English
Current location:home App StoreZ-blogPHPplug-intext
FinchUI AI Writing Assistant (Z-Blog Plug-in)

FinchUI AI Writing Assistant (Z-Blog Plug-in)

$59.38
Xinglan Studio(Seller)
hnyanshan@outlook.comqq91446648015639981097WeChat
QR code
classification
plug-in
update date
2026-09-30 01:26:26
Cumulative visits
8026
scores
No one has scored yet
label
FinchUIAI writing assistantDeepSeekbean bagKimithousand querywisdom spectrumChatGPTBatch generation of articlestimedly-published
service commitment
original worksfree upgradefree experienceUnlimited domain namespermanent useassist in the installationafter-sales protection
purchase quantity
Buy Now Add to shopping cart collection
  • introduced
  • parameters
  • update log
  • evaluation
  • guide

FinchUI AI writing assistantIs a practical AI content production plug-in for Z-BlogPHP. The core goal is not to just help you "write an article", but to integrate content generation, content rewriting, mapping, publishing, advertising, site distribution andscheduled taskString it into a complete set of sustainably running workflows. For websites that need to update content for a long time, it solves not the efficiency of a single creation, but the efficiency of the entire content production chain.

The current version supports Volcano Ark (bean bag）、Kimi、DeepSeek, Alibaba Cloud Refining (Thousand Questions), OpenAI (ChatGPT), xAI Grok, Xiaomi MiMo and other AI platforms. You can freely choose models based on budget, speed, writing quality and business direction, or you can assign different models to different batches of tasks in the same backend, taking into account cost control and content quality.

If you want to reduce repetitive writing time and increase the frequency of content updates, while retaining control over article quality, release pace and multi-site distribution, this plug-in will be easier than simply calling AI dialogue tools, and will be more suitable for corporate official websites, industry websites, long-term operations of content sites and website matrix projects.


Applicable groups and industries

This plug-in is suitable for website operators, corporate promoters, foreign trade promoters, industry portal editors, self-media teams and multi-website operation managers who need to stably update content. Whether you are updating corporate news on a daily basis, writing product introductions, organizing industry information, or laying out long-tail SEO content in batches or maintaining multiple columns or multiple sites, it can significantly reduce manual duplication.

Especially suitable for corporate official websites, independent foreign trade stations, B2B product stations, industry information stations, local portal stations, vertical content stations,website navigationStation and multi-site unified operation projects. It is suitable for continuous updates at a single station, and also for operating scenarios that require collaborative release from multiple columns, multiple content types, and multiple sites.


Model access and selection suggestions

If you are just starting to access, it is recommended to start with Volcano Ark or Alibaba Cloud Bailian first, which have relatively lower access barriers and more comprehensive model coverage; if you value the quality of Chinese writing and reasoning performance more, you can continue to introduce DeepSeek, Kimi, Grok or MiMo; if your business is biased towards overseas markets or international content, then consider accessing OpenAI. Combined this way, it often achieves a more practical balance between cost, stability and content quality.

Combined with the latest preset models built into the current project, it is recommended to use doubao-seed-2-0-mini-260215, qwen3.6-flash, mimo-v2-flash or gpt-4o-mini first for daily batch writing. These models are more balanced between speed, cost and stability; If you value the quality of writing, Chinese expression and overall completion more, recommend doubao-seed-2-0-pro-260215, doubao-seed-1-8-251228, deepseek-v4-pro, qwen3.6-plus, qwen3.6-max-preview, grok-4 or mimo-v2-pro; If the task is biased towards complex reasoning, structured analysis, or long-term planning, it is recommended to give priority to deepseek-r1-250528, kimi-k2.6, glm-5.1, or grok-4.20-reasoning.

It should be noted that OpenAI's GPT series models are limited to overseas users or websites for overseas business; if your server is deployed in China, it is not recommended to use this series of models.

In addition to the model itself, many users will also care about "whether this model can be searched online." Current plug-ins have whitelisted online search capabilities managed by model, and the background will clearly distinguish between models that support and do not support online search. The default whitelist currently covers corresponding models for platforms such as Volcano Ark, Kimi, Alibaba Cloud Refining, OpenAI, Grok and Xiaomi MiMo;DeepSeek currently does not provide online search options, so it will not appear in online search settings, which can significantly reduce misselections and mismatches.

If you need to further fine-tune the model output style, the current version also supports customizing the temperature by platform, and distinguishes between "thinking model" and "non-thinking model" settings. Except for official fixed-value models such as Kimi K2.6 and K2.5, which are preferentially processed according to official rules, other platforms can be adjusted manually; the default values have also been reorganized to think 0.7 and non-think 0.4 according to writing scenarios, supporting arbitrary decimal input, and can be restored to default values with one click.


Highlights of new functions after this supplement

The big model platform is more complete

In addition to the original mainstream platforms, the current version has added xAI Grok and Xiaomi MiMo support, and preset a more practical default model list for each platform. Alibaba Cloud Bailian's default models have also been simultaneously updated to a version that is more suitable for the current writing scene. A restore default button has also been added to the model setting area of each platform in the background, allowing you to quickly return to recommended configurations after trying to customize the model.


