# Slopify 2.0 to 3.0

Versions 2.0.1 to 2.5.0, released 25 to 27 September 2026, and 3.0.0, released 27 September 2026.

Through 2.x, Slopify learned to make more than one long video: a PDF of the article, a YouTube description with chapters, shorts, and real editing (cuts on sentence ends, transitions, a Look). 3.0 is about the steps around the video. It finds its own topics, reviews its own work, knows what a run cost, waits out plan limits instead of failing, and fills in YouTube Studio for you. You still press Publish. The narration sounds better too: one steady volume, real pauses between sentences, and tables and figures described instead of read cell by cell. The whole app has a new design, and every control now explains itself.

## Highlights

- Much faster videos. A 108-minute video with burned-in captions renders in about 12 minutes instead of over 2 hours.
- Shorts: the best 60–120 s moments of the narration become vertical clips with new images and word-by-word captions.
- Video editing: cuts that follow the narration, transitions, a Look, chapter cards, your own clips and animated images.
- A PDF of the article, and a YouTube description with chapters, hashtags and tags.
- Automatic reviews: a reviewer model checks the article, images, narration, thumbnail and shorts, and redoes what fails, up to a limit you set.
- Schedules that write their own topics from a series brief, and a calendar of the coming four weeks.
- Channels with a brand kit, a cast library for consistent characters, and episode memory.
- Multiple voices: audiobook, podcast, radio drama and interview, with MP3 and M4B files.
- Level the volume: every narration piece is brought to one loudness, and the video is mastered to −14 LUFS, the level YouTube plays at.
- Pauses between sentences (0.4 s by default) and, if you want them, between paragraphs.
- Narration that reads like speech: tables, figures, diagrams, equations and code are described in a short spoken passage, and shown on screen as a card while it is spoken.
- Projects in other languages, with word-by-word captions in nine of them.
- Prepare upload and the Slopify Studio browser extension fill YouTube Studio's upload dialog. You press Publish.
- Run cost on every project, including what CLI runs would have cost through the API and how much of your plan limit they used.
- A first-run screen, a sample project, audiobook and podcast demos, and a 60-second short as the first thing you make.
- A new design across the app, with an info button on every setting: 406 explanations in one catalogue.
- Patch notes inside the app, in Settings → Patch notes, opened once after each update.
- Slopify is now licensed under Apache-2.0.

## In 2.0.1 to 2.5.0

### Video
- Each image stays on screen for Seconds per image (15 by default) and the images take turns until the video ends. Before, five images under a 108-minute narration stayed up about 21 minutes each.
- Zoom (%) (22.5 by default) and Motion: Zoom in and out, Pan across, Mix of both, or Still.
- Silence at start and end (2 s by default), so narration doesn't start or stop abruptly.
- Rendering encodes each distinct image clip once and joins them, in bounded memory.
- Caption timing runs on the native speech engine with up to 8 threads: 3.6 minutes instead of 24.3 for a 109-minute narration.
- Cuts: Every N seconds, or Follow the narration (the default for new projects), which cuts after a sentence and at every chapter.
- Transitions: Cut, Crossfade, Fade through black, Slide, Wipe.
- A Look: vignette, film grain, colour grade, atmosphere (embers, dust, fog) and chapter title cards.
- Add your own video clips as shots, or animate images with Kling, Wan or Seedance models on fal.ai or Replicate, priced in the estimate.
- The Video section's files download from one Download menu next to one Open folder.

### Shorts
- The text model picks self-contained moments of the narration, cut on whole sentences, and each becomes a 1080×1920 short with new 9:16 images and big word-by-word captions.
- 1–10 shorts (3 by default), 15–180 seconds each (60 to 120 by default).
- Each short has a title, a description and hashtags to copy.
- Make this short again, Pick different moments (unchanged clips keep their images), and adjust each clip's start and end by hand.
- Optional background music that dips under the narration, also from Play.
- A new Shorts prompt kind in the Library.

### Document (PDF)
- An optional stage lays the article out as a PDF with a title page, clickable contents, drop caps, a Sources page and the thumbnail as cover. It runs on your machine, with no provider and no charge.
- Library → Documents: edit every PDF setting beside a live preview, and pick themes in Play and Edit project.
- Plain is the built-in theme. Installs that used the old DiceMaster look keep it as a theme of their own.

