Upcoming: gallery and recovery fixes
In preparation: keep gallery results and drafts when correcting filters or switching language; match repeat critiques to language, category and accessible history; recover interrupted critique and image requests without duplicate quota use; preserve pagination and ignore stale results after filter or account changes; hide private content after identity changes; improve downloads, mobile error pages and account recovery controls. Same-account session renewal keeps uploads and drafts and continues the requested action; signing out or changing accounts still clears private state. Account pages also keep their initial loading state consistent while restoring a session.
Upcoming: subscription recovery and image retry fixes
In preparation: keep subscription management available after payment issues, prevent out-of-order payment updates from blocking restored Pro access, and show an error when a retried image request has changed instead of returning an older task.
Critique and draft fixes
Critique retries no longer charge twice, language changes preserve image prompt drafts, sign-in and sign-up follow the site language, and retake checklist fields stay separate.
Bug fixes
Fixed several bugs to improve stability.
Gallery browsing is more direct
The Gallery no longer displays fixed featured critique cards. Browse public photos and open their full critiques directly.
The message center is available
Sign in and open the header bell or More → Message center to see task and account messages, filter unread items, mark them read, or archive them. System messages stay on; Gallery likes and product announcements are off by default and can be enabled in notification settings. Single-photo score feedback remains private, can be changed or withdrawn, and does not directly change scores.
Read a critique and choose the next adjustment
The English homepage explains what AI can observe and what to verify against your intent. The Gallery adds three current public critique examples with suggested actions, and daily-practice readers can continue with the shooting checklist and retake comparison guides. These actions are suggestions, not verified retake results.
Give feedback on scores; in-app messages are in preparation
Signed-in users can now privately mark a single-photo score as accurate, too high or too low, then change or withdraw their judgment. Feedback does not directly change scores. The in-app message center is not enabled yet and needs no setup now.
Quick Flash critique, stronger Pro analysis
Flash uses GPT-6 Sol for quick photo critique. Pro uses the stronger GPT-6.1 Sol for deeper analysis and improvement advice. Photo scoring, detailed feedback, and original/retake comparisons follow your chosen mode. Saved critiques keep their original model information, and progress statistics compare matching scoring profiles.
Controlled photo access
Photos and generated images use controlled, time-limited access to reduce exposure of direct storage links. Public Gallery browsing remains available. This change does not delete photos or alter saved critiques.
15-day Free critique history
Free critique history access changes from 30 to 15 days after each critique is created, including existing history. This change does not automatically delete original photos. Public works and share links keep their existing access rules, and Pro keeps permanent history. Save photos and critiques for longer-term reference.
More reliable critiques and photo browsing
This maintenance update is now live: quota accounting handles concurrent critiques, duplicate task processing is prevented, and upload confirmation checks the actual image. History pagination, photo recovery, and gallery scoreboard loading have also been improved.
Scoring refined, v5 gallery critiques reassessed for free
Scoring focuses more on the finished image: monochrome, muted color, and expressive shadows do not automatically lose points. v5 gallery critiques were previously reassessed for free; this time, 6 landscape photos have also had their scores and feedback updated under the revised rubric, with no critique quota charged and private historical critiques kept unchanged.
v5 gallery critiques reassessed for free
Scores and feedback have been updated with GPT-6 at no cost to your critique quota. Visit the gallery to see the results.
Photo critique upgraded to GPT-6
New workspace submissions now use GPT-6 for single-photo scoring, detailed critique, and original-versus-retake comparison. Saved historical results stay unchanged.
Photo scoring update, saved history preserved
New critiques use updated scoring ability, while saved critiques and gallery scores stay unchanged. Progress comparisons only use records scored the same way.
Workspace critiques use GPT-5.6
The workspace uses GPT-5.6 by default, with no model choice needed after upload. Access rules for saved critiques stay the same.
Photos first in the gallery, clearer critiques
Public photos now appear sooner in the gallery, with clearer critique evidence and controls, plus a retry when loading fails.
