Changelog

Gallery

• Fix The Gallery search field now has a clear surface, border, and readable text in light mode
• Improved The search and filter panel now uses the same compact close control as other dialogs and panels
• Fix Parent images use a subtle navigation overlay in light mode so more of the thumbnail remains visible
• Fix Gallery load failures now show a proper alert icon instead of a broken character

Interface

• Improved Primary green accents plus green and red statuses and action buttons now use the same vivid colors in light and dark modes
• Fix Previous and Next image controls no longer remain available after leaving Edit mode for Create

Admin

• Improved User requests and character suggestions now share the same section surfaces and spacing as other Admin tools, and All Users expands fully without an internal scrollbar
• Improved Character cards are larger with more padding, clearer thumbnails, and centered actions
• Fix Sidebar focus rings are no longer clipped, and the Admin heading now uses the shared serif display type
• New The Admin dashboard now counts Daily remix images separately from Super Evolution Remix and shows Memifier-created memes

Gallery

• Fix Search, filter, and refresh controls now use the same text size
• Fix Gallery card footers now use light surfaces in light mode and dark surfaces in dark mode
• Fix Edit-page image details, navigation, character suggestions, and action icons now match the selected color theme
• Fix Opening a gallery image now shows Edit Image immediately without briefly flashing Create Image

Interface

• Improved Monochrome icons now follow their control colors across themes and use clearer sizing in image attachment controls
• Fix The New Tab Extension control now uses matching text, icon, badge, and surface colors when hovered or focused
• Improved Dark, light, and system options now include recognizable icons and a checkmark for the selected theme
• Improved The site logo now matches its original desktop size while retaining the compact mobile header
• Fix Suggest a Character opens correctly from the profile menu again
• Improved Theme and profile dropdowns now share the same components and automatically close when the other menu opens
• Improved Newsletter and character suggestion inputs now share the same compact form-control styling
• Improved Primary and secondary actions now share button sizing, typography, padding, and corner radius across dialogs and the Memifier
• Improved Fraunces headings now use the same medium weight throughout the site

Interface

• Improved Generation, Gallery, Family Tree, Super Remix, Changelog, and Admin now share a quieter, image-led design system
• New Choose a dark, light, or system color theme from the header and keep that preference across visits
• Improved Shared navigation, dialogs, loading states, filters, cards, and administrative tools now use accessible reusable components while retaining existing workflows

Characters

• Fix Uploading images while editing a character now adds them without removing existing reference images
• New Character reference images can now be pasted directly into admin add or edit forms and the Suggest a Character form
• New Remove uploaded or pasted previews before saving, and delete existing character images while editing

AI Models

• Improved AI features now consistently use the configured Azure deployment names, so models can be upgraded without code changes
• Fix Daily remixes now recover when an AI response includes an extra structured-data wrapper

Image Generation

• Fix Aspect ratio choices now use contrasting text and backgrounds in Chrome dropdowns on ChromeOS
• Improved Edit prompts now preserve unaffected regions and correct quality drift without unnecessary restyling

Gallery

• Improved Gallery cards now reserve space while images load and settle into the masonry layout individually without scroll jumps
• Improved The next gallery page now loads through a browser observer instead of repeated scroll measurements
• Fix Scrolling near the end of the gallery now consistently loads the next page

Image Generation

• Improved Generating or editing an image now scrolls the page all the way to the top
• Improved Generation time now appears in translucent white beneath the rotating loading message instead of under the Generate button

Image of the Day

• Improved Daily actions now account for the source image and selected character

Characters

• New From an image’s Edit page, move and resize a selection by its corners, then use the crop as a reference for a new character suggestion
• Improved The Edit-page action now matches the person-plus icon in the profile menu
• Improved Suggest a Character keeps its desktop button styling below the image card on mobile with no full-width background bar

Gallery

• Improved Keep scrolling past the 500 most recent images to load older stored images

Profile Menu

• Improved Admin access now appears at the bottom of the profile menu behind a divider, and only shows for administrators

Memifier

• Improved Edit text directly in each layer row, with a clearer Re-memify action for existing Memifications
• New Add auto-sizing Comic Sans speech and thinking bubbles with draggable placement, eight tail directions, connected speech tails and cloud-shaped thinking bubbles whose thought trails stay attached as they resize
• Improved AI edits now warn when the current meme will be baked in and its layers will no longer be editable