### YouTube description
- An optional step writes a description with chapter timestamps and hashtags, plus a tags list, ready to copy.
- Chapter times come from the narration's word timing in the finished video, and are checked against YouTube's rules before you see them.
- A new Description prompt kind in the Library.

### Images
- Codex images: pick the model and effort, Ultra included. Up to 30 minutes per image, four at once.
- Establishing image: an optional reference image, made first or uploaded, that every video image, short and optionally the thumbnail follow for consistent characters, palette and style. It works with Codex, OpenAI, Google and fal.ai's FLUX.2 and Nano Banana models.

### Project page
- Five sections: Article, Audio, Images, Video and Document. Research, Sources and Pronunciation are tabs of Article, each with Copy as Markdown.
- The thumbnail is shown large at the top of Images.
- Pronunciations are shared: a new project copies the glossary terms of your other projects, and its own glossary wins.

### Schedules and Play
- A schedule takes a list of topics and runs one per scheduled run, filling a template keyword and the project title.
- The review refreshes itself after an edit, so Start and Queue no longer stay grey.
- Runs use the Library's current prompts, and the project title fills its keywords.

### Backups and notifications
- Export everything saves one .tar with every project and its files, the Library, templates, schedules, drafts, fonts, usage and non-secret settings. Provider keys are never included.
- Import a backup adds to this install without replacing anything. Schedules arrive paused, and importing the same backup twice adds nothing.
- Automatic backups every day at a time you choose, keeping the last few.
- Browser notifications, and an optional Notification URL (ntfy works), when a run finishes, fails or waits for you.

### Errors
- Every error says what failed, why, and what to do, using the names on screen or the exact command.
- A caption failure names the narration chunk that stopped matching its audio, with the words expected and heard.

## In 3.0

### Design
- A new design system across the app and slopify.stream: one palette, type scale, spacing grid and component kit, in dark and light.
- A left rail with six destinations and a channel picker, full width up to 1680 px, and a bottom bar on phones.
- Every setting, option and non-obvious action has an info button that says what it does, when to change it, its default, and what it costs or slows. Press it, or Tab to it and press Enter; Esc closes it, and it works on touch.
- All 406 explanations live in one catalogue, so a control is explained the same way in Play, Edit project, templates and schedules.
- Images and videos sit in proper frames with a lightbox and a real player.
- Library rows show Edit, Duplicate, Use in Play, History and Delete directly, instead of in a menu.
- Ctrl+K opens a command palette for Play and every project action.

### Home and the calendar
- Home shows what needs you, what is running, the next seven days, videos ready to upload, and this week's videos, spend and plan limits.
- Mark uploaded takes a finished video off Ready to upload.
- The calendar shows the next four weeks. Drag a topic to another day or schedule, and queue or reject suggested topics beside it.

### Play
- Play opens on Template and Topic. Everything else is folded into summary rows that open where they need attention.
- More topics are chips under "More videos from the same setup".
- A real style preview: six seconds rendered on your machine with the captions, the Look, transitions and a chapter card, drawn on your establishing image or a cast picture. No provider is called.
- One keyword list shows what each keyword feeds, in Play, Edit project and templates.
- Save as template from Play. Templates never keep one-off values such as the topic you typed.
- The estimate shows the total, with the price of each stage folded under Cost by stage.
- Start counts exactly the videos it will start, and the server refuses a Start that doesn't match the review.

### The project page
- A section rail (Article, Narration, Images, Video, Shorts, YouTube, PDF, Cost, Live, Settings, Checkpoints, History), and on the right the next action, the steps with their times, and the cost so far.
- One action named for its result replaces Resume, Retry stage, Re-run section and Rebuild: Continue the run, Remake 3 outdated images, Approve and render the video.
- Remaking a whole stage sits behind each section's More menu and asks first.
- A Live tab shows the article as it is typed, images as they land and the narration waveform growing.
- The article, research, sources and narration read in a reading view with contents, search and copy.
- Edit the YouTube summary, chapters, hashtags and tags by hand. A regenerated description keeps your edits and offers the new text with Use it, Keep mine and View diff.
- `{{Patreon}}`-style placeholders in descriptions fill from Settings → Channel links.
- Edit project → Images ticks prompts and changes image counts, and says what saving will add and remove.
- Edit project → Inputs → Channel moves a project to another channel.

