Comic Recap – Automated AI comic review video maker

Open Import from web link and paste the chapter reading pages, one per line. The app opens each page in a hidden window, scrolls screen by screen so lazy-loaded images finish loading, keeps only the images inside the reader and drops logos, banners and ads. The chapter title is taken from the page too.

For a whole series, paste the series page: the app scans the chapter list so you can pick a range, or you can ask it to add the next N chapters from a single link. Besides web links it accepts image folders, ZIP, CBZ and PDF files.

Before analysis, the app flags pages that are not part of the story: duplicates, scanlation banners repeated across chapters, tiny images, and credit, recruitment or promo pages at the start and end of a chapter. Flagged pages stay visible with a reason, and one click brings them back.

Vertical webtoon strips are cut into overlapping slices so the AI never misses a panel. The Pages screen lets you drag to reorder, rotate pages, and select many pages to exclude, restore or re-analyse at once.

A vision model reads every page, splits it into panels in reading order, describes each scene, recognises characters, actions, emotions and settings, and transcribes the speech bubbles. Each panel is scored for importance and visual quality so the video uses the right shots.

Characters are merged into one list for the whole series: name, aliases, appearance, role, relationships and even the natural Vietnamese pronoun to use in narration. In a new chapter the app recognises returning characters instead of treating them as strangers.

Each chapter's plot becomes an event timeline, plus a story memory with the synopsis, world facts and open threads so the next chapter continues the thread. Anything the AI is unsure about is marked low confidence for you to check.

Analysis results are cached per page: close the app midway and press Resume next time; add a chapter and only the new part is analysed. Characters and events can all be edited, merged and locked by hand.

The script is ready as soon as analysis finishes, no extra button: each event becomes a Vietnamese narration paragraph with sentences sized for easy listening, retold in new words rather than copied dialogue. The app estimates the duration of every paragraph and of the whole chapter.

The Check against images button has the vision model compare every paragraph with its own panels and rewrite any that contradict the artwork — sharply reducing made-up plot details.

Every paragraph can be edited directly, or rewritten by the AI following your instruction and the project's storytelling style: neutral, fast-paced YouTube, cinematic, light comedy, serious, literary, or your own channel guidelines.

Lock the paragraphs you like so later runs leave them alone; after editing a paragraph, re-record just that one instead of the whole chapter.

Local AI runs through Ollama on your machine: the qwen2.5vl vision model and the qwen3 writing model, with a lighter set for 8 GB machines. The app detects Ollama, guides the installation and downloads models with a progress bar, and offers four performance profiles from Fast to Quality. No fees, and your comics never leave your machine.

Cloud AI is for smoother, smarter scripts: choose Gemini or any OpenAI-compatible gateway and paste your API key. The key is encrypted with the macOS Keychain or Windows DPAPI. It usually costs only about 1,000–3,000 VND per chapter depending on the provider.

You can use the cloud only for writing and keep image reading local, or let the cloud do both. If the cloud has a network error the app falls back to local AI, and cloud tasks run in parallel for much faster processing.

The sidebar always shows which model is writing, which is reading images, whether each runs locally or in the cloud, and the tokens and estimated cost used so far — no surprise charges.

The app pairs each narration sentence with the best panel: the right event, the right characters, important and good-looking panels first, no repeats, and no text-only panels. Each shot's length follows the real narration audio.

Images are cropped smartly around each panel's focus area. If a crop is off, open the crop editor to drag it, press Auto to let the app redo it, or re-crop the whole chapter at once.

Each shot uses one of seven Ken Burns motions — static, zoom in, zoom out, pan left, right, up, down — with adjustable strength, and one of four transitions: cut, dissolve, fade and slide. Tall images scroll slowly over a blurred background instead of being squashed.

Text masking blurs speech bubbles or all text on the artwork so the video is your retelling, not a copy of the comic. Any shot can be split, deleted, reordered, mirrored, locked or test-rendered on its own.

VieNeu-TTS runs on your machine: download the model once (about 540 MB) and narrate offline, with no per-character fees and no limits. There are 25 Vietnamese male and female voices with Northern, Central and Southern accents in several styles: natural, storytelling, news and audiobook.

Reading speed goes from 0.6× to 1.6×. Audio is trimmed of silence and loudness-normalised so every shot sounds even. A per-series pronunciation dictionary makes sure character names, techniques and unusual terms are read correctly.

Subtitles follow the narration exactly and switch on with one tap. Choose any installed font, size, text colour, outline colour and thickness, shadow, side margins and bottom margin to match your channel.

The text watermark drifts slowly around the frame through eight anchor points, so it cannot be removed by cropping a corner. Customise the text, font, size, colour, outline, opacity and lap speed — building channel recognition and discouraging re-uploads.

Export H.264 MP4 using presets: YouTube 1080p, YouTube 1440p, Shorts/TikTok 9:16, square 1:1 and a 480p draft for quick checks. The 16:9, 9:16 or 1:1 frame is chosen when you set up the project.

Before rendering, the app runs a quality check: shots missing audio or images, shots too short or too long, repeated panels and overly long sentences, plus an AI continuity pass. After export, the Videos screen lets you review each chapter and join several chapters into one FULL video without re-rendering.