Easier gallery publishing and clearer controls
Find gallery publishing beside your critique, understand sharing status at a glance, and use clearer upload and generation controls in English, Chinese, and Japanese.
Clearer history, sharing, exports, and practice
Critiques and history are easier to browse, shared links open a photo and critique page, and exports keep the full text. Practice now connects an original and a saved goal to each comparison result.
Coming soon: organizing, retakes, and sharing after critique
Uploading for critique stays first; saved goals, retake comparisons, private tags and notes, practice logs, gallery browsing, share revocation, public profiles, image cards, and print summaries are being prepared.
A more consistent multilingual experience
Your language preference now carries through pages, errors, prompt examples, generation, and checkout without unexpected language switches.
Clearer high-score judgments and progress comparisons
High scores now get a closer check against the image, and progress comparisons use the same scoring standard; saved ratings and gallery scores stay unchanged.
Free gallery re-evaluation is complete
All 64 gallery works have been re-evaluated with the updated critique standard. Scores and feedback are refreshed without using account critique quota. Work links, likes, and gallery dates are unchanged.
Critiques that understand your photograph
Critiques respond to each photograph’s style and expression with more specific, consistent feedback. Free gallery re-evaluation will follow end-to-end validation.
More consistent reviews, credit reservations, and safer generation jobs
PicSpeak now uses one official GPT model for numeric scoring while keeping Qwen or GPT selectable for written critique. This release also strengthens public-result privacy, generation credit reservations, worker concurrency, storage compensation, and analytics attribution.
Photo critique and retake analysis upgraded
This release upgraded single-photo and retake analysis to the GPT-5.6 model used at the time. Saved results retain their actual model information. Today Flash uses GPT-6 Sol for quick critique, while Pro uses the stronger GPT-6.1 Sol for deeper analysis.
SEO/GEO authority and discovery hardening
PicSpeak completed separate traditional SEO and GEO audits and closed local crawl, SERP presentation, schema, and trust gaps. Invalid URLs return real 404s; titles, descriptions, H1s, Retake indexing, internal links, and caching are tightened; page entities and AI-discovery files expose stronger provenance.
Blog locale routing and cache hardening
PicSpeak now resolves unprefixed Lens Notes links with a private redirect to one stable locale-prefixed URL. The change removes shared-cache ambiguity, rejects forged locale headers, keeps social previews in the target language, and separates editorial curation from article modification dates.
Lens Notes localization and update visibility
PicSpeak now keeps Lens Notes aligned with the selected language, including fully localized Chinese and Japanese page chrome. The two refreshed guides about shoot-review-retake comparison and actionable feedback checklists appear first and expose their modification date.
Photography-coach frontend and core workflow redesign
PicSpeak now presents one coherent professional photography coach and efficient AI tool experience. Home shows a real critique artifact, Workspace, Review, Retake, and AI Create form a continuous practice loop, and supporting pages, responsive behavior, accessibility, and attribution follow the same system.
GPT-5.6 Terra Retake Coach and model-specific critique
PicSpeak now turns a single-photo critique into a measurable original-versus-retake loop. GPT-5.6 Terra compares both images in one request while the server calculates five-dimensional progress and saves evidence plus next-shoot actions; normal critique now exposes an explicit Qwen 3.5 / GPT-5.5 choice.
Fixed usage nav in the header and mobile tabs
PicSpeak now treats the Usage link as public header navigation in the app shell instead of hiding it behind authenticated state. Mobile tabs keep the Usage button too, and medium-width header spacing was tightened so the restored entry stays visible.
Frontend extraction, shared checkout, and JSON-LD hardening
PicSpeak now extracts repeated UI and utility logic from the generation page, account usage, review history, Gallery, and home page into reusable components and helpers. Credit-pack checkout, date filters, image-type copy, and locale date formatting are shared, and public JSON-LD scripts now use safe serialization.
Generate SEO fallback, news sitemap, and structured data
PicSpeak's /generate page now keeps its full interactive experience while exposing server-rendered template, quality, and size copy through an SEO fallback. robots.txt also advertises image and news sitemaps while protecting private paths, and Gallery, Affiliate, and Prompt Example structured data now live behind tested helpers.