The background settings are more compact and task management is more efficient

Originally, multiple model platform configuration areas would stretch the background page very long, but now it has been changed to the Tab switching method. The basic settings page and the new task page can be quickly switched by platform, and there is no need to scroll down to find configuration items. For sites with large tasks, although this adjustment does not change the original parameters, it significantly reduces the visual burden on background operations.

The task management page also adds the ability to modify task models in batches. After selecting multiple tasks, you can directly switch models and platforms in a unified manner, which is suitable for batch migration of historical tasks, unified replacement of lower-cost models, or quick adjustment of execution plans when a certain platform is temporarily unavailable.


The library ability is more practical and the picture matching hit rate is higher

In terms of gallery management, in addition to the original Unsplash, Pexels and Volcano Ark AI maps, two more easily accessible free map sources, Pixabay and Openverse, have been added. In this way, even if you don't want to rely entirely on AI to generate pictures, you can first make up for most article photo matching needs through free libraries.

In terms of image matching logic, the current version also optimizes the hit method of local libraries. The system no longer only looks at a single keyword, but will integrate article tags, image keywords returned by AI, and original task input content to match pictures in the local gallery. In particular, it has been revised for the scenario where manually uploading pictures uses Chinese tags., which can match existing material more stably.


Core highlights

Automatic release of website navigation

FinchUI AI Writing Assistant is not only suitable for ordinary article publishing, but also supports automatic content production for website navigation topics. After entering the target URL, the plug-in will use built-in crawling logic to extract the title, description, keywords and main content of the page, and then hand it to the AI to generate a completely structured and SEO website introduction article, reducing the time needed to manually organize the information and write the introduction.


AI creation

The plug-in supports multiple creation modes such as keyword creation, title creation, website navigation, and article rewriting. For sites that need continuous updates, you can generate content in batches around topic terms, industry terms, product terms and long-tail terms; for existing content assets, you can also continue to revitalize old articles through rewriting tasks to maintain site activity and content coverage.


Batch tasks vs. scheduled tasks

Batch tasks are one of the core implementation capabilities of this plug-in. You can add a large number of tasks at one time, and set parameters such as AI model, classification, author, review status, and release time. The release time also supports random allocation of scope, which is more suitable for the update rhythm of real websites. Coupled with the planning task interface, you can combine "batch generation" and "timedly-published"Really string it together to achieve unattended automated updates.


Batch rewriting of articles

After selecting the article rewrite type when adding tasks, you can enter a single article ID, multiple comma-separated IDs, continuous ranges, and mixed formats to create rewrite tasks in batches. When rewriting, AI will try its best to retain the image tags and basic attributes of the article in the original text, and only update the title, body, summary and tags, which is suitable for refreshing old texts, supplementing inclusion, adjusting expression methods, and improving content coverage.


Prompt template

The plug-in supports a complete prompt template system, which not only provides default templates such as keyword creation, article title, website navigation, and article rewriting, but also supports custom templates, import and export templates, and variable substitutions under different task types. This part is very practical for users who need to continuously tune prompt words and precipitate writing methodology. Each template also provides a version with pictures, which is convenient for direct linkage with the automatic picture matching process.


Editor AI Assistant

If you use both FinchUI wangEditor5 Editor, you can also embed the AI assistant button directly in the editor toolbar. It supports 12 auxiliary capabilities such as rewriting, error correction, continuation, summary generation, title optimization, label extraction, Chinese-English translation, tone adjustment, content expansion and content streamlining. It is suitable for continuing to speed up during the manual refinement stage, rather than completely relying on One-time generation.


Library management and automatic drawing

The built-in gallery management system supports obtaining pictures from Unsplash, Pexels, Pixabay, and Openverse, and also supports calling the Volcano Ark Wensheng model to generate AI original pictures. All pictures are stored through the Z-BlogPHP standard attachment system, which is consistent with manual upload logic, and is compatible with the fui_oss object storage plug-in, making it easy to automatically synchronize to cloud storage.

The system can intelligently insert pictures based on article content, tags and image keywords. The background also supports manual uploading, editing tags, batch deleting, viewing sources and counting usage times. For websites that need to establish a dedicated material library, this part can significantly improve the efficiency of drawing and reduce the problem of content homogenization caused by repeated use of drawings.


Advertising management

The plug-in has built-in advertising management functions, which can uniformly maintain text advertisements, custom code advertisements, image advertisements and mixed advertisements, and supports automatic insertion in the task of creating and rewriting articles. You can select multiple insertion locations such as the beginning of the text, the end of the text, or the beginning of the text and the end of the text according to the style of the site. At the same time, you can control the expiration time, replace old advertisements in batches, or remove old advertisements from historical articles, and put content production and advertising in the same set of processes.