### Automatic reviews
- A reviewer model checks the article, every image, the narration, the thumbnail and every short, and saves its verdict and reasons beside the item.
- Per stage: Off, Flag only, or Flag and redo. Redo is limited per item (2 by default, 0–5), then the item is kept and flagged.
- Overrule accepts a flagged item. Redo makes it again and reviews it again.
- Claude Code and Codex review pictures. Gemini CLI and OpenRouter review text only, and Play says so.
- Review prompts are a new Library kind. Reviews count in Usage and in the estimate.

### Schedules
- A schedule can carry a series brief and ask an LLM for its next topics when fewer than 10 are queued (you can change the number).
- Topics are queued directly or held for your approval, and never repeat a title the schedule or any project already used, near-duplicates included.
- A Notification URL gets "N new topics are waiting for you".
- Topics can set any keyword, written one per line, as a table, or pasted as YAML or JSON.

### Channels
- A channel page with Brand, Cast, Templates and Schedules. Existing templates and projects move into "My channel", unchanged.
- The brand kit sets caption font and colours, chapter cards, an end screen, intro and outro, document theme, language and ambient sound for its projects.
- Cast members (characters, creatures, places, objects) have aliases, a description and up to four reference pictures. When an image's brief names a member, their pictures go with the request.
- Episode memory: each finished episode leaves a short summary, and a new episode's prompt gets up to five related ones.
- Existing videos: paste titles or import a YouTube Studio CSV. Topic suggestions skip them.
- A CSV import lists its titles with a tick each, so videos of other channels in the same Studio export stay out. "Keep only titles containing…" ticks the titles holding that text, and the filter is remembered per channel for the next CSV.

### Multiple voices
- Formats: Narration (one voice, as before), Audiobook, Podcast, Radio drama and Interview.
- Speakers have a name, role, voice, pace and pronunciations. Audition plays each speaker's first line, priced first and only on click.
- Cast members can have a voice, so recurring hosts and characters sound the same in every episode.
- A Script prompt kind writes speaker turns. An audiobook can instead have its article split into speakers, keeping every word.
- One narration request per turn, using ElevenLabs Text to Dialogue for consecutive eleven_v3 turns. Changing one speaker's voice remakes only their turns.
- Captions carry the speaker. Podcasts and interviews get a speaker panel with portraits from the cast.
- The Video stage can also make an MP3 and an M4B with chapter markers.
- Narration Preparation works for multi-voice runs: each speaker's turns on Inworld TTS-2 get delivery cues (Inworld's bracketed directions and non-verbal sounds) suited to who says them. Captions and word timing keep the clean words.

### Other languages
- A Language picker in Play, Edit project and a channel's brand kit. English projects are unchanged.
- The article, narration preparation, description, chapters, tags, Shorts titles and scripts are written in the project's language. Your prompt itself is never edited.
- Word-by-word captions in Spanish, German, French, Italian, Portuguese, Dutch, Catalan, Polish and Czech, using a free 248 MB multilingual model downloaded on first use. Other languages get sentence-by-sentence captions, and the project page says so.
- Voices list the languages they speak, and Play shows the ones for the project's language.
- Captions fall back to a bundled Noto Sans font when your font lacks the language's letters.

### Level the volume
- Every narration piece (a chunk, an intro or outro part, a speaker's turn) is measured and brought to −20 LUFS by one fixed gain before the pieces are joined, so voices and requests no longer jump in loudness. The Narration section says what it did, for example "the spread was 16.6 LU, now 0.2 LU".
- The long video and the shorts are mastered to −14 LUFS with peaks under −1.5 dBTP. The audio-only WAV, MP3 and M4B are mastered to −18 LUFS with peaks under −3 dBTP. Both are measured on the finished files.
- The ambient bed and shorts music stay ducked under the levelled voice.
- Volume is set in dB from the recommendation, or as a percentage, from −10 dB to +4 dB.
- Settings → General sets the default for new runs (on). Play's Export row and Edit project → Pauses and volume set it per project.
- Turning it on for an existing project makes no speech, word timing, description, shorts pick or review again. Uploaded narration is never re-levelled; only the exports' master reaches it.

