changelog

2026-05-21

(v0.2026.05.21.02)

• None.

• Extracted browser final-video upload into a shared service with progress labels for preparation, upload, final save, and server fallback upload.
• Added file-size-based timeout handling for large MP4 direct uploads to reduce long-video upload failures.

• Fixed large production MP4 uploads continuing into API fallback after direct upload failure, which could cause final-video uploads to fail again.
• Added regression coverage for browser upload timeouts, upload progress, and large-file fallback policy.

2026-05-21

(v0.2026.05.21.01)

• Added an R2 public asset sync workflow with a versioned manifest for static site assets.
• Added Moonshot Kimi K2.6 as the default text model for drama and text workflows.

• Moved logos, favicons, social previews, blog images, payment icons, avatars, and feature media to versioned R2 URLs.
• Updated content, navigation, SEO metadata, and MDX image handling to use remote asset URLs consistently.
• Increased image generation route duration limits for slower premium image models.
• Improved the drama creator hero spacing on landing pages.

• Fixed social preview image metadata when the configured preview image is an absolute remote URL.
• Improved image API errors for non-JSON upstream responses by including HTTP status and response preview.
• Added regression coverage for R2 asset URL replacement, Kimi K2.6 defaults, and non-JSON image API failures.

2026-05-20

(v0.2026.05.20.01)

• Added final preview playback in the post-production stage using director videos in sequence.
• Added browser-side final MP4 export with upload and persistence as a project final-video asset.

• Added a direct video upload flow with signed upload URLs, completion confirmation, and API upload fallback.
• Improved large video upload timeouts and storage response handling to reduce final MP4 upload failures.
• Updated video model capability detection for Seedance 2.0, start/end frames, audio support, and more accurate duration options.

• Fixed the Seedance 2.0 page resolving to the wrong model alias.
• Fixed inaccurate reference-image limits and audio capability labels for some video models.

2026-05-12

(v0.2026.05.12.01)

• Planned the drama post-production stage for assembling director shot videos into a final MP4.

• Added an implementation plan covering stage modeling, project persistence, composition APIs, UI, agent runtime integration, and verification.
• Clarified the MVP scope: preserve source audio, allow silent segments, ship concat-only assembly first, then add BGM, trim, skip, and reorder controls.

• None.

2026-05-05

(v0.2026.05.05.01)

• Added a Codex pet ZIP upload endpoint that returns a downloadable public package URL after generation.
• Added Codex Pet Generator to the English and Chinese use-case navigation menus for easier access.

• Replaced long base64 install scripts with short ZIP download commands, making Windows and macOS / Linux installation more reliable.
• Added bilingual post-install guidance for refreshing Custom Pets, selecting the new pet, and showing the Codex Avatar.
• Updated spritesheet composition to write atlas cells frame by frame based on each state frame count, with image smoothing disabled for sharper pixel output.

• Added regression coverage for package uploads, install commands, navigation order, and atlas composition, including signed-out, invalid WebP, and unavailable storage cases.

2026-05-04

(v0.2026.05.04.01)

• Added a bilingual Codex Pet Generator SEO tool page for creating Codex custom pet files from prompts and optional reference images.
• Added a dedicated Codex pet generation API that uses signed-in user credits and generates the base image plus nine animation state rows one image at a time.
• Added browser-side spritesheet.webp composition, pet.json generation, ZIP download, and copy-paste install scripts for Windows, macOS, and Linux.

• Added shared Codex pet helpers for atlas rules, animation states, manifest output, prompts, and install script generation using the 8x9, 192x208 cell contract.
• Added state previews, generation progress, model credit display, and reference image upload support so users can inspect the pet before installing it.

• Fixed Codex pet generation failing when an image model does not support transparent backgrounds by avoiding unsupported transparent parameters and adding a fallback retry path with regression coverage.

2026-04-27

(v0.2026.04.27.01)

• Added a bilingual GPT Image 2 image generator landing page with text-to-image, image editing, multi-reference fusion, high-resolution workflow copy, and dedicated SEO content.

• Moved GPT Image 2 to the first position in the AI Image menu and added an animated New badge for better discovery.
• Added sitemap coverage, locale message registration, and fallback model support for the new GPT Image 2 page.
• Added product marketing context documentation to keep future positioning and landing-page copy aligned.

• Added regression coverage for GPT Image 2 page messages, navigation order, and animated navigation badge rendering.

2026-04-16

(v0.2026.04.16.01)

• Added five new English blog posts covering AI animation basics, beginner tool comparisons, AI animation workflows, AI anime singing characters, and whether AI will replace animators, along with matching AI-generated illustrations.