Website management and multi-site group sending

If you operate more than one site, the plug-in also supports multi-site management. Through API addresses and keys, you can uniformly access multiple Z-BlogPHP sites in one background, synchronize remote classification and author information, switch corresponding data by site when creating tasks, and achieve cross-site content distribution. Planning tasks also support site parameters and classification parameters, making it convenient for single-station execution, classified execution, and multi-station parallel processing.

The supporting receiver plug-in "FinchUI AI Writing Assistant Multi-website Group Sending Site" can be downloaded from https://app.zblogcn.com/?id=52794 Get it, and you can complete remote publishing through the security key after accessing it.


Instructions for use

  • The large model interfaces accessed in plug-ins are essentially paid services. Some platforms will provide test quotas or free quotas for new users, but if you want to use them stably for a long time, you still need to pay according to the charging rules of each platform. Buying a plug-in itself does not mean that you are qualified to call the large model unlimited times and unlimited quantities.
  • Manual review is always recommended for AI-generated content. Whether it is creating new articles, automatically rewriting or website navigation content, whether it is ultimately published, whether to adjust titles and summaries, or whether to retain accompanying pictures, it is recommended to conduct secondary editing based on site positioning and business needs.
  • The frequency of execution of planned tasks is also recommended to be reasonably controlled. Although excessively high call frequency can increase output speed in a short period of time, it will also bring additional interface costs and server pressure, especially in multi-site parallel task scenarios, it is necessary to control the pace.
  • Please abide by relevant laws and regulations, ensure the compliance of content generated, published, and collected and organized, and use it reasonably in accordance with the interface rules, content rules and image copyright rules of different platforms.


How to apply for a large model API Key

  • If you are not familiar with how to apply and configure large model API keys, don't worry. We can provide remote assistance to help you complete platform registration, real-name authentication, API Key creation, and parameter filling in the plug-in.
  • The usual process is to first select the platform you want to use, complete account registration, real-name authentication, service opening or account recharge according to the platform requirements, then enter the developer background to create an API Key or call the key, and finally fill in the key in the corresponding platform settings area in the plug-in background.
  • If you are not sure which platform to choose, or want to give more practical model suggestions based on your site type, budget and server environment, we can also help you determine an access plan together. The plug-in itself is not limited to the few platforms that are currently built-in. As long as there are appropriate API interfaces and clear requirements, it can continue to be expanded in the future.
AI content tools are not just installed and finished. Whether model selection, interface configuration, scheduled tasks and release processes can run stably also affects the actual user experience.

  • One purchase, long-term use

    One-time purchase for permanent useno mandatory subscriptionPlugin fees do not include large model API call fees

  • Continuous upgrades and compatibility maintenance

    Permanent free upgradefunction iterationbug fixcompatibility optimization

  • Experience first, then decide

    You can experience the core process firstConfirm the content scenario that suits youReduce trial and error

  • Configuration assistance is not missing

    Assist in applying and configuring API KeysAssistance in selecting modelsAssist in setting up scheduled tasksAssist with accessing the editor and multi-site publishing

  • Multi-channel after-sales support

    work order systemQQWeChatremote assistanceBeginners can follow the process to complete the configuration

Batch creation and rejuvenation of old texts

Instead of just writing an article, turn long-term content updates into schedulable tasks
Create tasks around keywords, titles, URLs, and existing articles; once the amount of content is up, it can still be generated, rewritten, and published in batches according to the rules.
  • core

    Four content input methods

    Keywords creationTitle creationwebsite navigationarticle rewritingA set of background coverage for new content and old content updates

  • core

    batch creation task

    Text input line by lineTXT file importCreate multiple tasks at onceSuitable for keyword library and topic selection list

  • recommend

    Batch rewriting of old text by range

    Article IDs support single, multiple, range, and mixed inputsThe range is limited to 1000 articlesSupport for summarizing articles by categoryManual ID and classification results are automatically deduplicated

  • core

    Controllable retention of important information in old text

    The original label can be retainedKeep original picturesCategory, author, post time, and comment settings remain unchangedUpdate only title, body, summary, and tags

  • recommend

    The URL content can be converted into an introductory article

    Grab website titles, descriptions, and keywordsFailure to fetch is automatically downgraded to basic URL analysisThe original URL can be saved to the specified article field

Task management and schedule release

Put build, check, retry, and publish into a traceable content workflow
Long-term posting is most afraid of scattered tasks, no records of failures, and runaway release time. Here, task status, batch adjustment, and plan execution are centrally managed.
  • core

    complete mission lifecycle

    not performedqueuegenerationhas completedfailurepublishedTask execution results can be traced

  • recommend

    Filter and adjust by operational dimension

    Keyword, status, author, time, task type, site filterBatch retry failed tasksModify models, templates, and publishing settings in bulk

  • core

    Publish parameters clearly set once

    authorclassificationaudit statusRelease timeframerandom release timeAI automatic classification and alternate classification

  • core

    Timed task execution continuously

    Pagoda Panel Timed TaskSecurity code verificationIP whitelistProcessing 1~ 10 tasks per batchExecution scope can be limited by site or category