### Pauses between sentences
- New runs lengthen the quiet after every sentence to at least 0.4 s (Pause between sentences, 0 to 2 s), with an optional longer Pause between paragraphs (0 by default).
- Set them on Play's Export row or in Edit project → Pauses and volume.
- A pause is added in the middle of the quiet a sentence already ends with. It never shortens a pause or cuts a word, and word timing runs afterwards, so captions, cuts, shorts, chapters and M4B marks stay in sync.
- In multi-voice scripts the pauses go inside a speaker's turn. The gap between turns stays separate.

### Narration that reads like speech
- Describe tables and figures in the narration (on by default for new runs and templates): tables, figures, Mermaid and ASCII diagrams, equations and code blocks are each described in a short spoken passage by the run's text model, instead of being skipped or read cell by cell. Leave code out drops code instead.
- Set it in Play under Audio Advanced, or in Edit project → Providers. Each described block adds one text-model call to the estimate.
- Lists are spoken as sentences ("First, … Then, … Finally, …"), blockquotes as quotes, and footnote markers and bare URLs are no longer spoken.
- Descriptions are written in the project's language. Claude Code and Codex are shown a figure's own picture when the project has it as an uploaded image.
- Captions follow what is spoken. The article, PDF and reading view keep the real table.
- Show tables and figures on screen (on by default with describing): each described block is drawn as a card on your machine, in the brand kit's title font and colour, and shown in the video exactly while its description is spoken. Shorts show it upright. Cards are listed under Images → From the article.

### Narration
- Library → Aliases: words the narrator says differently from how they are written, such as `Dr.` as `Doctor`. The article and captions keep the written words.
- An optional ambient bed (rain, fire, wind or your own file) under the long video's narration, ducked under the voice.
- More images for long videos: "N images per hour" or "one every N minutes", up to 240 per run.
- A glossary entry narration can't use (sounds English doesn't have, ARPAbet, a word-count mismatch, a second different pronunciation, an unreadable line) is skipped and read as ordinary text, instead of blocking every narration request. The Pronunciation tab and the rebuild review list skipped entries by number and reason.
- Table glossaries whose IPA column has no slashes are read as IPA.

### YouTube Studio
- Prepare upload lists what Studio asks for, in its order, each with Copy and a done tick, for the video and each short.
- The Slopify Studio extension for Chrome, Chromium and Firefox fills the upload dialog after you drop the video in: title, description, thumbnails, playlist, audience and tags. It never presses Publish, and any field it can't fill gets its own message with the text copied.
- The extension follows Studio's live Details page as of 27 September 2026: title and description are typed with real line breaks, tags become chips one by one, and three thumbnails go to A/B Testing when it can be filled. The playlist list and the A/B Testing dialog are not verified yet.
- AI use is answered Yes only in YouTube's three cases, and Prepare upload says which in YouTube's words. An AI narrator or AI pictures alone are No.
- The three cases are marked by a Real person switch per voice (Settings → Voices), a Draws photorealistic pictures switch per image prompt (Library → Prompts), and The uploaded clips are real footage in Prepare upload. All are off by default, and a channel's Always Yes or Always No still wins.
- Generated thumbnails can come in threes, for YouTube's Test & compare.

### Run cost and providers
- A Run cost tab on every project: the actual cost per stage and model, with tokens, narration characters, images, video seconds and time per stage.
- Claude Code, Codex and Gemini runs show $0 on your plan with the API equivalent, and the share of the 5-hour and weekly limit used where the CLI reports it.
- When a CLI's plan limit runs out, the step waits ("Waiting for Codex limits (resets at 14:00)") and carries on by itself after the reset, also after a restart.
- Every key field has setup steps and a Test button. Check all checks every CLI, key and chosen model.
- The model catalogue updates itself at start and daily. Retired models are flagged, with Switch and Switch all for the templates, schedules and projects that use them.
- On first launch Slopify finds the installed CLIs and says you can make a video with no API keys.

### When something goes wrong
- Rate limits, timeouts and dropped connections wait about 2, 4, 8 and 16 minutes and run again by themselves, also across restarts.
- A failed step stops only the steps that need it. A run whose video was made ends "done with problems".
- Failed steps show the button that fixes them: Sign in to Codex, open the provider's settings, Free space, Switch model, Edit prompt, or Soften and retry for a refused image prompt.
- Updating while a project runs waits for it and installs afterwards, natively and in Docker.