AI Markdown mirrors, llms.txt, and WebSite JSON-LD
PicSpeak now exposes /ai-content/*.md Markdown mirrors for the home page, Lens Notes, Prompt Library, Gallery, and Updates so AI crawlers can cite concise public summaries. llms.txt lists the same mirrors, and WebSite JSON-LD now includes search and Updates subscription actions.
Web Vitals reporting, canonical redirects, and Blog speakable data
PicSpeak now sends Core Web Vitals and browser performance metrics into product analytics, aligns production canonical redirects on https://www.picspeak.art, and marks BlogPosting JSON-LD with speakable headline and intro selectors.
IndexNow, image sitemap, and author entity page
PicSpeak now exposes an image sitemap, an Asa Zhou author entity page, and a post-deployment IndexNow submission path. The blog index also renders static article content on the server while keeping view counts as a small client island.
Structured data, demo review SEO, and Updates metadata
PicSpeak public content pages now expose stronger machine-readable signals: blog posts add breadcrumb JSON-LD, public demo reviews add Review / CreativeWork / Rating JSON-LD, and the default /updates page owns page-level metadata with shared hreflang alternates.
Blog-specific OG images and public page caching
PicSpeak blog posts now generate 1200x630 social images from each article title and category. Public home, blog, updates, Prompt Library, legal, and affiliate pages also get CDN cache headers, while workspace is removed from the public sitemap.
Homepage H1 SSR fallback
PicSpeak now renders the homepage H1, description, and three core feature cards in server HTML, so search engines and AI crawlers can read the core product meaning without waiting for client hydration.
Security hardening, CI gates, and accessibility polish
PicSpeak now enforces scoped upload and photo-proxy tokens, atomic rate limits, safer request-audit boundaries, pooled outbound HTTP calls, and new database indexes. The frontend also improves generated-image rendering, SEO defaults, keyboard menus, modal focus behavior, and GitHub Actions CI.
Privacy and terms pages, plus static asset recovery
PicSpeak now has public Privacy and Terms pages linked from the footer, and the local production preview no longer lets Clerk intercept Next static CSS and JavaScript assets. The homepage loads with styles again on localhost.
Prompt Library expansion, example retakes, and operational health
The AI Create prompt library now has 50 examples, detail pages can carry an example target into the workspace for retake practice, the trilingual Blog adds a prompt workflow article, and the backend now exports a daily operational health snapshot for tasks, costs, payments, and public-content checks.
AI Create example handoff, credit recovery, and analytics splits
Prompt Library examples can now carry their prompt and recommended settings into AI Create. When credits run out, users can start a credit pack or Pro checkout in context, return to the original task after payment, and analytics now breaks down example usage, reuse, credit recovery, and generation unit economics.
Analytics weekly report, retake practice loop, and in-task reading
PicSpeak now connects critique suggestions back to the workspace as concrete next-shoot targets, lets users read full Lens Notes articles inside review and generation waiting pages, and expands product analytics with AI Create, Prompt Library, retake, locale, and data-health reporting.
Full SEO pass, real product OG image, and crawlable prompt examples
Completed a full PicSpeak SEO / GEO audit and aligned AI critique, AI Create, and the public gallery across metadata, sitemap, llms.txt, a real product Open Graph image, and crawlable GPT Image 2 prompt example pages.
Runtime migrations, API stability, and homepage language cleanup
Runtime schema management now runs through Alembic baseline and incremental migrations. Guest creation, audit logging, blog view counts, gallery recommendations, and AI generation downloads gained stability guards, while the frontend now has tighter API URL, default locale, html lang, demo critique, and image-processing behavior.
Curated GPT-Image-2 prompt library
The AI Create page now includes 30 curated GPT-Image-2 prompt examples with output images, author/source attribution, and localized title/prompt fields. Users can tune quality, ratio, and style below the prompt box, then apply an example as a draft.
Content bundles, homepage SSR translations, and task stability
Homepage, blog, updates, and review copy now sit behind clearer component and JSON bundle boundaries. Root and locale homepages render with initial translations, header controls are shared, and generation task status reads no longer commit the database session.
