruptions / empty replies are all settled through a single settlement exit, eliminating incorrect and duplicate charges
• Adjusted account change record times to a full date-time format accurate to the second (YYYY-MM-DD HH:mm:ss) for precise reconciliation
• Added a login device & remote-login detection system: records login IP (resolved to carrier and city/district), browser, operating system, and device type; added "Login Devices" to the personal center to view the device, IP (region), login and expiry times, and online status of the last 5 logins
• Added precise online status determination: based on dual verification of server-side session validity and token expiry; only the most recent valid login is shown as "Online", all others are shown as "Offline / Session expired", consistent across the user end and admin backend
• Refactored the super / admin account session security model: super frontend and admin backend sessions are slotted by login IP — the same IP can keep frontend and backend sessions at the same time, while logins from different locations force each other offline and record the IP (so a stolen login is noticed immediately); the user end allows only one device online at a time; the admin demo account supports concurrent multi-device sessions, and admin backend logins are not overridden by frontend logins, making it easy for multiple people to demo the backend together
• Added sensitive information masking: IPs and regions are masked when viewed by demo / non-super-admin accounts (IP keeps the first two segments, region keeps the province with city/district replaced by **), effective on both the user end and admin backend
• Added secure cloud storage uploads: when cloud storage is enabled, uploads go to the cloud, anonymous writes to the local disk are prohibited, and guest uploads are limited by short-term and daily counts and by capacity
• Added file upload type control: officially supported types are shown first, and the server directly rejects script / code files to avoid invalid uploads
• Added in-dialogue file preview (PDF / Word / PPT / Excel / Text / Markdown / HTML): supports online viewing, printing, and downloading, and shows details such as type and size; file tags use unified hover tooltips, with a hover-state corner radius matching the message bubbles; Coze Agent dialogues use the same implementation
• Added a pre-selected model system to the admin backend: updated new-generation models by vendor, including OpenAI / Claude / Gemini (only used for binding key pools, does not affect the model name actually called upstream)
• Added a points consumption column to the admin backend dialogue list: shows the points consumed by each AI reply for easy reconciliation
• Added a "Login IP Records" column to the user data list: shows IP (region) and device identifier with adaptive column width; when there are multiple logins, a "More" button opens a popup with the last 5 full records (device / IP region / login and expiry times / online status), with dark theme support
• Added adaptive flow layout for multiple images in user messages: fixed, constant spacing between images (PC 8px / mobile 4px) and even spacing for any number of images (including odd counts such as 3 or 5); up to 4 per row on PC and 2 per row on mobile, wrapping automatically with the last row centered
• Added multi-image adaptation for different browser zoom levels: when zooming to 110% / 125% / 150% narrows the bubble, the number of images per row is automatically reduced and kept centered, so images no longer overflow or get cropped
• Adjusted the global announcement popup: on mobile the height is dynamically limited by the visual viewport with a smaller vertical footprint, so the close and action buttons are no longer hidden by the browser toolbar (PC layout unchanged); the page background gets a frosted-glass blur when the popup opens; the fade at the bottom of the body text shows/hides dynamically with the scroll position (disappears automatically at the top / bottom); the title is pure black / pure white depending on the theme; unified card corner radius
• Adjusted the top membership card color scheme: changed the purple gradient to a champagne-gold gradient, with the title, subtitle, and action button colors adapted to ensure readability
• Adjusted the unified corner radius standard for popup cards: model marketplace (outer card and inner model cards), package purchase, top points card (including the inner balance card), check-in reward (including the tip card), and global announcement
• Adjusted the system action button color standard: the check-in button, the announcement "Don't show again for a week" button, etc. now use a unified high-contrast black-and-white monochrome style (dark background with light text in light mode, light background with dark text in dark mode), consistent with the model marketplace "Switch / In use" buttons
• Adjusted popup close buttons: check-in, package purchase, login devices, and other popups now use the global rounded "×" close button, with dark theme support
• Added horizontal scroll navigation for package purchase: the package list supports horizontal scrolling with the mouse wheel, plus new left/right "‹ ›" arrow buttons (smooth paging on click, continuous scrolling on hover on PC, automatically grayed out at the edges, automatically hidden when the packages don't overflow), consistent with the model marketplace vendor switcher
• Adjusted the personal center "Login Devices" card: unified card corner radius and close button, device types now use standard monochrome icons, and non-current sessions are uniformly shown as "Offline"
• Adjusted the AI face-swap painting module: images can no longer be uploaded or submitted, as there is currently no relevant API available on the market (globally), and current policies do not allow it
• Other new features, security optimizations, UI improvements, etc.....