### First five minutes
- A first-run screen shows the CLIs found, Make a 60-second short, the samples and four starter packs (Sleep lore, True crime, History, Science explainers).
- Three read-only samples: The Library of Alexandria, an audiobook of the opening of The Wind in the Willows (a narrator and two character voices), and a two-host podcast on the Antikythera mechanism (a speaker panel with painted portraits). Make my own copy turns one into an ordinary project with nothing to rebuild.
- The samples are spoken by Inworld voices and painted with the Codex CLI, with the volume levelled and pauses between sentences. Each video measures about −14 LUFS, and the demos' MP3 and M4B about −18 LUFS.
- Settings → Backup & storage → Restore samples brings all three back.
- A "What's new in 3.0" tour on the first launch after updating, and an updated tutorial.

### Patch notes
- Settings → Patch notes shows the newest release's notes in the reading view, with earlier versions folded under Earlier versions. Also in Ctrl+K (Show patch notes) and Settings → About.
- After an update, the new version's notes open once by themselves. A fresh install doesn't see them. On a major update the What's new tour shows instead, and its last step has Read the full patch notes.

### Library
- Prompts and intros and outros keep every version, with a side-by-side word diff, restore, and a list of what uses them.

### Storage, install and upkeep
- New installs keep projects, backups and exports in Documents/Slopify. The database, settings, models and logs stay in the hidden data folder.
- Existing installs keep their folders. Settings → Backup & storage → Your files has Open folder, Move to Documents/Slopify and Choose another folder. Every file is copied and checked before Slopify switches, and the old folder is kept.
- New Docker installs mount Documents/Slopify too. `update --docker --projects-dir documents` moves an existing one there.
- Trash: deleted projects, prompts, intros and outros, templates and schedules stay 30 days in Settings → Trash.
- Keep outputs only frees a finished project's working files. Disk space shows the trash separately.
- Settings → General → Start Slopify when I log in uses your account's own start-up list, with no administrator password. Slopify starts quietly, without a browser tab, and turning it off removes exactly what it added. The first-run screen and the terminal ask once.
- In Docker the screen says whether Docker itself starts at login, and where to turn that on. Slopify never changes Docker's settings.
- Docker runs from one compose file: `npx @gentbajko/slopify --docker` installs, `npx @gentbajko/slopify@latest update` updates, natively or in Docker.
- Docker updates wait for running work, keep a checked recovery copy of your data, and put the previous version back by themselves if the new one doesn't answer. Installs from 2.5.0 or earlier are taken over in place, with the same data and project folder.
- Run cost and plan limits work through the Docker host bridge too.
- Backups carry channels, cast pictures, episode memories, prompt history, reviews, run cost and the trash.

### Licence
- Slopify is now licensed under Apache-2.0, with a NOTICE crediting the author. The licence covers the code, not the Slopify name or logo.

## Fixes

- Resuming a project made before 1.6 after its narration text changed no longer fails with a database constraint error.
- Videos with captions no longer fail on "subtitle model download was interrupted". The caption model now ships with the Docker image and downloads in the background on npm installs.
- A caption-model lock left by a crash no longer blocks captions in Docker.
- Undoing an edit or restoring a version reuses finished work instead of queueing it again.
- Waiting steps are no longer duplicated while narration chunks arrive. Held work says "Waiting for Resume".
- Old duplicate queued steps are removed.
- Cancelling a CLI step always stops the CLI.
- Backup imports no longer time out on slow connections.
- Docker updates keep only the newest recovery copy of your data.
- Edit project says "unsaved" only when something changed.
- The rebuild review shows CLI work as "$0 on your plan" with the API equivalent, instead of an unknown cost.
- Edited captions keep their speaker on multi-voice runs.
- Multi-word narration aliases no longer leave double spaces.
- API starts take cast speakers' current voices, as Play, batches and schedules do.
- Extra videos no longer get "2" or "3" appended to keyword titles.
- A multi-voice run whose script or speaker split was written by the text model no longer fails to store that step.
- The rebuild review folds its per-item list into a summary with counts, shows every blocked item's reason above Start rebuild, and says next to the button what still keeps it off.