Memifier

• New Memify any gallery image from the Edit screen with as many draggable text layers as you need
• New Style each layer with font, size, fill, outline and alignment controls
• Editable Reopen a Memification to change its original layers without baking text over text
• Connected Memifications carry their source lineage, display a gallery badge and can continue through the regular AI Edit Image workflow

Characters

• New Type @ in the prompt box to bring up a character menu, and keep typing to narrow it down — fuzzy matching means @pork finds The Porker
• Improved Adding characters from the character menu now writes them to the end of your prompt, like “a hedgehog on a bike - With @The Porker & @Sonic”
• Improved Deleting a character’s name from the prompt removes it from the selection, so the two always stay in sync
• New Added characters are highlighted in the prompt box, and Backspace, Delete and the arrow keys treat the whole name as a single item instead of one letter at a time

Characters

• Improved Characters are now ordered by how often they’ve been used to generate images, instead of a manual sort order
• New Characters added in the last 14 days float to the top of the list so they get a fair chance to be discovered
• Faster The character list is served from a cached manifest instead of a database lookup on every page load, refreshed daily and on every character change
• New The admin character grid shows all-time usage counts, plus a Rebuild Order button to recompute the order on demand

Super Evolution Remix

• New Admins can run a finite chain of 1-20 successful remixes, with each completed image becoming the next image’s parent
• New A dedicated Super Remix gallery keeps experimental low-resolution results separate from the main gallery and search
• Reliable Progress, cancellation, failure limits, queue recovery, and stale-lock handling keep batches bounded without branching the lineage

Security & Authentication

• Private content enforcement — Hedgehog feed and content APIs now require sign-in, so unauthenticated clients can no longer browse image feed, conversations, messages, reactions, or profile pictures
• Protected image actions — Image generation, editing, and analysis endpoints now require authentication, preventing anonymous API usage against model-backed features
• Defense-in-depth access checks — Access control is now enforced at both the Static Web App route layer and API handler layer, reducing the risk of accidental public exposure

Reliability & Verification

• Auth guard regression coverage — Added automated tests that validate authenticated-principal parsing and confirm protected APIs keep authentication guards in place

Azure OpenAI API Guidance

• API version clarified — Confirmed that image generation and edit calls still require a dated api-version on deployment-style Azure OpenAI endpoints, while Responses API calls use /openai/v1/responses without an api-version query
• Current preview version check — Confirmed 2025-04-01-preview is the current preview version to keep for the existing image endpoint pattern

Maintenance Mode

• New Dedicated maintenance page with the site logo and a friendly "down for maintenance" message, so visitors see a proper notice instead of an error when the site is offline
• New The GitHub deployment workflow can now be swapped between maintenance mode and the live site by copying a single backup file

Search & Analysis

• Fix Image analysis (used for search text) now reliably completes instead of being silently killed by the server, which was causing many images to have no search text
• Fix Image analysis descriptions are no longer truncated mid-sentence
• New Reanalyze button in the admin image inspector to re-run analysis on any individual image
• New Search Analysis backfill tool on the admin page finds all images missing analysis and processes them with a live progress counter

Characters

• New Select named characters (like Hedge Hogan) when generating or editing images to automatically include their reference photos and descriptions, helping the model reproduce them consistently
• New Collapsible Characters panel on the generate/edit page shows character thumbnails you can toggle on or off
• Improved Selecting characters reduces your available upload slots (10 total), and the dropzone updates to show the adjusted limit
• New Character reference images appear in their own “Characters” section when viewing image details, with names shown on hover
• New Suggest a Character — submit character ideas with reference images from the profile menu for admin review
• New Admins can create, edit, and delete characters from the admin page with reference image uploads and display ordering
• New Pending character suggestions appear on the admin page for one-click approve or deny
• New The Admin menu item shows a notification badge when character suggestions are waiting for review
• Improved Selected characters are cleared when you start a new session, create a new image, or load an existing image for editing
• Improved Admin page shows image stats and pending suggestion count at the top for a quick overview
• Improved Characters on the admin page are displayed as a compact card grid instead of a tall list
• Improved Styled file upload buttons in the suggest dialog and admin character form match the site design