AI Create, image credits, and retake references
PicSpeak now includes a standalone AI Create page, review-linked retake reference generation, generation history, downloads, and reuse. Image generation credits, a 30-credit redeem code, and a 300-credit checkout pack are wired into billing and usage.
Content-to-workspace conversion paths
Blog, Gallery, and Home now send readers back to the workspace with intent-specific CTAs, while the analytics snapshot can track content views through clicks, uploads, and first critique completion.
Pro value repositioning and FAQPage duplicate fix
Pro is now framed as next-shoot guidance, complete review, and progress tracking instead of just deeper analysis. Locale home pages also no longer emit duplicate FAQPage structured data.
Growth loop, replay guidance, and next-shot checklists
History now surfaces a recent-vs-previous growth loop, review detail separates same-photo verification from new-photo retakes, and Flash suggestions are shaped into cleaner next-shot checklist actions.
Auth hardening, safer task retries, and more stable requests
Tightened CORS and cookie behavior, corrected webhook error codes, added activation-code rate limiting, locked JWKS cache refreshes, and made frontend requests cancellable.
Backend & frontend modular refactor
Split oversized backend routers and the review detail page into single-responsibility modules. No user-facing changes — this is a code organization update that improves maintainability.
Blog post view counts are now visible
Added public view counts for blog posts and wired article detail visits to increment the count automatically.
llms.txt, Enhanced Schema.org, and Localized Updates Pages
Added llms.txt for AI search visibility, expanded Person and SoftwareSourceCode structured data, and launched locale-pinned /zh/updates, /en/updates, and /ja/updates routes.
Blog Module, Gallery Sort, and Dark Theme Refinement
Launched a trilingual blog with 6 SEO articles, added gallery sorting (recommended, latest, top score, most likes), and recalibrated the dark theme to a warmer tone.
Locale-Pinned Home Routes, SEO Signals, and Gallery UI Refactor
Added /zh, /en, and /ja locale-pinned home pages, wired hreflang and JSON-LD signals across public routes, and moved scattered home, gallery, review, and workspace copy into shared i18n keys while splitting the gallery UI into focused components.
Gallery Ranking, Image Rendering, and Lint Gates Updated
The public gallery now ranks work with score-weighted freshness, uses stable paging cursors for that combined order, restores reliable gallery card image rendering, and ships a real frontend lint baseline.
Activation Codes and China Purchase Flow Added
Added an Afdian plus activation-code path for Chinese Pro purchases, tightened expiry-aware billing sync, and exposed redeem entry points across the home, review, promo, and account surfaces.
Moderation, Score Labels, and Gallery Cards Refined
Relaxed public-gallery moderation for clearly allowed borderline content, expanded review score labels to 10 localized tiers, and stabilized gallery cards for landscape images.
Home, Review, and Gallery Experience Refresh
Unified the home header with the app shell, added a sign-up entry, fixed review-detail display issues, and improved public gallery paging and thumbnail delivery.
Public Gallery Navigation and Filters Updated
The public gallery now supports history-style filters, URL-synced state, reliable return-position restoration, and thumbnail-only card loading.
Pro Launch Offer and Direct Checkout Updated
Standardized the launch-offer messaging, expanded Pro entry points, and sent purchase buttons straight to checkout.
Scoring Upgrade and Gallery Recommendation Logic
Unified scoring behavior across modes and added score-version-aware recommendation logic to the public gallery.
Public Gallery Likes and SEO Improvements
Added persistent gallery likes for signed-in users and tightened SEO metadata for public-facing pages.
Stricter Scoring Prompt and Correct Gallery Count
Tightened the AI scoring prompt and fixed the gallery header so the total count is correct from page one.
Public Gallery Browsing, Moderation, and Paging
Rolled out the public gallery workflow with moderation, guest browsing, gallery cards, and stable paging behavior.
History, Sharing, Replay, and Favorites Connected
Connected history filtering, sharing, export, replay analysis, and favorites across the frontend and backend.
