Changelog

New Features

• Edge Item Picker — Right-click any connection to visually browse what's flowing through it. Images and videos show as tiles, text as readable rows. Pick a specific item or choose "All Items" to pass everything downstream. Connections display badges ("All" or "#3") so the active selection is always visible at a glance
• Faster AI Generation — Video and image generation jobs complete noticeably faster by receiving instant completion signals from the server instead of repeatedly checking for results
• Background Saving — Running a workflow no longer pauses to save first. Saves happen seamlessly in the background, and the app skips saving entirely when nothing has actually changed
• Render Progress — Caption and slideshow renders now display a real-time progress bar instead of just a spinner
• Video Merge Upgrades — Auto-detects input resolution from the first video (no more fixed 720p), drag-and-drop video ordering in the sidebar, supports up to 10 input videos, and downloads all clips in parallel for faster processing
• Persistent Output Edits — Text edits made in the sidebar (e.g., tweaking AI-generated copy) now save to the database and survive page refreshes
• Caption Timing Control — New word grouping speed setting in the caption sidebar for fine-tuning how words appear on screen

Improvements

• Compact, Polished Nodes — Tighter node layouts across the board. Image AI uses a clean 2x2 grid, Video AI controls are logically grouped, and Captions show only Style + Position on the node with the rest in the sidebar
• Better Connections — Connections now render above nodes so arrows are always visible, have a wider click area for easier interaction, and auto-select the first item when connected
• Captions & Video Merge Official — Both modules graduated from experimental to full production status
• Consistent UI Styling — All dropdowns, scrollbars, and interactive elements are now fully themed to match the app design across all platforms
• Settings Quick Access — Gear icon next to the Run button for one-click access to node settings
• Faster, Smarter Saving — Saves transfer less data, skip unnecessary work, and thumbnail generation is throttled to keep things smooth
• Unsaved Changes Warning — Browser warns before closing a tab with unsaved workflow changes
• Correct File Downloads — Downloaded files now use the correct format extension (WebP, JPEG, PNG) for proper OS previews and thumbnails

Bug Fixes

• Caption and slideshow renders were stuck at 0% progress throughout the entire render
• Caption cancel button wasn't working
• Video Merge "First Only" audio mode was incorrectly mixing all audio tracks
• Aggregator outputs were losing their structure when passing through connections
• Close Iterator to Gate connections were silently dropping values
• Edge context menu showed stale data until manually refreshed
• Clicking an image in output history did nothing
• Closing the sidebar could leave behind stale state
• Several iterator and loop execution edge cases fixed

Removed

• Add Audio Node — Removed in favor of Video Merge's built-in audio controls

New Features

• Iterator Loop Execution — New flow-control nodes for batch processing on the canvas. Connect an Iterator to split a list (like scene descriptions) into individual items, run each through any chain of AI nodes, then collect results with Close Iterator. Supports parallel execution (up to 10x concurrency) for faster batch processing
• Slideshow with Built-in Captions — Slideshow node now renders captions directly in a single Lambda pass, removing the need for a separate Video Captions render step
• Moving Images Product — New product page for generating slideshow videos from product assets
• Node Color Theming — Each node category now has a distinct accent color (sky, emerald, violet, amber, etc.) for better visual organization on the canvas, with matching focus rings and sidebar highlights
• Animated Canvas Connections — Edges between nodes feature animated dashed lines with a glow effect on hover
• Maintenance Mode — Instantly take the site offline via a Vercel Edge Config toggle — no redeployment needed. Includes a branded maintenance page, admin bypass via HMAC cookie, and API 503 responses with Retry-After headers
• Visitor Analytics — Vercel Web Analytics integration for page views, visitor tracking, and demographics

Improvements

• Context Menu Reorganized — "Add Node" uses nested submenus grouped by purpose (Input, AI, Compose, Flow) with stability badges (Beta, Alpha) on experimental nodes
• Toolbar Streamlined — Only Input and AI nodes in the top toolbar; Compose and Flow nodes accessible via right-click context menu
• Gate Node Redesigned — Candidates display in a sidebar gallery by default with a toggle for inline display. Multi-item array outputs expand into individual selectable candidates
• Caption Presets Refined — Replaced broken presets with Pop (energetic social media) and Cinema (professional). Font selector uses a curated dropdown of 10 fonts instead of free-text input
• Slideshow Animations Unified — Single preset selector replaces 6 separate controls. Two refined presets (Smooth, Dynamic) with 3-phase entrance/motion/exit architecture
• Slideshow Quality — Minimum resolution raised to 1080p. Edge expansion rewritten for cleaner reflections
• Custom Style Notes — Free-text style notes from the Global Style node now reach the final AI prompt across all 7 pipeline layers