• Expanded the blog batch with stronger conversion-oriented internal links and complete image assets for each new article.
• Added a shared duration parsing utility plus regression tests so drama workflows can consistently understand values like 3秒, 3s, and 4.2 seconds.

• Fixed inconsistent shot-duration parsing across director video generation, workspace totals, exports, and storyboard schema normalization, reducing estimation, billing, and display mismatches.

2026-04-15

(v0.2026.04.15.01)

• Added a new bilingual Wan 2.7 AI image generator landing page with standard / pro model switching, gallery content, and dedicated SEO copy.
• Added four new English blog posts covering AI animation workflows, full anime episode production with AI, selfie-to-animal persona ideas, and animal-to-human image transformation.

• Added Wan 2.7 AI to the homepage AI Image navigation and moved it to the first position, with sitemap coverage included.
• Updated pricing copy so plan examples now describe script generation as 1 credit = 1 script, making credit expectations easier to understand.

• Fixed mismatched example anchors on the AI Video Generator and Seedance 2.0 pages, and removed the duplicated video gallery section to avoid broken jumps and repeated content.

2026-04-14

(v0.2026.04.14.02)

• Added an AIBlog.Tools partner badge to the homepage footer.
• The site now remembers the user's last selected locale and redirects unprefixed visits to that language when applicable.

• Enabled NEXT_PUBLIC_LOCALE_DETECT_ENABLED by default and documented it in the example environment file so fresh deployments can turn on locale detection out of the box.
• Extended the locale cookie lifetime so language preference persists more reliably across sessions.

• Fixed public-page caching behavior removing pending locale Set-Cookie headers too early, which could prevent language preference from being saved after a locale switch.

2026-04-14

(v0.2026.04.14.01)

• Added six new English blog posts covering AI animal image and talking-animal workflows, AI animation vs traditional animation, AI sprite animation, anime-to-realistic image conversion, AI photo animation tools, and anime AI art prompts.

• Improved blog MDX rendering so internal and external links are styled more clearly, and statically imported article images can reuse their original dimensions.
• Refined the drama screenwriter stage view with more stable character-age formatting and clearer character and episode cards.
• Added a pinned-post rule for blog listings so the changelog stays at the top of the post feed.

• Fixed MDX images losing static import metadata, which could cause article images to render with fallback dimensions.
• Fixed the screenwriter stage view duplicating the Chinese age suffix when age values already included 岁.
• Fixed blog sorting so newly published posts no longer push the changelog out of the top position.

2026-04-10

(v0.2026.04.10.02)

• Storyboard character validation now reads character mentions from screenwriter prose and canonicalizes punctuation drift such as · versus . into the normalized character name.

• Sign-in, sign-up, and verify-email pages now share one auth layout shell for more stable centering, width, and text wrapping on narrow screens.
• Removed transitional helper copy from drama workspace stage pages to keep the stage headers and review views cleaner.
• Moved custom theme spacing tokens under the --wire-spacing-* namespace so Tailwind spacing tokens are no longer overridden.

• Fixed storyboard character checks incorrectly flagging names as unknown when they only appeared in prose or used punctuation variants, and aligned auto-fill with the canonical name.

(v0.2026.04.10.01)

• Added a new AI Character Generator landing page (EN/ZH) with dedicated prompts, gallery content, CTA flow, and sitemap coverage.
• Added a page-style drama workspace mode that lets users switch between page view and the existing canvas view, with focused panels for screenwriter, storyboard, image designer, artist, and director stages.
• Expanded short-drama genre templates with 12 new directions, including cyber cultivation, esports redemption, interstellar son-in-law, and anti-involution workplace arcs.

• Refreshed the AI Anime Generator page with richer SEO copy, new “What is / How to” sections, stronger internal linking, and localized generator messaging.
• Deferred analytics, customer service, global overlays, and root utilities to reduce first-load overhead, while hardening same-origin auth base URL handling.
• Unified generator prompt/style assembly, theme loader registries, and static icon registries so dynamic pages and generator blocks are more reusable.

• Fixed unstable drama screenwriter/storyboard validation by detecting missing episodes, scenes, and character data more accurately, and by repairing certain malformed array/object JSON responses from the model.
• Fixed recovery paths for missing storyboard character names and image-designer prompts so existing project data can be reused inside the workspace and canvas flows.
• Fixed canvas snapshot save/load behavior when no storage provider is configured, avoiding unnecessary failures in environments without R2 storage.

