Manga Colorizer AI — Batch Comic Page Colorization

Turn black-and-white manga, manhwa, and comic pages into fully colorized images in one batch run — line art preserved, already-colored pages skipped for free, and every finished page delivered as a public URL.

Manga Colorizer AI takes a list of publicly accessible page image URLs (JPG, PNG, or WebP), colorizes each page with AI, and stores the results in the run's key-value store with a public URL recorded in the dataset. Point it at a whole chapter and get back one clean row per page with before/after links.

Who uses it

User	What they get
Scanlation & fan-translation teams	Colorized pages ready for lettering and localization
Webtoon & comic publishers	Fast color passes on line-art pages before final polish
Content creators & social media managers	Colored manga panels for posts, thumbnails, and reels
AI & data pipeline builders	A batch colorization step with structured dataset output and public URLs
Collectors & archivists	Colorized editions of public-domain manga and woodblock pages

What it does

• 🎨 AI colorization that preserves line art — the model adds color only; every inked line, screentone, and panel border stays untouched.
• 📚 Whole-chapter batching — pass every page URL of a chapter in one input; the run processes them all with controlled concurrency.
• 🆓 Free default provider — works out of the box with a public Hugging Face colorization space, no API key required. A pluggable provider system also supports Replicate with your own token.
• 🖼️ Style presets — anime, manga-flat, comic, and watercolor grading, forwarded to prompt-aware providers and recorded on every row.
• 🔍 Optional 2x upscale — double the resolution of every output page with a high-quality Lanczos filter.
• 🗜️ Output format choice — save colorized pages as png, webp, or jpg.
• 💸 Fair billing — pages that already contain color are auto-detected, passed through, and never charged. You pay only for pages actually colorized.
• ⚙️ Concurrency control — tune maxConcurrency (default 3) to match your throughput and the backend's limits.
• 🔗 Before/after URLs on every row — each dataset item links the source page and the colorized result, plus status, dimensions, style, and language metadata.
• 🌐 Language-aware metadata — set the originalLanguage (Japanese, Korean, Chinese, English, other) and targetLanguages so downstream OCR/translation steps read text regions correctly.

Use cases

Colorize a whole manga chapter

Collect the page URLs of a chapter (from your CMS, a MangaDex-style listing, or your own storage), paste them into imageUrls, and run. The Actor colorizes pages three at a time by default and hands you a dataset of public result URLs in the same order.

Pre-color line art for webtoon production

Studios use flat color passes to speed up production. Run your inked pages through the manga-flat preset to get a consistent base coat, then ship the public URLs straight to your artists.

Restore and color public-domain manga

Classics like Hokusai's manga sketchbooks are public domain but printed in monochrome. Colorize them for new editions, prints, and social content — already-colored plates are detected and skipped for free.

Feed a translation pipeline

Every dataset row carries originalLanguage and targetLanguages, so an OCR/translation actor can pick up the colorized pages and localize text bubbles without extra bookkeeping.

Why choose this scraper

Reason	Detail
Works with zero setup	Free Hugging Face provider needs no API key, no account, no proxy
Honest pricing	$0.10 per colorized page — and already-colored pages are detected and never charged
Batch-first design	Whole chapters in one run, with before/after URLs in every row
Production controls	Concurrency limit, output format, 2x upscale, style presets
Pluggable backend	Switch to Replicate with your own token, or point at your own Hugging Face space for guaranteed capacity
Structured output	Dataset rows + public key-value store URLs, ready for pipelines and integrations

How it compares

	Manga Colorizer AI (this Actor)	parseforge / modern-manga-colorizer	muhammadafzal / manga-colorizer	jungle_synthesizer / ai-manga-image-colorizer
Price per page	$0.10	$0.45	Per-run compute	Per-run compute
Already-colored pages	Skipped, never charged	Charged	Charged	Charged
Free no-key mode	✅ Built in	❌	❌	❌
Style presets	4 presets	❌	❌	❌
2x upscale toggle	✅	❌	❌	❌
Output format choice	png / webp / jpg	❌	❌	❌
Before/after URLs per row	✅	Partial	❌	❌
Concurrency control	✅	❌	❌	❌
Store traction (Oct 2026)	New	705 total users, ~1 monthly active	6 total users	5 total users

Configure the run

Minimal: colorize a chapter with the free provider

{
  "imageUrls": [
    "https://thumb.wikimedia.org/wikipedia/commons/thumb/c/c6/Archers_%28Kyujutsu%29_Page_7_by_Katsushika_Hokusai_1817.png/960px-Archers_%28Kyujutsu%29_Page_7_by_Katsushika_Hokusai_1817.png",
    "https://thumb.wikimedia.org/wikipedia/commons/thumb/f/f6/Archers_makiwara_straw_target_%28Kyujutsu%29_Page_9_by_Katsushika_Hokusai_1817.png/960px-Archers_makiwara_straw_target_%28Kyujutsu%29_Page_9_by_Katsushika_Hokusai_1817.png"
  ]
}

Full control: watercolor style, 2x upscale, WebP output, Replicate provider

{
  "imageUrls": [
    "https://example.com/chapter-12/page-001.png",
    "https://example.com/chapter-12/page-002.png",
    "https://example.com/chapter-12/page-003.png"
  ],
  "originalLanguage": "korean",
  "targetLanguages": ["english", "spanish"],
  "stylePreset": "watercolor",
  "outputFormat": "webp",
  "upscale2x": true,
  "maxConcurrency": 5,
  "provider": "replicate",
  "apiToken": "YOUR_REPLICATE_TOKEN",
  "replicateModel": "cjwbw/ddcolor"
}