Bug Fixes

• Edge item selector now works for all multi-item nodes (Image AI, Voice AI, Close Iterator), not just Text AI
• Caption positioning broken on portrait/mobile videos — bottom position rendered at center
• Caption presets not driving visual styling in the rendered video (changing preset now updates font, colors, animation, and effects)
• Canvas click-through where the controls header blocked clicks in the gap between left and right control groups
• Parallel execution not triggering for pre-existing Iterator nodes (missing default value)
• Iteration progress display clipped on narrow Iterator nodes

New Features

• Slideshow Node — New composition node that compiles multiple AI images into a polished video. Connect 2--20 images, choose from 5 transition types (fade, slide, wipe, clock wipe, iris), and add Ken Burns motion effects with configurable intensity
• Audio-Synced Slideshows — Connect a Voice AI node to the Slideshow's audio input and set timing to "auto" -- images automatically distribute across the voiceover duration so visuals stay in sync with narration
• Slideshow Image Panel — Drag-to-reorder sidebar panel for arranging slideshow images. Ordering persists across re-generations so you don't lose your arrangement
• Voice AI Speed Control — 0.7x--1.2x speed slider for voiceover generation. Faster speeds produce shorter audio, perfect for fitting narration to a target duration
• Voice Search & Preview — Browse 100+ voices with search by name and gender filter. Play free preview samples before selecting -- no credits spent until you generate
• 2 New Voice Models — ElevenLabs v3 (most expressive, best for dramatic narration) and Flash v2.5 (fastest and cheapest, ideal for drafts and iteration)

Improvements

• Captions Node Rebuilt — Completely redesigned with 6 style presets (karaoke, minimal, bold, boxed, outline, typewriter). Customize fonts (10 Google Fonts), colors, backgrounds, outlines, positioning (top/center/bottom), 4 word animations (scale, color, both, none), and 4 entrance effects (fade, slide, pop, none). Preset names show a modification indicator when you customize individual fields
• Voice AI Per-Character Billing — Pricing now scales with text length instead of flat per-call. Short texts cost less, long texts cost proportionally more -- matching how the voice provider actually charges
• Slideshow Timing Modes — Choose between manual (fixed duration per image) and auto (driven by connected audio length). Images ordered by source node position on the canvas (top-to-bottom, left-to-right)
• Slideshow Output Options — 4 aspect ratios (9:16, 16:9, 1:1, 4:3), 2 resolution levels (720p, 1080p), and a random option for both motion style and transition type that varies per image
• Voice Speed Indicator — Voice AI canvas node now shows the current speed when it differs from default, so you can see at a glance which voices are adjusted
• Prompt Template Resolution — All AI providers now share a single template resolution system, eliminating duplicate logic

Bug Fixes

• Kling error 1303 (parallel task limit) now stops retrying immediately instead of hitting the same quota wall 2 more times
• Kling image max concurrent tasks capped to 6 to match account limit
• Kling image generation producing N*N images instead of N -- batch count and iteration count are now independent
• Image AI series mode now passes all generated images to downstream nodes (previously only the first image flowed through)
• Kling image provider skipping prompt template resolution -- now applies modifiers and style correctly
• Voice AI language parameter now forwarded to ElevenLabs API for language-optimized generation
• Server-side execution (product workflows) for captions and slideshow nodes now creates proper job records
• VideoCaptions composition no longer crashes when no video source is provided (renders captions on black background)
• Status-poll rate limit raised from 60 to 120 per minute to handle concurrent async jobs without 429 errors

New Features

• Kling V3 Video Models — Two new video models (V3 Standard and V3 Omni) with flexible 3-15 second durations. V3 Omni is now the default for new Video AI nodes
• Native Audio on Video — V3 models generate AI audio that matches the video content. Separate pricing for sound on/off so you only pay for audio when you use it
• Kling Image Generation — Three new image models for different needs: V3 (fast generation), O1 (reasoning-enhanced for complex prompts), and V3 Omni (full-featured with 4K output, series mode, and cinematic controls). Choose from 8 aspect ratios and up to 4K resolution
• Image Series Mode — Generate 2-9 related images in a single pass from a reference image — perfect for film storyboards, social media carousels, product shot collections, and brand asset sets with visual consistency. Set a specific count or use "Auto" to let the model decide
• Multi-Shot Video — Create videos with up to 6 shots using the storyboard editor. Auto mode lets the model plan transitions, or use Custom mode to define each shot's prompt and duration for full creative control
• Multi-Reference Images — O1 and V3 Omni support up to 10 reference images per generation for style transfer, multi-view generation (front/side/rear), image editing, and character swaps. Address each in your prompt with <<<image_1>>>, <<<image_2>>>, etc.
• Auto Aspect Ratio — O1 and V3 Omni intelligently select the best aspect ratio based on your prompt content
• Model Guides — Per-model documentation with tips, capabilities, and usage guidance. Access from the floating guide button, organized by Video AI and Image AI categories
• 12 Kling Prompt Templates — Optimized templates for cinematic stills, storyboard series, ad campaigns, multi-view, portrait series, HD detail, and mood boards
• Model-Scoped Templates — Prompt templates are tagged with compatible models and automatically filter when you switch models
• Template Descriptions — Each prompt template shows a description below the selector explaining what it does