  • advanced

    Task configuration is migrable

    Export JSON by filterCross-site import tasksRe-map classification, author, and article IDs when importing

Multi-model and hint templates

Select models by task, budget, and content requirements, rather than being tied to a single platform
Different sites, columns, and tasks have different requirements for speed, cost, networking capabilities, and content style. Models and templates can be configured according to actual operating strategies.
  • core

    Seven types of mainstream model platform access

    Volcano Ark (Bean Bag)KimiDeepSeekAlibaba Cloud Bailian (Thousand Questions)OpenAIxAI GrokXiaomi MiMo

  • recommend

    Separate control of model parameters

    Different models can be selected for different tasksDifferentiate network search capabilities by modelThinking/Not Thinking Temperature Set SeparatelySupport for arbitrary decimals and restore default values

  • core

    Built-in common content templates

    Keywords creationTitle creationwebsite navigationarticle rewritingNews TodayTechnology Industry NewsIncluding online search and version with pictures

  • recommend

    Cue words can be precipitated for a long time

    custom templatevariable substitutionTemplate Import ExportOne-click recovery built-in templateAvoid being unable to return after wrong modification or deletion

Gallery, automatic mapping and visual marking

Bring the generated articles closer to a publishable state and gradually precipitate reusable image assets
Content publishing should not be stuck on finding pictures, adding labels and filling in Alt descriptions. Gallery capabilities revolve around image sources, content matching, visual recognition and sorting of existing materials.
  • core

    Multi-source image unified management

    UnsplashPexelsPixabayOpenverseVolcano Ark AI mapmanually uploadThe picture enters the Z-BlogPHP standard attachment system

  • core

    Match by content and automatically map

    article tagtask original keywordimage keywordsweighted hit strategyArticles can be automatically inserted using the image version template

  • recommend

    The visual model complements the image information

    Output Chinese label and Alt descriptionSupport Volcano Ark, Kimi, OpenAI, Grok, Alibaba CloudIt can be automatically triggered when uploading an image without a label.

  • advanced

    Existing galleries can be sorted out in batches

    Single sheet or batch AI markingConcurrent count adjustable from 1 to 20Default Concurrency 3Time-out seconds configurableReal-time display of completion progress

  • optional

    Image source and storage controllable

    Customize image source priorityCache →Pexels→Pixabay→Openverse→Unsplash→AICompatible with fui_oss object storage plug-ins

Editorial refinement and advertising

AI completes the basic work in batches, and the manual can continue to control the expression and transformation details on the editing page
Automation doesn't mean giving up human moderation. Editor AI and ad management allow content to retain a controlled refinement and operation entrance before and after publication.
  • recommend

    12 AI aids in the editor

    Need to cooperate with FinchUI wangEditorRewrite, correct, continueAbstract, title optimization, tag extractionChinese-English translation, tone adjustment, content expansion and streamlining

  • performance

    Edit response policies can be set individually

    Deep thinking is turned off by defaultValue by Non-Thinking TemperatureMore suitable for editing page real-time assistance

  • core

    Ads can be automatically inserted with content tasks

    Ads can be selected for both new and rewritten tasksText, images, custom code, mixed adsbeginning, end, beginning + end

  • recommend

    Centralized advertising lifecycle management

    expiration date controlAutomatic stop delivery when it expiresExpired advertisements in the article can be cleaned regularlySupport batch replacement and usage statistics

Multi-site mass delivery and site-side access

One background arranges content tasks for multiple sites, reducing multi-background switching
Suitable for enterprise multi-site, industry station group and content matrix. Obtain the classification and author of the target site through the site-side publishing interface, and arrange tasks independently by site.
  • core

    One background management multi-site publishing

    Access PHP sites that support publishing interfacesSelect the target site when creating a taskTask list support for site filteringRemote classification and author information can be synchronized

  • recommend

    Common PHP CMS site-side support

    Z-BlogPHP can be directly connectedWordPress, Yunyou CMS site-side plugins can be contacted by the author for request.Other PHP programs can contact the author to add site-side plugins for free

  • safety

    The remote publishing interface has a validation boundary

    fui_ai_bridge receiverArticle publishing, classification, author interfaceAPI Key AuthenticationRemote sites can be enabled or disabled individually

  • performance

    Multi-site scheduling tasks independently

    Each site can configure an independent scheduled task URLSite parameters define the target siteCATE parameter limited target classificationMultiple site tasks can be processed in parallel

Use boundaries and environmental requirements

Explain the cost, audit responsibilities and operating environment in advance in order to stabilize the process for a long time
The plug-in is responsible for organizing the content production and publishing activities; model call costs, content review and site compliance still need to be responsible for the operator's own business.
  • recommend

    AI call costs need to be budgeted separately

    Plugin purchase fee does not include model API call feeDifferent models, networked search, and visual capability billing are differentIt is recommended to configure by content volume and publication frequency

  • safety

    The generated content should be manually reviewed

    Check for factual accuracyCheck business expressionCheck image copyrightCheck ad content and compliance requirements