🔨 Fix

• Fixed an issue where, in special cases, repeatedly triggered Midjourney / Niji failure callbacks refunded points multiple times; painting task refunds are now idempotent
• Fixed an issue where the global announcement popup's height constraint failed on mobile, with the content nearly filling the whole screen and making the buttons hard to tap
• Fixed an issue where some built-in models had leftover default upstream configurations; the admin backend model configuration is now uniformly authoritative
• Fixed several stability, security policy, and UI detail issues

[V6.9.5] (Released on 2026.08.22): Added a dialogue navigation bar for the chat large-model module, a dialogue demo data feature, and "Gallery Picks / My Paintings" switch tabs with separate loading logic; refactored the personal center page UI; implemented Chinese font subsetting; performance optimizations: optimized for the iOS single-page memory limit, etc.....

🚀 Feat

• Added a dialogue navigation bar to the chat large-model module: automatically enabled when the current dialogue group contains ≥ 3 "user questions"
• Dialogue navigation bar features: collapsed state, expand on hover, text truncation, adaptive + scroll loading, click-to-locate, etc.
• Added a dialogue demo data feature: not restricted by the "System Function Test Demo Site" switch, and can be freely assigned and removed
• Added a "Dialogue Group List" page to the admin backend, allowing any user's dialogue groups to be designated for site-wide demo display (up to 5)
• Demo dialogue groups are visible to all users and read-only, with an orange "Dialogue Demo Data" label in the title; the owner is not restricted, and deleting the group automatically removes the demo designation
• Added "Gallery Picks / My Paintings" switch tabs at the top of the painting list page, with separate loading logic
• Added a switch configuration for the system voice playback feature
• Refactored the personal center page UI with a brand-new card style and animations
• Refactored the painting module operation tips UI
• App plaza cards on mobile changed from three per row to two per row, with the logo and text moderately enlarged (PC unchanged)
• Adjusted professional painting models on mobile: the input box is not focused and the keyboard does not pop up by default when entering the page; after the user taps the input box, it focuses and accepts input normally
• Optimized painting module list positioning: chat-style following — on completion, the list follows the latest result only if the user is "still at the bottom"; if the user has scrolled up to browse, it won't interrupt
• Performance optimization: Chinese font subsetting — a single font weight reduced from 4.2MB → 0.9MB, covering all commonly used Chinese characters; rare characters automatically fall back to system fonts; the original font source is kept as a backup; SEO structure optimization
• Performance optimization: single-page resource download size reduced by about 50%
• Performance optimization: App plaza resources on mobile are loaded in batches, significantly reducing the risk of crashes on low-memory devices
• Other new features, security optimizations, UI improvements, etc.....

🔨 Fix

• Fixed repeated reloading/refreshing when iOS devices (Safari / WeChat / Chrome) accessed the "App Plaza" and hit the iOS single-page memory limit
• Fixed the issue where "Send verification code" was not clickable when reopening the phone number binding popup

[V6.9.4] (Released on 2026.07.27): Added the standalone Nano Banana 2 painting module; refactored the system painting module architecture with an extracted reusable foundation and asynchronous task orchestration; added support for forcing specified users offline; 200% performance optimization for painting submission configuration queries on the submission path; added support for the "System Function Test" feature, etc.....