Input fields

Field	Type	Required	Default	Description
imageUrls	string array	✅	—	Public URLs of manga/comic pages (JPG, PNG, WebP).
originalLanguage	enum	No	japanese	Language of the text on the source pages.
targetLanguages	string array	No	["english"]	Languages you plan to localize into; stored per row.
stylePreset	enum	No	anime	anime, manga-flat, comic, or watercolor.
outputFormat	enum	No	png	png, webp, or jpg.
upscale2x	boolean	No	false	Double output resolution (Lanczos).
maxConcurrency	integer	No	3	Pages processed in parallel (1–10).
provider	enum	No	huggingface-space	Free Hugging Face space, or replicate with your token.
hfSpaceUrl	string	No	Public MangaColorization space	Point at your own duplicate space for guaranteed capacity.
apiToken	string	No	—	Required only for the replicate provider or private spaces.
replicateModel	string	No	cjwbw/ddcolor	Any Replicate colorization model (owner/name).

Output

Every page produces one dataset row with before/after URLs:

{
  "pageNumber": 1,
  "sourceUrl": "https://example.com/chapter-12/page-001.png",
  "colorizedUrl": "https://api.apify.com/v2/key-value-stores/STORE_ID/records/page-0001.png",
  "status": "colorized",
  "alreadyColored": false,
  "charged": true,
  "stylePreset": "anime",
  "outputFormat": "png",
  "upscaled2x": false,
  "width": 960,
  "height": 1316,
  "provider": "huggingface-space",
  "originalLanguage": "japanese",
  "targetLanguages": ["english"],
  "meanSaturation": 0.012,
  "error": null,
  "processedAt": "2026-10-04T09:30:00.000Z"
}

• status is colorized, already-colored, or failed.
• The colorized image binary lives in the run's key-value store (page-0001.png, page-0002.png, …) and colorizedUrl is its public link.
• Already-colored pages are stored as-is (converted to your chosen format) with charged: false.
• Failed pages appear with an error message so one bad URL never kills a chapter run.

Pricing

Pay per result — you are billed only for pages actually colorized:

Event	Price
Actor start	$0.001
Colorized page	$0.10
Already-colored page (auto-detected)	$0.00 — free

Pages colorized	Cost
10 pages	$1.00
100 pages	$10.00
1,000 pages	$100.00

That is 78% cheaper per page than parseforge/modern-manga-colorizer's $0.45 — and their actor charges even for pages that were already in color.

Free users

On the Apify Free plan you can preview the Actor with a 10-result free run — enough to colorize a pilot chapter and check quality before paying anything. The default provider needs no API key, so the preview works with zero setup.

Run it

1. Open the Actor in the Apify Console and click Try for free.
2. Paste your page URLs into imageUrls (or keep the prefilled Hokusai sample pages).
3. Pick a style preset, output format, and optional 2x upscale.
4. Click Start, then open the Dataset tab for before/after links or the Key-value store tab to download the images.
5. Prefer code? Call it via the Apify API or the JavaScript/Python SDK with the same JSON input.

Use with MCP (AI agents)

This Actor is MCP-ready. Add the Apify MCP server (https://mcp.apify.com) to Claude, Cursor, or any MCP client and ask:

"Use the manga-colorizer Actor to colorize these 24 pages of chapter 12, anime style, webp output."

The agent passes your imageUrls and options as the input JSON and returns the dataset rows with public colorized URLs.

Troubleshooting

Q: The run failed with "The replicate provider needs an API token." You selected provider: replicate without a token. Add your Replicate API token to apiToken, or switch back to the default huggingface-space provider, which needs no key.

Q: A page row says failed with an HTTP error. The source URL was unreachable or did not return an image. Make sure every URL is publicly accessible (open it in an incognito window) and points directly to a JPG/PNG/WebP file, not an HTML page.

Q: The colorization timed out or returned a 503. The free Hugging Face space sleeps when idle and rate-limits heavy traffic. The Actor retries with backoff automatically; for guaranteed capacity, duplicate the space to your own Hugging Face account and set hfSpaceUrl to your copy (plus apiToken if private).

Q: My sepia/aged-paper scans were colorized instead of skipped. That is intended — aged paper is monochrome, so the pages are treated as black-and-white and colorized. Only pages with genuinely varied colors are detected as already-colored.

Q: Can I use my own colorization model? Yes. Set provider: replicate, add your apiToken, and put any owner/name colorization model into replicateModel. Your style preset is forwarded as the prompt for prompt-aware models.

FAQ

Question	Answer
Do I need an API key?	No — the default Hugging Face provider is free and keyless.
What image types are supported?	JPG, PNG, and WebP inputs; PNG, WebP, or JPG outputs.
How many pages per run?	A whole chapter — the input is an array, processed with maxConcurrency parallelism.
What counts as a chargeable result?	Only pages that were actually colorized. Already-colored pages and failed pages are never charged.
Does it preserve line art?	Yes — the model is trained for line-art colorization; ink lines, panel borders, and screentones are preserved.
Where do the images go?	The run's key-value store, with a public URL in each dataset row.
Can I run it from an AI agent?	Yes — via the Apify MCP server or the Apify API/SDK.
What languages can it read?	Set originalLanguage (Japanese default) and targetLanguages; the metadata flows through to your OCR/translation steps.