  • operating environment

    Z-BlogPHP 1.7+PHP 7.0+MySQL 5.6+CURL extensionConfigure at least one available large model API Key

t like2
version number
1.2.7
work ID
fui_ai_writer
Author's official website
https://en.finchui.com/
procedural requirements
1.7
PHP version
7.0
package size
269.34 KB
  • 202609-30

    v1.2.7

    Fix the issue of repeatedly creating tasks by refreshing the page after adding tasks

    • repairFixed the problem that after adding a task, it still stays on the POST request page, and refreshing the browser will repeatedly submit the original data source and create the same task again; after the task is processed, it will be redirected to the task list instead.
  • 202609-24

    v1.2.6

    Fix gallery compatibility issues after object storage deletes local files

    • repairFixed an issue where the gallery misjudged remote attachments as unusable when deleting local copies after uploading the object store plug-in; gallery display, cache matching, and visual marking were changed to the final access address returned by the Z-BlogPHP attachment system.
  • 202609-11

    v1.2.5

    Optimize OpenAI calls, scheduled tasks, and template state management

    • changeOpenAI validation, article generation, and editor calls preferentially use the Responses API, and compatible lines that do not support this endpoint automatically fall back to Chat Completions.
    • repairRetain HTTP status codes, error messages, error types, and BigInt returned by upstream APIs to avoid misreporting HTTP 403 errors such as locale restrictions as insufficient model permissions.
    • optimizeWhen the plug-in is installed for the first time or the existing configuration is missing, the security check code and the scheduled task URL are automatically generated, and the task settings page adds manual triggering and execution prompts for all, this and remote site URLs.
    • safetyScheduled task security code regeneration is changed to POST requests with CSRF verification, and the backend configuration entry is restricted to access only when enabled in App Center or Finch Store according to packaging specifications.
    • repairFix an issue where prompting templates to be unenabled and reopened still shows enabled; disabling templates continues to remain in the template management list, but no longer appears in the template options for adding tasks and bulk modifications.
  • 202608-27

    v1.2.4

    Optimize multi-site API configuration and connection testing

    • changeThe website management cancels the default API address splicing and program type fallback, and uniformly uses the complete site-side API address filled in by the user; the site root URL is automatically extracted from the API address.
    • repairConnection testing uses the API address and key from the current form instead to avoid unsaved or modified configurations being tested at the old address.
  • 202608-14

    v1.2.3

    Online search changed to built-in model capability presets, and improved writing and image model presets

    • changeRemove the independent networked search model whitelist setting; the task page will automatically determine whether networked search can be enabled according to the built-in and verified model capabilities of the plug-in, and custom and unknown models are not supported by default.
    • changeUpdated default models and descriptions for Volcano Ark, Kimi, DeepSeek, Bailian, OpenAI, Grok, and MiMo; DeepSeek V4 remains on the official platform only, and Kimi adds K3.
    • changeOpenAI presets supplement GPT-5.6 Luna, Terra, Sol, GPT-5.5, GPT-5.4, GPT-5.4 mini, and GPT-5.4 nano; their networking capabilities are reserved for actual OpenAI or compatible interface verification and cannot be regarded as OpenAI official confirmation.
    • changeThe basic settings and model suggestions on the About page show only the current default presets; the Volcano Ark image model is changed to an independent configuration to avoid the image model from entering the article text model list.
    • repairBailian Network Search uniformly calls validated Qwen models through the Responses API; the default model migration only covers the full history built-in list and does not affect the manually maintained model list.
  • 202607-31

    v1.2.2

    Enhance multi-model structured output, task execution, and management entry security