2026-03-15

(v0.2026.03.15.01)

• Expanded the Loli image and video style picker with a much larger curated style library, adding more previewable looks across anime, illustration, western, game/3D, cute, eastern, and Korean categories.
• Added style-library tooling to mirror preview assets to R2 and generate runtime JSON payloads from snapshot data.

• Unified style-tab construction for the Loli image and video generators so legacy presets and the expanded style library are assembled from one shared source.
• Upgraded style search to match localized names, group labels, tags, and keyword metadata for faster discovery inside the style modal.

• Fixed lazy-loaded style preview wrappers so thumbnails fill their cards consistently in the modal and selected-style preview surfaces.
• Restored image title passthrough in the shared LazyImage wrapper so preview metadata is preserved.

2026-03-14

(v0.2026.03.14.01)

• Added free-credit progress bars to the Loli image and video generators, showing remaining new-user credits directly under Generate.
• Added a searchable style gallery modal to the Loli image and video generators, with richer preview coverage across base, US, JP, and game looks.
• Added model-aware video controls for Pollinations models, including Veo start/end frames, per-model duration options, and optional video sound when supported.

• Unified user credit payloads so /api/user/get-user-info and /api/user/get-user-credits both return total remaining credits plus free-credit summaries.
• Enriched Pollinations video model metadata with text/image mode support, reference-image limits, reference layout, duration options, and audio mode to drive the frontend consistently.
• Replaced the static public/sitemap.xml with a dynamic Next.js sitemap that merges localized routes and blog entries.

• Fixed Pollinations video generation requests hitting the wrong upstream path by switching backend requests from /image/... to /video/....
• Fixed model-specific video validation so unsupported reference images and out-of-range durations are rejected or normalized before submission.
• Fixed generator guidance and validation around reference uploads so Veo-style start/end frame flows match actual model capabilities.

2026-03-09

(v0.2026.03.09.01)

• Added an enlarged preview dialog to the Loli image generator with zoom, pan, and quick history-image preview.
• Added a Prompt Enhance toggle to both image and video generators, enabled by default and persisted across visits.

• Simplified the prompt action area in both generators by removing the inactive Optimize button and keeping the prompt workflow cleaner.
• Wired the prompt-enhance setting through the full request path so image and video generation follow the current toggle state.

• Fixed image and video generation APIs not honoring the user-facing prompt-enhance setting, so generation parameters now match the UI.

2026-03-04

(v0.2026.03.04.01)

• Added a new Nano Banana 2 AI image generator landing page (EN/ZH), plus an alias redirect from /ai-image-generator/nanobanana-2 to /ai-image-generator/nanobanana2.
• Added Nano Banana 2 entries to landing navigation and sitemap coverage.
• Added preset prompts and one-click shuffle fill in the Loli video generator.

• Upgraded director video billing to per-second charging: estimates, approvals, and deductions now use credits per second × total duration, with billed duration and per-second credits returned in responses.
• Switched Pollinations default video model to grok-video and expanded fallback model coverage (including grok-video and seedream5).
• Standardized video generation on 1080p: frontend resolution is locked to 1080p, and backend maps width/height (1920×1080 / 1080×1920) for provider requests.

• Fixed incorrect Seedream 5 alias mapping (seedream5 -> seedream5 instead of nanobanana-pro).
• Fixed mismatch between displayed and charged director video credits by estimating credits from shot durations.
• Fixed inconsistent duration normalization between single-shot and batch video generation flows.

2026-03-02

(v0.2026.03.02.01)

• Agent runs now support inline renaming in the workbench, with backend support to persist config.name.
• Agent timeline now includes persisted user.message events (user input, approval/reject notes, pause reasons) mixed with assistant/status/tool events by timestamp.
• Added one-click stage artifact export from the output tree, including structured files for script/storyboard/images/videos.

• Improved Pollinations text generation flow: auto-select streaming for high token limits, normalize streaming deltas, and parse content_blocks consistently.
• Improved video export robustness by carrying cloud/original/source URL candidates and falling back across sources during packaging.
• Refreshed EN/ZH landing page media assets by replacing placeholder Picsum images with hosted production assets.

• Fixed duplicate/overlapped text output in Pollinations streaming responses.
• Fixed non-retryable Pollinations 400 cases (max_tokens > 4096 requires stream=true) to fail fast instead of looping retries.
• Fixed long auto-start creativity prompts being truncated by URL length limits (now transferred via session storage).

2026-02-26

(v0.2026.02.26.01)

• Added a unified PromptInput input component (autosize, Enter submit, action tooltip) and integrated it into the Agent console input area.
• Added an AI138 partner badge link in the homepage footer.