Improvements

• Sound + Video Reference Safety — V3 Omni automatically disables sound when a reference video is connected, preventing unsupported configurations
• Sound-Aware Billing — V3 models have separate pricing for audio on and audio off. V3 Omni has 8 billing tiers covering mode, video input, and sound combinations
• Resolution-Aware Image Billing — Kling image generation priced by resolution tier (1K, 2K, 4K)
• Dynamic Settings Panel — Switching between Google and Kling image models updates the settings panel to show only relevant options (resolution, series mode, aspect ratios)

Bug Fixes

• Video sound toggle state leaked as true when switching from Pro to Standard mode
• Image series mode could be enabled without reference images (now validated server-side)
• Kling V2.6 Pro no-audio pricing corrected from 3c/s to 5c/s to match actual provider cost
• All Kling V3/V3 Omni/Image pricing updated to official rates

Improvements

• Stronger Authorization — Workflow output endpoints now verify you own the workflow before returning data
• Polling Timeout — Stuck AI jobs automatically time out after 10 minutes instead of spinning forever
• Unicode Error Messages — Error messages from multilingual AI providers (Kling, ElevenLabs) now display correctly instead of being stripped to empty strings
• Upload Rate Limits — Uploading reference images and videos no longer consumes your AI generation quota
• Inngest Resilience — TikTok research function restructured into 5 idempotent steps for clean retries; image generation no longer risks exceeding step size limits
• Admin Billing Performance — Billing queries now filter to last 90 days and only fetch user emails when needed

Bug Fixes

• Fixed TikTok Research nodes always running in "content" mode during product runs regardless of user setting
• Fixed Story AI nodes using the wrong rate limit bucket in product runs
• Fixed TikTok and Brand Analysis poll callbacks passing undefined to success handlers
• Fixed Voice AI using a hardcoded model ID instead of the registry default
• Fixed credit hold race condition — credits are now reserved before AI work begins
• Fixed pinned iteration outputs being ignored in Story AI and Trend Selector nodes
• Fixed captionEdit ghost reference in captions error context (renamed to videoCaptions in v0.15)
• Fixed batch executor logging being completely disabled — errors during "Run All" are now visible

Improvements

• Explicit Credit Holds — Credits are now tracked through a dedicated hold lifecycle (held → settled/refunded/expired) instead of immediate deduction, preventing balance drift
• Fractional Credit Pricing — Credit pricing now supports decimal values for more granular cost-per-operation tuning
• Balance Reconciliation — New audit view detects drift between ledger balance and hold records

Bug Fixes

• Fixed balance inflation bug where sync route error handling ordered billing operations incorrectly, causing credits to be double-refunded
• Fixed expired holds not restoring credits before marking the hold expired (permanent credit loss on TTL expiry)
• Fixed settle/refund RPCs accepting cross-user requests — now enforces user_id ownership guard
• Fixed Inngest settle-billing applying a harmful refund fallback on completed jobs (retry handles recovery)
• Fixed video generation onFailure overwriting cancelled status with failed when user cancels during processing
• Fixed admin billing dashboard using wrong unit conversion (CREDITS_PER_DOLLAR instead of CENTS_PER_DOLLAR)
• Fixed captions status route writing execution records with mismatched billing reference keys

New Features

• Node Iterations — Run text, image, and video AI nodes multiple times (1–8) to generate several versions at once. Pin your favorite output with "Use This" to control what flows downstream
• Gate Node — Add review checkpoints to your workflow. The Gate collects outputs from upstream nodes and lets you pick the best one before continuing — with image thumbnails, video previews, or text cards depending on the type
• Credit Billing — See exactly how many credits each AI node costs before you run it. Credits are reserved upfront, refunded if something fails, and tracked in your profile
• Product Steps — Products with Gate nodes automatically break into steps. The pipeline pauses at each Gate for your review, then continues when you pick a candidate or retry
• Image Size Options — Choose 1K, 2K, or 4K resolution on supported image models
• Credit Profile Page — View your credit balance, transaction history, and usage breakdown by operation type
• New Prompt Templates — Added UGC/Casual templates for images and videos, plus Character/Object Design templates for isolated subjects