Upload Drop Zone

• Improved The upload area now says “Drop images here to add” instead of “to edit”, which better describes what happens when you attach images
• Fix The paperclip icon in the drop zone is now visible on dark backgrounds instead of rendering as black-on-dark

Content Safety

• Fix When the image model accepts a prompt but blocks the generated image later, the site now detects this and shows the content safety warning instead of silently timing out
• Fix The content moderation warning now clears automatically when you navigate away from the page

Admin Tools

• New Each image on the admin page now has an Inspect button that shows the full database object in an overlay
• Fix The Rebuild Manifest button on the admin page works again

Images API Migration

• Improved Image generation and editing now call the image model directly instead of routing through an intermediary, eliminating silent content-oriented rejections that were causing generation failures
• Improved Content moderation detection is now simpler and more reliable — a single clear error code instead of three ambiguous failure modes
• New Search indexing now happens in the background after you see your image, so generations feel faster while images remain searchable

Performance

• Improved Post-processing (image resizing, uploads, and database writes) now runs in fully parallelized phases, shaving seconds off each generation

Image Quality

• Fix The hedgehog instruction now works correctly even when asking to remove or change the background

Testing

• New Integration test suite covering all image paths: generate, edit, transparency, inpainting, content filtering, and search analysis
• Improved Tests use fixed input images for reproducible results — run with --save to inspect generated outputs
• New Pushing to main now automatically runs tests when API files have changed, blocking the push if any fail

Edit Image Timeout Fix

• Fix Edit requests now send image URLs directly to OpenAI instead of downloading and re-encoding them, making edits significantly faster
• Improved The original prompt is now always looked up for edits, giving the AI better context for your changes
• Improved Post-processing (image resizing, uploads, and database writes) now runs in parallel instead of sequentially

Generation Timeout Fix

• Fix The OpenAI API calls in both generate and edit functions now have an explicit 3-minute timeout, preventing the function from hanging indefinitely on a stalled model response during its 45-second execution window
• Improved Documented that the 40-second client-side timeout is intentionally set just below Azure Static Web Apps’ hard 45-second API limit, so the client cleanly aborts and switches to polling before the platform kills the connection

Edit Instruction-Following Improvements

• Improved Your edit instruction is now the leading text the AI sees, instead of being buried after long context — the model follows edit requests more accurately
• Improved When editing an existing image, the context sent to the AI uses your original short prompt (capped at 300 characters) instead of the expanded 1000+ character revised description, preventing context bloat during back-to-back edits
• Improved Quality policies and the hedgehog requirement now live in a separate system message, so they don’t compete with your actual edit instruction for the model’s attention

Responsive Layout

• Fix On narrow screens, the profile avatar now stays in the upper-right corner next to the logo instead of dropping down to a new row below the navigation

Caption & Sources Layout

• Improved The image caption and source thumbnails now sit side-by-side: prompt text on the left, source images on the right
• Improved Attached source images now load 128px WebP thumbnails instead of the full 1024px originals for faster display
• New The parent image thumbnail shows an angle-right arrow overlay, distinguishing it from attached images that open the lightbox
• New The prompt text now has a “Prompt” heading, matching the “Attached” and “Parent” labels on the source thumbnails
• Improved Source labels shortened to “Attached” and “Parent” for a cleaner look
• Fix Closing the lightbox now clears the previous image, preventing a flash of the old image when opening a new one

Content Safety Improvements

• New When the model declines to generate without an explicit error (e.g. copyrighted characters), the model’s explanation is shown instead of a generic error
• Improved Removed speculative timeout-based content filter detection that was causing false positives; only explicit API confirmations are reported now
• New 91 automated tests covering all content moderation paths

Concurrent Generation Fix

• Fix Image polling is now scoped to the specific conversation — fixes a bug where two users generating at the same time could receive each other’s images
• Improved New generations and edits create a conversation ID client-side, ensuring polling always targets the correct request