🚀 Feat

• Refactored the system painting module architecture, extracting a reusable foundation and adding asynchronous task orchestration
• Developed the standalone Nano Banana 2 painting module (gemini-3.1-flash-image, gemini-3-pro-image)
• Admin backend supports custom Nano Banana 2 API integration and custom extended model versions (e.g., adding a custom gemini-3.1-flash-lite-image Banana model)
• Supports text-to-image, reference image painting, custom resolution and aspect ratio, and other parameters
• Supports "Paint the Same" reuse (syncs the prompt, resolution, aspect ratio, and other parameters)
• Nano Banana 2 image editing API supports multiple smart reference images
• Supports base multiplier billing configuration for image editing, and private storage (original images, thumbnails)
• Supports dynamic display of points prices for frontend users, points refund logic for failed paintings, and a scheduled compensation task for Nano Banana 2 paintings
• Optimized the painting VO aggregation API, and added synchronized frontend display of private storage progress status
• Added support for forcing specified users offline
• Added email whitelist configuration (only allow registration from specified email domains; "+" alias emails can also be blocked)
• Sensitive configuration value masking (unified entry point for queries by non-super-admins)
• API address values keep only the protocol + top-level domain suffix; improved masking for keys / tokens
• Added character limit logic for large-model dialogue input submission, with an admin backend configuration switch
• Added composite indexes for painting data to improve query performance
• 200% performance optimization for painting submission configuration queries on the submission path
• Added support for the "System Function Test" feature: when enabled, all users' AI generation requests (dialogue, painting, video, etc.) are blocked at the system level, no upstream large-model APIs are called, and only a demo-site notice is returned. Suitable for scenarios where generative AI filing has not yet been completed, effectively avoiding compliance risks
• Optimized the issue on mobile where prompts could not be entered and submitted again after submitting a painting, preventing post-generation polling scroll from collapsing the input bar
• Other new features, security optimizations, UI improvements, etc.....

🔨 Fix

• Optimized the issue on mobile where prompts could not be entered and submitted again after submitting a painting, preventing post-generation polling scroll from collapsing the input bar

[V6.9.3 Major Update] (Released on 2026.06.23): Major refactor of the GPT-Image-2 painting module, upgraded system risk-control sensitive-word detection, AI large-model dialogue core refactored into decoupled upstream/downstream connections + an asynchronous execution model, end-to-end reliability hardening of the streaming (SSE) pipeline, full persistence of dialogue data, multi-session concurrency isolation, visualization of stopped / abnormal states, system UI refactors and optimizations.....