    • repairFix the HTTP 400 issue caused by mismatch between thinking and temperature parameters when Kimi K2.5/K2.6 turned off thinking mode, and keep the compatibility behavior of old tasks inheriting global settings
    • repairEnhanced AI JSON output parsing to distinguish response truncation from formatting errors, and increased output budget for a bounded retry if structured output fails
    • repairFixed Grok thinking parameter mapping, different thinking mode timeout policies, and lost upstream HTTP error messages
    • repairFixed the problem that concurrent scheduled tasks may receive and publish the same task repeatedly, increased task collection lock and fatal error status recovery
    • changeUnify successful task status, article ID, publication time, and result data writing; only failed tasks are allowed to retry to avoid repeated generation of published articles
    • changeDebugging The original response is only saved in debug mode, and the resident task log no longer records body-related content such as URLs, titles, labels, and image keywords
    • repairEnable TLS certificate verification for editor AI and web crawl requests, and complete permissions and CSRF verification for log, task, gallery, template, site and API verification entries
    • changeUnify debug log directory and .log/.json file statistics and cleaning behavior, and strengthen independent log cleaning entry
    • changeCustom prompt automatically appends fact safety, prompt word injection protection and task type special project constraints at runtime, does not modify the template content saved by the user, and continues to use the original JSON return structure
    • changeAdapt output Token, temperature and reasoning parameters according to OpenAI and Grok model capabilities, and support model-level thinking mode uniformly for networked and non-networked requests
    • changeOpenAI GPT-5/o series and Grok 4 series preferentially use JSON Schema structured output, incompatible custom lines will automatically remove structured parameters and bounded retries
    • changeUpdate OpenAI with Grok default model presets, only affect new installations and restore defaults, do not overwrite existing user model list
    • changeAdapt enable_thinking, reasoning_effort and thinking parameters according to Bailian Qwen, DeepSeek, GLM, Kimi and MiniMax model families to avoid mixing different model parameters on the same platform
    • changeDeepSeek and Bailian prioritize JSON Object, bean bag Seed models that support structured output prioritize JSON Schema, and reserve automatic degradation when parameters are incompatible
    • repairRecognize that the model returns empty content and incorporate a bounded retry to reduce task failures caused by occasional empty responses from DeepSeek JSON Output
    • changeUpdate Volcano Ark, Bailian and MiMo default model presets, remove the MiMo V2 series that has been offline on June 30, 2026, only affect new installations and restore defaults
    • changeFollow Retry-After and retry at most once when the AI request encounters HTTP 429, 500, 502, 503, 504 to avoid instantaneous limited viewership or service fluctuations directly causing the task to fail
    • changeThinking about parameter incompatibility cache increase 30 days valid period, automatic re-detection model ability after expiration, compatible with subsequent platform upgrades
    • changeExport tokens according to model capacity limits, and apply security caps to GPT-4o and legacy bean bag models to avoid generating HTTP 400 beyond model limits.
    • repairFilter AI content for dangerous tags, HTML event attributes, srcdoc, and scripting protocol URLs before publishing, while preserving normal article layout and image content
    • changeThe security code of the planned task is changed to the cryptographic security random value generated by the random_bytes, the existing security code remains unchanged, and the new format is used when regenerating
    • changeThe first output budget of structured articles has been increased from 4096 to 8192 Tokens, and the minimum has been increased to 16384 Tokens when truncating, null responses or invalid JSON retries, and continues to be subject to the security upper limit of each model
  • 202607-22

    v1.2.1

    Scheduled task support Post articles by city sub-station

    • newSoft integration with FinchUI city sub-plugin: When the fui_citysubsite is enabled and the article City data field mapping is available, the "Publish Settings" of the scheduled task adds the city selection, and the new article on this site can be published to the designated city sub-station; if you select National, you will keep the national article. This setting is not displayed when data field mapping is not installed, not enabled or not completed, and the original writing and publishing process is not affected.
    • changeCity attribution only applies to new articles on this site; it is automatically cleared when switching remote sites or using article rewriting tasks, and the city field is not passed to the fui_ai_bridge universal remote publishing interface. Task import and export will retain the city code, but only when the current city substation environment is valid and the city is still enabled.
  • 202606-16

    v1.2.0

    Add visual marking, enhance map hit and optimize the writing and editor AI experience

    • newAdded visual marking capabilities: Support Volcano Ark, Kimi, OpenAI, Grok, and Alibaba Cloud Refining, and output Chinese tags and Alt descriptions
    • newNew visual marking switch, model selection and Prompt template editing are added to the map matching settings, allowing you to customize the identification strategy according to site needs
    • newLibrary management adds single AI marking and batch AI marking buttons to support supplementary marking of historical pictures
    • newA new "Batch Marking Concurrent Numbers" configuration item has been added to the drawing setting (default 3, adjustable 1 - 20), and batch AI marking of the gallery is changed to slicing parallel requests based on the concurrency number at the front end
    • newA "visual marking timeout (seconds)" configuration item (default 180) has been added to the drawing settings to uniformly control the maximum waiting time for a single visual model request
    • newThe rewrite task adds a "Keep the original tag" option, which forces the original tag to be used when rewriting to avoid tag drift
    • newRewriting tasks support batch selection of target articles by category. You can check multiple categories to automatically summarize article IDs and manually fill ID to remove duplication.
    • newReminds template management to add a "Restore Built-in Template" button, which can restore overwritten or mistakenly deleted built-in templates with one click
    • changeThe upload process supports automatic triggering of visual marking when no label is filled in, reducing manual maintenance costs
    • changeLocal cache hits are changed to a weighted matching strategy that combines article tags, task keywords and image keywords to improve matching accuracy
    • changeThe gallery batch AI marking button displays the progress in real time during the execution process (completed/total number and successful number)
    • changeThe default value of max_tokens for a single AI writing request has been increased from 2000 to 4096, making it difficult for long text scenes to be truncated.
    • changeThe article publishing process was changed to forcibly refresh the tag count, and the "Update Tag Count" switch was removed to avoid inaccurate tag cloud data caused by omissions
    • changeThe editor AI turns off the deep thinking mode by default and configures the value according to the non-thinking temperature, significantly improving the response speed of AI calls in the editor.
    • repairMake up the default configuration for visual marking, upgrade and migrate to avoid empty models or empty prompt words appearing after upgrading old sites
    • repairFixed an issue where the "Keep original label" check box was abnormally visible in the overwrite task form
  • 202605-26