Other Changes

• New Added a link to the changelog in the user profile dropdown menu
• New Added a reusable Copilot prompt file that automatically summarizes session changes and updates both changelog files
• New A git pre-commit hook now reminds you to update the changelog before committing
• Improved Removed shortcuts.md — learnings are now captured in the structured prompt file

Content Safety Feedback

• New When Azure OpenAI blocks a request due to content safety filters, the site now detects it instead of showing a generic error
• New Content filter warnings appear as an amber banner directly below the prompt input — no more confusing error toasts
• Improved The warning clears automatically when you start typing a new prompt
• Improved Content moderation is returned as a normal 200 response, so there are no red errors in the browser console

Source Image Provenance

• New When you create or edit an image using pasted, drag-and-dropped, or file-picked images, those source images are now saved alongside the result
• New Source thumbnails appear below the prompt caption on the edit page — click any to view full size in the lightbox
• Improved Deleting an image also cleans up its associated source images from storage
• Improved Prompt caption now sits below the image at all screen sizes instead of overlaying it
• Improved Download and Share buttons appear inline next to reaction emojis for a more compact layout

Search & Filter

• New Full-text search — search the gallery by what's in the images: scenes, objects, colors, mood
• New AI-powered image descriptions — every image now has a rich text description generated by gpt-5-mini vision
• New BM25 relevance ranking — search results ranked by relevance using Cosmos DB full-text search
• New Multi-reaction filtering — select one or more reaction types to filter the gallery
• New Combined search + filter — use text search and reaction filters together
• New Unified filter panel — search and reactions in a collapsible panel, open with click or / shortcut

Responses API Migration

• Improved Image generation and editing now use the Azure OpenAI Responses API with gpt-5 as the mainline model
• Improved Every prompt is rewritten into a detailed scene description (1000+ chars) for richer search metadata
• Improved All 449 existing images analyzed and indexed for search

Search & Filter UI Polish

• New The "Search & Filter" button now highlights when the filter panel is open
• New A "Reset Filters" button appears in the toolbar when any filters are active
• Improved All toolbar buttons now use consistent solid-fill hover styles — red for reset, green for the others
• Improved Filter panel now smoothly slides open and closed with a fade transition
• Improved Search input redesigned as a solid rounded pill with a lighter fill background
• Fixed Header navigation no longer wraps to two lines on mobile screens

Other Changes

• Improved Analyze button removed from the image action bar (API endpoint kept for future use)
• New 156 unit tests covering image processing, prompt construction, API response shapes, and more

Hedgehog-Themed Loading Messages

• Improved Status messages now appear directly on the loading overlay instead of as toast notifications
• New Loading messages are now hedgehog-themed (e.g. "Lovingly adjusting quills...", "Carefully nibbling around the edges...")
• New While waiting, the overlay cycles through fun follow-up phrases like "Still foraging for pixels..." and "Rustling through the bushes for your image..."

Gallery Auto-Refresh

• New Gallery automatically checks for new images — every minute at first, slowing to every 10 minutes
• Improved Auto-refresh pauses when you switch tabs and resumes when you come back
• Improved Clicking the site logo while on the gallery now refreshes it
• Improved Navigating away and back resets the check frequency

Site Migration and Official Domain

• New We official now, hogs! New domain: hedgehog-generator.com
• Improved Moved off Supabase backend and refactored to Node.js running on Azure
• Improved Migrated all site assets and database to Azure

Authentication

• New So our AI model doesn't get spammed by randos, the site now requires log in with a Microsoft account, but with the caveat that all image generation is all still anonymous
• New User badge in header showing logged-in alias with sign-out link and optional user profile picture feature

Inpainting Mode

• New Draw directly on images to define areas for the AI to edit, with brush/eraser and adjustable size
• New "Use inpainting mode" toggle in the options panel when editing an image
• New Mask toolbar with Brush/Eraser switcher, size slider, Undo, and Clear controls
• New Masks sent via the OpenAI mask parameter with dimension validation

Transparent Background Option

• New Background dropdown in settings to choose between Solid and Transparent backgrounds
• New Generate and edit functions now support true PNG transparency via Azure OpenAI
• New Checkerboard pattern behind all images (edit view, gallery, visualizer) to reveal transparent areas