🚀 Feat

• Refactored the AI dialogue core into an asynchronous execution model, fully decoupling the "backend ↔ large model" upstream connection from the "frontend ↔ backend" downstream connection
• Upstream requests now run independently of the client lifecycle: switching sessions / navigating to other modules / closing the page / exiting the client does not interrupt generation or persistence
• Extended system risk-control sensitive-word detection to the painting module: GPT-Image image generation / image editing is intercepted at the system level before submission
• Extended system risk-control sensitive-word detection to Midjourney / NIJI / DALL-E text-to-image: prompts in the input box are intercepted when they hit sensitive words
• Large-model dialogue now supports native camera photo upload (native camera on mobile + webcam devices on PC)
• Mailer email service now supports configuring whether SMTP uses SSL; if not set, it is determined automatically by port: 465 is true, other ports (e.g., 25/80/587) are false (configured in .env, see the latest .env.example file)
• Adapted to official Claude / GPT-5 / o3 series models, which do not need the temperature parameter, with the frontend disabling it accordingly
• Added custom resolution support to the GPT-Image-2 painting module
• Optimized the GPT-Image-2 painting module to submit the size parameter strictly within the range required by the official API
• Optimized how the GPT-Image-2 painting module displays aspect ratio, resolution, and quality definitions to frontend users, to avoid misunderstandings
• Added GPT-Image-2-Dev painting model integration with separate billing configuration in the admin backend (this model does not support the quality parameter)
• Refactored and optimized GPT-Image-2 painting concurrency limits: supports per-user concurrency and system-wide concurrency
• Added frontend display switch configuration for each painting model (MJ / Niji, Dalle3, GPT-Image-2 related painting models)
• Optimized the GPT-Image-2 painting module: when "Edit Image" is not enabled in the admin backend, painting editing is unavailable on the frontend
• Added handling for GPT-Image-2 reference image painting where oversized images may be blocked by the server or gateway
• Optimized permissions for the system Admin demo-site account viewing backend user data (including AI large-model dialogues, AI painting results, and system resource image URLs): not viewable
• Refactored the model marketplace UI, adding display of model capability descriptions and a latest-model badge
• Implemented a high-fidelity Apple liquid-glass button effect (painting module action buttons, top buttons)
• Painting models now support a painting progress bar and a blurred background image UI at the bottom
• Optimized the user-end UI: action buttons at the bottom of AI large-model reply bubbles changed to icon-only + hover tooltips
• Added sensitive-word detection to the mind map module, reusing the same risk-control pipeline as the dialogue core, validated on submission
• Painting / mind map sensitive-word hits are now handled the same way as in the dialogue module: intercepted upfront, not submitted to the upstream API, no balance check, no charge, instant response
• Added full logging of violation records triggered by sensitive words in painting / mind map; violations by both users and guests are synced to the admin backend "Violation Detection Records"
• Added the OpenAI file protocol parsing method for document files in large-model dialogues (i.e., supports the Chat Completions OpenAI file protocol and URL concatenation for documents)
• Adjusted large-model dialogues to prefer the real usage returned by the upstream SSE; falls back to the original local system estimate when missing
• Adjusted all large-model dialogues to a unified 10-minute timeout (to accommodate Base64 conversion + large request bodies + potentially slow upstream docx/pdf parsing)
• Added a security policy for user file uploads in large-model dialogues (enforced interception): server-executable script extensions are directly rejected by the upload API with a 400 response, preventing webshells from being triggered via reverse access
• Added automatic fallback when uploaded PDF files exceed the size limit (under the OpenAI file protocol)
• Added large-model configuration for extra billing on PDF Base64, and support for setting the maximum document size
• Added upfront interception when AI video (Pika) upstream integration parameters are not configured: uniformly returns "The system has not configured the related large-model feature", without sending a request or charging
• Upgraded streaming responses to incremental persistence, changing from "save once at the end" to "save while generating" during SSE according to a throttling strategy
• Added session resume display: content already received is retained after the client disconnects, and switching back to an in-progress session directly loads that round's snapshot without resending the request
• Added a closed-loop active stop pipeline: clicking "Stop answering" sends an interrupt signal that explicitly cancels the upstream request, stopping losses and saving quota in time
• Added support for saving the partially received content when actively stopped, marked with the "User actively stopped the question" status
• Added a fallback for first-token timeout / empty responses: when the upstream doesn't respond for a long time or the final response is empty, it uniformly falls back to "No data returned / API response timed out" and is persisted
• Added synchronized frontend display of fallback states, preventing dialogue bubbles from being stuck on "Thinking" or blank
• Added full request logging, based on "whether a request was actually sent to the large model": success / timeout / exception / active stop are all guaranteed to have corresponding records in the persistence layer
• Added support for reconciling full logs against upstream API billing (upfront interceptions where "the request was never sent", such as insufficient balance or sensitive words, are excluded)
• Added the dialogue status labels "User actively stopped the question" and "No data returned / API response timed out", shown on both the frontend and admin backend
• Added multi-session concurrency isolation: UI writes from streaming callbacks are validated against the owning session group scope, so asynchronously completed answers do not update a session that is not currently active
• Optimized the recovery logic after leaving a page to idempotent loading (based on persistence-layer snapshots), eliminating duplicate requests
• Other new features, stability enhancements, security optimizations, UI improvements, etc.....

🔨 Fix

• Optimized boundary handling when switching session groups / pages during streaming generation, preventing asynchronously completed answers from being written into other sessions in special cases
• Optimized the recovery mechanism when leaving and returning to a page, preventing duplicate requests or duplicate charges in special cases
• Optimized persistence logic for client interruption scenarios (switching / closing the page / active stop), preventing missing or incomplete dialogue data in extreme cases
• Optimized the fallback display on empty responses / exceptions, preventing dialogue bubbles from being stuck on "Thinking" or blank
• Optimized AI video (Pika) upstream error handling: HTML error pages / API exceptions returned by the channel are no longer passed through to the frontend as-is; they are only logged on the backend and replaced with a friendly message

[V6.9.2] (Released on 2026.05.25): Added the standalone GPT-Image-2 painting module (text-to-image, reference image painting, image editing, custom quality / quantity / aspect ratio and other parameters), custom GPT-Image-2 API integration, custom extended model versions, brand-new reference image upload and UI animations, a website live-stream entry, refactored painting module UI, security optimizations, other new features, various UI improvements...