    v1.1.9

    Enhance batch modification of tasks and add task import and export

    • newTask management adds exporting structured JSON based on current filtering results, and supports importing tasks to other sites to create new tasks
    • newTask import preview supports remapping prompt templates, models, publishing sites, categories, authors, advertisements, and rewriting task original article IDs
    • changeThe batch modification function is upgraded to batch modification of common task settings, and supports batch adjustments such as prompt templates, models, release sites, categories, authors, review status, and release time.
    • changeWhen a task is not checked manually, batch modification will be applied by default to all pending tasks that meet the current filtering conditions, and the quantity will be clearly prompted when confirming.
    • repairFixed the problem where the old default check logic only works on the currently paginated 50 tasks to avoid batch modification range being truncated by pagination
  • 202604-29

    v1.1.8

    Fix known issues

    • repairFix known issues
  • 202604-24

    v1.1.7

    Add refined temperature configuration and optimize upgrade compatibility logic

    • newAdd independent thinking/non-thinking temperature settings for each platform, supporting arbitrary decimal input and one-click restoration of default values
    • repairFixed temperature compatibility issues in kimi-k 2.6 and kimi-k 2.5 in interface verification, batch tasks and editor AI
    • repairFixed a problem where the Volcano Ark API Key may be automatically filled and changed by the browser when you click to verify and then save it
    • changeSynchronize the default model, default online search whitelist and default temperature preset, and adjust the default temperature for non-Kimi platforms to think 0.7 and non-think 0.4
    • changeOptimize the upgrade migration order and migration strategy to prevent old migrations from overriding new default values, and avoid overriding user-defined editor AI prompts during upgrade
  • 202604-23

    v1.1.6

    Enhance online search configuration and supplement news information tape template

    • newAdd a unified whitelist setting for online search models, clarifying that models that do not support online search will no longer be displayed and will not continue to participate in the saved whitelist
    • newTwo default prompt templates have been added: Today's newsletter template (online search)(with pictures) and Technology Industry Information template (online search)(with pictures)
    • changeToday's newsletter template and technology industry information template are optimized for official production use, and the "Online Search" logo is uniformly added
    • changeUpdate Alibaba Cloud Bailian default model and default online search model presets, while retaining default models such as deepseek-v3.2, glm-4.7, kimi-k2-thinking
    • changeThe classification display of this station in the task management list is optimized to "Classification Name (Classification ID)" to facilitate verification of classification ownership
    • repairOnline search tasks extend API timeout times and optimize timeout prompts according to scenarios, reducing 120-second request timeout errors
    • repairCompatible with scenarios where Pagoda URL-GET scheduled tasks do not carry User-Agent, preventing scheduled tasks from reporting errors in advance during the loading phase of third-party plug-ins
  • 202604-18

    v1.1.5

    Added Grok / MiMo, multiple free gallery sources and model batch management

    • newAdded support for xAI Grok and Xiaomi MiMo large models, covering basic settings, task execution, editor AI and default model presets
    • newThe task management page has added the ability to modify task models in batches to facilitate the unified migration of providers and models used by historical tasks
    • newNew Pixabay image sources and Openverse public copyright image sources have been added to the gallery, supplementing the more accessible free image sources
    • repairFixed a problem where manually uploaded Chinese tagged pictures hit unstable when publishing tasks with pictures. Local gallery matching was changed to integrate article tags, AI image keywords and original task keywords
    • changeChange multi-platform model configuration and task model selection to Tab switching to reduce the height of background pages
    • changeSupplement the default model presets and restore default buttons for each platform, and update the list of recommended models for Alibaba Cloud Bailian
  • 202604-02

    v1.1.4

    Optimize task management stability and unify the background paging experience

    • repairFixed the problem that the task still prompted the addition success when the database write failed, and the new result was changed to judge based on the actual write status
    • repairFixed a performance hazard caused by large data volume query in total task list statistics, and changed it to database COUNT statistics
    • changeOptimize the logic of batch deletion and clearing tasks, and change it to batch deletion of database to improve stability when processing a large number of tasks
    • changeAdd the existence check of the task table before adding a task to avoid the task table being missing in an abnormal state and the inability to continue writing
    • changeUnify paging rules for scheduled tasks, advertisements, templates, and website management pages, and display up to 10 digital page numbers
    • changeAdd home page, previous page, next page, and last page buttons for background paging, and add paging disabled style
    • changeThe website management list supplements the real paging ability, and it is no longer fixed to read only the first 100 records
  • 202603-23

    v1.1.3

    Added multi-site management function, fixed 502 error in planned tasks, and improved task execution stability