Simplified Settings

• Improved Removed Quality and Style dropdowns — always uses HD quality and Vivid style
• Improved Settings now show just Size and Background in a clean single row

High Input Fidelity for Edits

• Improved Image edits now use high input fidelity, preserving finer details like faces, logos, and textures

Custom SVG Icons

• Improved Replaced emoji icons with custom SVG icons across the entire app
• New Action buttons use new icons: eye (Analyze), file-download (Download), share-square (Share)
• Improved Gallery nav, refresh, close buttons, empty state, visualizer badge, and file upload all use SVG icons

Navigation Overlay

• Improved Prev/next buttons now float over the image instead of beside it
• Improved Buttons hidden by default and fade in on hover for a cleaner look

Gallery Image Navigation

• New Prev/next arrow buttons on the edit view let you step through gallery images
• New Left/right arrow keys navigate between images in the edit view
• Improved Navigation always available — loads full image list automatically

UI Cleanup

• Fix Reactions and action buttons now properly centered under the image
• Improved Removed the noisy "Image loaded" toast notification
• New Click an image in the edit view to open it full size in a lightbox overlay

Jar of PASTE™ Reaction

• New Added Jar of PASTE™ reaction emoji to the picker
• New Database migration to support the new paste reaction type

Visualizer Loading State

• Improved Visualizer waits for all thumbnails to download before calculating layout
• New Loading spinner shown while images are downloading
• Improved Grid fades in smoothly once images are ready

Bandwidth Optimization – Multi-Size Images

• New Images stored in three sizes: original (1024px), medium (512px), thumbnail (128px)
• Improved Aspect ratio preservation instead of square crops
• Improved Gallery loads medium images, reducing bandwidth ~76%
• Improved Visualizer uses thumbnails, reducing bandwidth ~95%
• New Backfill complete — all 227 existing images processed
• Improved Graceful fallbacks for older images without size variants

Image Sizes & Analysis

• New Multiple image size options for generated images
• New AI-powered image analysis with automatic tagging
• New Backfill tool to analyze existing images
• Improved Gallery now displays image metadata

Firehog Hearthog Reactions

• Improved Replaced llama-fire emoji with new firehog emoji
• Improved Migrated all existing llama-fire reactions to firehog
• New Updated Hearthog reaction graphic

Gallery Masonry Performance Optimization

• Improved Rewrote gallery layout engine for smoother performance
• Improved GPU-accelerated positioning with CSS transforms
• Improved requestAnimationFrame batching for paint-synced updates
• Improved ResizeObserver for efficient resize detection
• Improved Batched DOM operations to minimize reflows
• Improved CSS containment with contain and will-change properties
• Improved Left-to-right ordering with true interlocking masonry

Family Tree Visualizer Redesign

• New Responsive masonry layout for family tree cards
• New Hierarchical tree visualization with connector lines
• Improved All family trees fully visible by default
• Improved Responsive columns: 3 on large, 2 on tablet, 1 on mobile
• Fix Preview popup now shows images at natural aspect ratio

Reactions

• New Emoji reactions on gallery images with custom emojis
• Improved Reaction counts and display on gallery items

Family Tree & Editing

• New Family Tree visualizer to explore image lineage
• New Image editing with inpainting via mask drawing
• Improved Navigation with shared header component

Visualizer Page Improvements

• Fix Rewrote tree connector logic for proper parent-to-child lines
• Improved Full-screen layout with content scrolling under sticky header
• Improved Removed redundant visualizer-specific header
• Improved Preview panel now appears near the clicked image

Gallery Enhancements

• New Infinite scroll — loads 16 images at a time
• New Masonry layout with left-to-right reading order
• New "Original" tags for top-of-lineage images
• New Loading spinner with "Loading gallery…" text
• Improved Gallery items fade in together to prevent layout jumping

UI/UX Design Updates

• Improved Cooler, muted sage green and gray color palette
• Improved Smaller, more minimal navigation header
• Improved Individual bordered filter pills
• Improved More subtle shadows and hover states

Bug Fixes

• Fix Restored sticky header by moving positioning to #shared-header
• Fix Restored frosted glass backdrop blur on header
• Fix Added missing loading overlay CSS for edit view
• Fix Added missing CSS for toggle switches