🚀 Feat

• Developed the standalone GPT-Image-2 painting module
• Supports text-to-image, reference image painting, image editing, custom quality / quantity / aspect ratio and other parameters
• Supports "Paint the Same" reuse (syncs the prompt, aspect ratio, quality, quantity, and other parameters)
• GPT-Image-2 supports online editing of generated painting results, including brush / box-selection editing
• GPT-Image-2 image editing API supports multiple smart reference images (up to 8)
• Admin backend supports custom GPT-Image-2 API integration and custom extended model versions
• Supports billing configuration for quality multipliers (low, medium, high, auto), the base generation multiplier, and the base image editing multiplier
• Billing rule: ceil(base × quality multiplier × quantity)
• Supports private storage (original images, thumbnails)
• Supports dynamic display of points prices for frontend users, and points refund logic for failed paintings
• Supports a scheduled compensation task for GPT-Image-2 paintings
• Refactored the painting module UI
• Brand-new reference image upload UI and animations
• Unified painting VO aggregation API
• Optimized the custom application workbench UI
• Added "One-click cleanup of user-deleted paintings" (soft delete -> physical delete)
• Refactored MJ/Niji instruction integration logic for different reference images (if a reference image is not uploaded, its corresponding -- instruction is not included)
• Added a website live-stream entry, with automatic PC + H5 detection
• Mobile supports URL Scheme invocation code, and an intermediate operation prompt page inside WeChat
• Added support for split-table aggregated queries across painting categories (frontend aggregated API / admin backend split-table and aggregated queries)
• Optimization: when using the official WeChat Pay API, the JSAPI payment flow is only used in "WeChat environment" + "mobile", avoiding the issue where payment was impossible inside the PC WeChat browser
• Optimized admin backend data query performance with millions of dialogue and painting records: concurrent count execution, using RedisCacheService
• Added user snapshot logic triggered by prohibited words in the risk-control module
• Under the PC Notion style, added automatic collapsing of multiple menu bars and an external-open UI indicator
• Fully masked usernames, email addresses, etc. for the admin backend admin demo-permission account
• Added question/reply labels to large-model AI dialogue records
• Other new features, security optimizations, UI improvements, etc.....

🔨 Fix

• Fixed permission issues in some scenarios for preset applications, and the issue where they could not be copied

[V6.9.1] (Released on 2026.04.08): Redis distributed lock optimization for Coze batch Agent import, new system multi-language internationalization (full i18n) + admin one-click AI internationalization, added support for Midjourney latest V8 flagship model painting, login/registration page refactor, security optimizations, other new features, various UI improvements.....

🚀 Feat

• Added Redis distributed lock: when batch importing large amounts of metadata from Coze configuration, prevents duplicate data and optimizes batch import speed and database write performance
• Added system multi-language internationalization (full i18n: not conventional frontend framework fixed-content i18n — supports backend configuration info + backend response info internationalization)
• Supports automatic website language internationalization based on user client access IP + browser default language setting auto-internationalization
• Added one-click AI internationalization for all configuration content in the admin backend
• Added support for Midjourney latest V8 flagship model version painting (development complete; V8 is currently in Alpha stage officially — once officially released and supported by the API, it will be directly available)
• Added support for MJ V8 high-consumption multiplier billing and admin backend settings
• Professional painting module now supports dynamic display of points prices for each painting model and each painting operation
• Refactored login/registration page: brand-new creative animated login interface (built from scratch using native CSS + Vue 3 + Composition API, enhanced creative version)
• Refactored AI large model dialogue attachment upload UI, with upload progress status indicator support
• Optimized app-store Agent backend logic and user-end UI
• Under Notion style, sidebar added "View More Menu" feature
• Added "Ctrl+K / Command (⌘)+K" shortcut for creating new AI conversations
• Model marketplace modal UI optimization, added support for model provider sliding buttons (PC and mobile H5)
• Removed active input outer border display from all system input fields, replaced with tactile shadow effect
• Optimized default first-time light/dark theme logic: prioritizes user PC/mobile device theme color, then falls back to system time (day/night)
• Optimized user registration bonus model points logic with security improvements to prevent attackers from exploiting special methods or scripts to modify bonus amounts
• Admin backend cleanup of failed painting data, long-unfinished painting tasks, and other batch import operations now includes real-time logs, progress logs, and popup UI display
• Added API key pool: delete all key configurations, and batch select and delete API keys features
• Added API key pool: when adding a KEY, select unconfigured key model feature, and added support for displaying the number of keys under that model
• Optimized backend response messages to use friendly user-facing prompts
• Other new features and various UI improvements.....