    • newNew website management features: Support for adding, editing, and deleting remote sites, and automatically synchronize remote site categories and author lists
    • newAdd fui_ai_bridge receiver plug-in: Remote sites provide API interfaces such as article publishing, classified list, and author list after installation
    • newWhen adding tasks, you can choose to publish them to this site or remote site, and the category and author list are automatically switched
    • newThe task list supports filtering by site, and batch retry and clearing operations support site-wide filtering
    • newThe scheduled task URL supports site parameters, and scheduled tasks can be configured separately for each site to achieve parallel processing
    • newThe scheduled task URL supports the cate parameter, which can limit processing to only tasks of the specified category
    • newThe scheduled task setting page displays a dedicated URL for each site and supports one-click copying
    • repairFixed a 502 error caused by a request timeout when a scheduled task generated multiple articles
    • changePlanning tasks now return an HTTP response and close the connection immediately after verification. The tasks continue to be processed in the background, completely eliminating the 502 caused by the timeout of reverse proxy (Nginx, etc.)
    • changeSupport in PHP-FPM environment to completely close HTTP requests through fastcgi_finish_request() and continue execution
    • changeAdd set_time_limit(0) and ignore_user_abort(true) to prevent PHP timeout or client disconnection from causing task interruption
    • changeReset the PHP execution time limit (300 seconds) before each task execution to prevent long-lasting tasks from being forcibly terminated
    • changeAdded monitoring of the total execution time of batch tasks (upper limit of 600 seconds), and automatically stopped remaining tasks when timeout
    • changeRemove unnecessary sleep(2) delays between tasks and reduce the total time spent planning tasks
    • changeNew type verification is added to batch parameters, which is limited to the range of 1-10 to prevent abnormal input
  • 202603-09

    v1.1.2

    New advertising management features

    • newAdd a complete advertising management module: support four advertising types: text, custom code, image, and mixed
    • newAds can be associated when adding tasks, and AI automatically inserts ad content after generating articles
    • newAd insertion is supported in both new article and article rewrite task modes
    • newSupport three insertion positions: beginning of text, end of text, beginning of text + end of text
    • newSupport setting of advertising expiration time (y-month-yy
    • newAdded expired ads cleaning portal (cron_clean_ads.php), which can be automatically executed every day in conjunction with pagoda panel
    • newSupport batch removal or batch replacement of advertising content from articles
    • newThe advertising management page provides statistical cards (total/enabled/used/expired)
    • changeAd HTML adopts anti-interception naming conventions to avoid being misrecognized by browser ad interception plug-ins
  • 202602-15

    v1.1.1

    Added editor AI assistant function

    • newIntegrate FinchUI wangEditor AI functions
    • newProvides 12 AI-assisted writing functions (rewriting, error correction, continuation, summary, etc.)
    • newSupport customization of prompt word templates for each AI function
    • newAdd an editor settings tab to centrally manage editor AI configuration
    • changeIntelligently preserve pictures and HTML tags in articles
  • 202601-20

    v1.0.6

    Added library management, automatic photo matching and article rewriting functions

    • newNew gallery management system: Support Unsplash, Pexels free gallery search and Volcano Ark AI production pictures
    • newIntelligent matching of pictures based on article tags and AI keywords to automatically insert article content
    • newGallery management interface: supports AJAX batch upload, label editing, filtering, and batch deletion
    • newNew article rewriting task function: Support two task types: new article and article rewriting
    • newRewriting tasks support single ID, multiple IDs, continuous ranges (such as 100-200), and mixed format inputs
    • newAdd two template types: article rewrite and article rewrite (with pictures) and default template
    • changeAI deeply rewrites and retains all pictures of the original text, leaving the original text classification/author/release settings unchanged
    • changePerfectly compatible with the fui_oss object storage plug-in (via the Z-BlogPHP standard attachment system)
    • changeNew task type filtering and color label display in the task list
  • 202512-10

    v1.0.5

    Added Alibaba Cloud Refining and OpenAI support

    • newAdded Alibaba Cloud Refining AI model support (using DashScope compatible interface)
    • newAdded OpenAI model support (support for custom API Base URLs)
    • changeOptimize the recommended list of models and focus on writing scenarios
    • changeImprove multi-platform model configuration and verification functions
  • 202511-20

    v1.0.4

    Optimize plug-in configuration page

    • changeOptimize the prompt text and button styles on the plug-in configuration page
    • changeAdd model suggestions
  • 202510-15

    v1.0.3

    Added multi-platform API support

    • newIncrease support for DeepSeek's official website API
    • newIncrease support for Kimi's official website API
  • 202509-20

    v1.0.2

    Add summary function and improve multi-platform support

    • newAdd AI to generate article summaries to improve SEO effectiveness
    • changeImprove multi-AI platform support
  • 202508-09

    v1.0.1

    original version was

    • newSupport three creation modes: keywords, titles, and website
    • newComplete task management system
    • newVolcano Ark AI model integration
    • newPrompt template system
    • newAutomation of scheduled tasks
    • newAutomatic crawling of website content
    • newAI automatic classification function
    • newURL field saving function

guess you like

hot commodity
hot Tags
related tags
FinchUI Store Edition

Share

Copy Link

Xinglan StudioAdd customer service qq.

Working hours: 9:00-22:00
Saturday and Sunday: 14:00-22:00
wechat
Scan the code to add customer service WeChat