• Added summary=1 lightweight mode to /api/agent/run so run lists fetch only required fields with lower payload cost.
• Switched the Agent creation panel to lazy-load text model options, and refined selector widths plus input-area styling.
• Added auto-scroll to the latest conversation message, and stabilized run-list refresh/selection behavior.
• Added internal documentation for Agent conversation and event flow.

• Fixed legacy run config compatibility: records with only model are normalized into contentModel/controlModel.
• Fixed input action conflicts during active runs by unifying primary action button logic and reducing invalid-trigger cases.

2026-02-23

(v0.2026.02.23.01)

• Added an Agent workbench (/agent, /agent/runs, /agent/runs/[runId]) with run creation, auto/single-step execution, checkpoint approval, cancel/continue, event timeline, and artifact tree preview.
• Added Agent backend routes (/api/agent/run, /step, /approve, /cancel, /events) to support tool-calling across stages 1-5.
• Added new Agent data models and migrations: drama_agent_run and drama_agent_run_event, including persisted run state, locking, and events.
• Added a public /anime-ai-agent landing page (EN/ZH) and updated sitemap coverage.

• Moved project quota checks into the service layer, with unified PROJECT_LIMIT_REACHED responses in project/image/video project creation paths.
• Upgraded project deletion to "hide first, cleanup async": storage objects and canvas snapshots are cleaned up before hard delete.
• Added Agent settings: drama_agent_control_model and drama_cost_agent_control, and wired them into the credit system.
• Raised local dev minimum DB pool capacity (to reduce auth/session queue stalls) and increased max duration for Agent step/video routes.

• Fixed multiple SSR/Hydration issues (media query hooks, theme/toast mounting, sidebar skeleton flicker).
• Added scan limits for video history lookup to avoid overly deep scans on large datasets.
• Updated director canvas preview sizing to respect video aspect ratio and reduce preview distortion.
• Fixed duplicated chat text when parsing content_blocks.

2026-02-13

(v0.2026.02.13.08.20)

• Added a global Pricing modal (responsive dialog/drawer) and a TopBanner CTA action to open it.
• Added a weekly flash deal countdown and discount display in the pricing section.

• Updated default Drama/Pollinations text model to kimi and refreshed the Pollinations fallback list.
• Removed the outdated public/robots.txt (now relying on the dynamic Next.js robots route).
• Simplified the Seedream 5 landing page rendering by removing duplicate FAQ JSON-LD injection.

• Fixed Drama chat model resolution priority and ensured logs/metadata record the effective model/provider.

2026-02-11

(v0.2026.02.11.10.28)

• Added landing pages and blog posts for new AI models: Qwen Image 2, Seedream 5, and Seedance 2.
• Added automatic storyboard character auto-fill plus quick-edit metadata for missing mentions.
• Added a new Seedance 2 animated icon.

• Image/video generators now support default models and alias mappings for cleaner model selection.
• Updated AI image generator landing copy, CTAs, and sitemap coverage for new pages.
• Added shared OpenGraph/Twitter metadata for landing pages and improved centered hero layout.

• Unified storyboard character parsing to avoid false empty/missing character states.
• Resolved model aliases correctly in generation requests to prevent mismatched backend models.

2026-02-07

(v0.2026.02.07.19.23)

• Added Drama text-model selection sourced from Pollinations, including display names and NEW badges.
• Added automated image retention cleanup with a cron endpoint and configurable retention days.

• Pollinations model list now surfaces model titles/descriptions and defaults to Claude Large for text.
• Switched default image/video aspect ratios to 9:16 with updated ratio presets.
• Improved dashboard navigation and table action dropdown behavior for active states and hydration.

• Refined drama chat retry caps for no-JSON responses.
• Removed JSON code fences from screenwriter prompt to avoid formatting artifacts.
• Defaulted Pollinations video aspect ratio when missing.

2026-02-01

(v0.2026.02.01.19.05)

• Storyboard shots now keep a clear on-screen character list (empty shots supported).
• Added a “Missing characters” hint with quick edit in the storyboard table.

• Storyboard rendering and exports now prioritize the character list for consistency.

• Prevented saving characters that are not in the screenwriter list.

2026-01-31

(v0.2026.01.31.09.16)

• Added a video generation page.
• Updated Drama AI workflow and chat route for new capabilities.
• Added model selection for all five stages (writing, storyboard, image design, art, directing) with different credit costs.

• Refreshed landing pages and localization content.
• Optimized the pricing section design on the homepage.
• Added a Vercel production env file for smoother deployment.

• Fixed build-time type errors.
• Fixed checkout price lookup issues.
• Fixed an edge case when cancelling chat snapshots.