🔨 Fix

• Fixed the issue where GPTs agents found via full-web API search could not be added if the agent name or description contained emoji characters
• Optimized mobile experience: fixed the issue on iOS + third-party in-app browsers where the bottom input box would be pushed behind the menu bar

[V6.9.0] (Released on 2026.03.01) - Major system refactor version: 2026 full UI overhaul with brand-new card / Notion-style themes, immersion mode, video creation module UI refactor, multiple custom components, MJ painting submission logic optimizations, and many new features, etc.....

🚀 Feat

• 2026 user-end UI major refactor: brand-new card style / Notion-style themes, allowing users to freely switch the UI
• Optimized MJ painting submission logic: when the API submission result is "queued", the backend will no longer treat it as a submission failure exception (it will be treated as successfully submitted)
• Added AI dialogue immersion mode (on PC only the large-model conversation content is shown for a more focused dialogue experience)
• Under the Notion-style UI, unified management of AI dialogue history and application dialogue lists in the sidebar
• Developed a floating side menu bar (effective in AI dialogue immersion mode, automatically displayed when the mouse hovers over the left area)
• Added "Add Menu" feature (left-side menu items can be quickly collapsed/collected via right-click), allowing users to freely remove/collect function modules into "Add Menu"
• Added support in the admin backend for Agent application management to preview Coze Agent persona and reply logic information when editing
• Added support in the painting plaza and video creation module for hovering over video works to auto-preview playback, including video duration and progress bar display
• Brand-new painting input box UI (replicating the Jidmeng AI painting input UI), which automatically expands or collapses based on input status and painting list state (scroll up, scroll down, bottom)
• Adjusted painting scroll positioning logic so that submitting a painting no longer auto-scrolls to the bottom; it only scrolls when the painting status changes
• Refactored the top bar UI with a brand-new, clean layout and design
• Added top-bar model balance cards (related data is loaded on click/hover)
• Added top-bar personal center card (shown on click/hover)
• Personal center card supports displaying points details, membership status, membership expiry time, and membership activation button
• Refactored the video creation module UI and adapted it for mobile H5
• Added a droplet-style switching animation for the bottom menu on mobile
• Brand-new video player component (colors, buttons, layout)
• Refactored the promotion plan page layout (fixed PC-side promotion statistics panel, with the records list on the right scrollable in a table)
• Brand-new model icons (basic model balance, advanced model balance, professional painting balance, Agent points balance)
• Top bar now automatically adapts to different module pages and shows the corresponding model balances
• Brand-new greeting messages and App plaza top-tip typewriter animations
• Optimized AI dialogue module operation button UI
• Refactored AI dialogue module new-conversation page UI, supporting display of the small explanatory text from previous versions
• Added brand-new hover/click animations for the menu bar modules
• Added an "About Us" info button (showing custom official website, user agreement, privacy policy, site information, ICP number, public security record number, etc.)
• Optimized AI Agent module page UI
• Adjusted AI large-model dialogue information (no longer displays the personal avatar)
• Optimized admin backend operation UI
• Other feature additions and optimizations.....

🔨 Fix

• Fixed the issue where MJ painting submissions without a scale value from the frontend would fail to be written to the database and directly report "submit Action task failed..."
• Fixed the issue of Coze Agent dialogues where, in special cases (continuous same-type characters), content overflowed the dialogue background