Improvements

• Fresh Visual Theme — New warm parchment light theme replacing the dark slate look
• Better Sidebar History — Pinned output moves to the top of the history; "Use This" button becomes an "Active" badge when selected
• Image Selection in Products — Gate image candidates now show as a tile grid with larger thumbnails instead of tiny rows
• Simpler Pricing — Credit costs match provider costs 1:1 (1 credit = 1 cent), no markup
• Video costs now scale with duration (5s, 8s, 10s priced proportionally)
• Caption rendering cost reduced to 15 credits
• Edit nodes (trim, merge, audio merge) clearly marked as "Free"
• Story AI node reorganized with prompt template + style selectors
• Cancel product runs instantly, even while waiting at a Gate

Bug Fixes

• Fixed canvas flickering/zooming during node execution and saving
• Fixed Story AI style dropdown showing wrong options
• Fixed Voice AI credit cost badge showing blank
• Fixed cost badges missing on Website Research and TikTok Research nodes
• Various Gate node reliability fixes

New Features

• Outputs Survive Refresh — Your AI generation results are now saved to the database. Close your browser, come back, and your outputs are still there
• Template Version Control — Create, rename, and manage versions of your prompt and style templates. New versions start as drafts before going live
• Additional Instructions — All AI nodes now have a dedicated "Additional Instructions" field, always visible regardless of which template you pick
• Cancel Research Jobs — TikTok research and brand analysis can now be stopped mid-execution
• Multi-Platform Trends — Trend system now supports Instagram, YouTube, and Facebook alongside TikTok
• Image Attachments in Tickets — Attach screenshots to your bug reports and feature requests

Improvements

• Workflows save much faster (smaller save payloads)
• Smoother sidebar performance when switching between nodes
• Consistent settings layout across all AI nodes
• Better trend pipeline reliability with progress tracking
• Template reordering is now instant

Bug Fixes

• Fixed a loop that caused the sidebar to flicker on certain nodes
• Fixed template content being partially overwritten on save
• Fixed required fields not being validated on nodes with multiple inputs
• Fixed async jobs not recovering properly after page refresh

New Features

• Navigation Bar — Consistent header across all pages with active page indicator
• Bug Reports & Feature Requests — Submit feedback from anywhere in the app via the floating button
• Full-Screen Text Editor — Expand text inputs into a large dedicated editor with markdown preview
• Edit AI Outputs — Edit text generated by AI before it flows to the next node, with "Edited" badge tracking
• Custom Product Pages — Build unique landing pages for your products with drag-and-drop sections
• Multiple Product Outputs — Products can now have multiple output nodes with independent labels
• Draggable Toolbar — Reorder your node toolbar by dragging. Keyboard shortcuts follow your custom order
• Cancel Running Jobs — Stop any in-progress job mid-execution
• Video Reference for Text AI — Connect a video to the Text AI node for video-aware text generation
• Changelog — This popup! Shows you what's new each day

Improvements

• Admin pages load instantly (no more spinners)
• Trend tools merged into a single, cleaner interface
• Better product page layout with separated inputs and outputs
• Improved auth protection and security
• Faster template loading with client-side caching

Bug Fixes

• Image and video outputs now reliably survive browser refresh
• Fixed various product system issues (input validation, output rendering)
• Fixed TikTok Research "Analyze" button when connected to another node
• Nodes no longer spawn off-screen
• Voice AI language setting now works correctly

New Features

• Products — Turn any workflow into a shareable product with its own page. Run it, close your browser, and come back to finished results
• Background Execution — Workflows now run on the server. No need to keep your browser open — results are saved automatically
• Global Styles — Set a visual style once and it applies across all your AI nodes. Keep your brand look consistent without repeating settings
• Collected Videos — Browse and manage your library of trending video references
• Jobs Dashboard — Track all your active and completed runs in one place
• Redesigned Navigation — New signup page, cleaner user menu, and improved app header

Improvements

• Better settings panels for input nodes — more control right in the sidebar
• Improved trend browsing with video previews
• All AI nodes now support style templates

Bug Fixes

• Fixed video generation issues with Kling
• Improved reliability for Google AI image, video, and text generation
• Fixed Story AI style inconsistencies