RedInk

Quick Verdict

RedInk is for creators who publish image-and-text posts daily and lose most of their time to layout and cover art. Its scope is deliberately narrow: type one sentence, get a multi-page outline, then render the cover and each content page from that outline, with the title, body, and tags written alongside. The capability that makes this work is Nano Banana Pro’s text rendering (gemini-3-pro-image-preview) — accurate on-image typography is exactly where ordinary text-to-image models fail, and it is the reason a generated cover is publishable rather than a blurry approximation.

Two things to weigh first. It is self-hosted, not SaaS: one Docker command starts the service, you supply the API keys, and image generation cost tracks your own API usage. And the license is non-commercial — the README carries a CC BY-NC-SA 4.0 badge — so brand accounts, agencies, and anyone embedding it in a commercial product need to settle permission first. For an individual creator willing to configure keys, it is far faster than manual layout.

Best For

• Individual Xiaohongshu creators: Publishing regularly, with layout and covers as the bottleneck
• Multi-platform content operators: Need consistent cover and content-page sets
• Educational and explainer accounts: Outline-driven flow suits splitting long posts into card sequences
• Self-hosters: Happy to run Docker and bring your own keys in exchange for no subscription
• Not ideal for: people who will not touch deployment or API keys; brand accounts and agencies that need commercial licensing; designers who want pixel-level control — this is templated batch generation, not a design tool

Key Features

• One sentence to outline: A topic produces a multi-page outline (defaults to 5 pages when unspecified), with per-page editing and per-page custom descriptions — the project recommends leaning on those descriptions to control style. Generation typically takes 15–30 seconds
• Dedicated cover step: The cover is generated as its own stage, aimed at matching your personal style with accurate text and coherent visuals, because the cover drives click-through
• Batch content pages: Pages render in batch with up to 15 concurrent by default, and any single page can be regenerated without redoing the set
• Captions included: Titles, body copy, and tags are generated too, so you get the post rather than just images
• Swappable providers: Text works with the native Google Gemini API or any OpenAI-compatible endpoint; images work with Gemini image models or an OpenAI-compatible image API. Configure via the settings page or text_providers.yaml / image_providers.yaml
• Concurrency switch: Sequential by default to suit rate-limited accounts; enable concurrency in settings for speed, up to 15 at once
• History and recovery: Records persist, and refreshes or retries reuse existing images instead of spending upstream quota again

Use Cases

• Explainer posts: Topic → 5-page outline → cover plus four content pages plus captions, drafted within half an hour
• Product reviews: Per-page descriptions control information hierarchy while keeping the series visually consistent
• Series production: Generate multiple posts in one style without re-tuning each time
• Cross-platform reuse: Covers also work for other portrait or square placements
• Reworking old posts: Split an existing long article into an outline and re-render it for redistribution

Pricing

The software is free. The Docker image histonemax/redink starts with a single command — docker run -d -p 12398:12398 -v ./history:/app/history -v ./output:/app/output histonemax/redink:latest — then open http://localhost:12398 and enter keys on the settings page. docker-compose and local development are also supported (Python 3.11+, Node 18+, pnpm, uv, with one-click start scripts for macOS, Linux, and Windows). The image ships without any API keys.

Real cost sits in the image API. Accurate cover typography depends on a high-quality image model like Nano Banana Pro, and a five-page post means five or more image calls, which is the dominant expense at volume. The project also warns about quotas: keep high concurrency off on GCP trial accounts or you will trigger rate limits and failed generations. The current version is v1.4.3 (2026-06-30), which is also the latest commit — a noticeably slower cadence than earlier in the year, so factor maintenance risk into your decision.

The license deserves its own note: the README carries a CC BY-NC-SA 4.0 badge (attribution, non-commercial, share-alike), and GitHub does not recognize it as a standard open-source license. Personal non-commercial use is fine; commercial use needs the author’s confirmation.

Pros

• Tightly scoped: covers the full outline → cover → content pages → captions loop for one real format
• Uses Nano Banana Pro’s text rendering to solve the long-standing problem of unusable Chinese titles in generated images
• Single pages can be regenerated, saving both quota and time
• Providers are swappable for both text and images, avoiding vendor lock-in
• One-command Docker deployment with data and history staying local
• Caption generation is built in, removing one tool switch

Cons

• Non-commercial license (CC BY-NC-SA 4.0) requires separate permission for commercial or agency use
• You supply image API keys, and cost at volume tracks image model pricing
• Defaults depend on Google models, so users in mainland China need to configure alternatives
• Image generation is slow and the page must stay open, which suits batch work rather than instant feedback
• Single-maintainer project with its last update on 2026-06-30, leaving long-term upkeep uncertain
• Style is driven by the model and prompts, with less precise layout control than a design tool

Alternatives

FAQ

Is RedInk free?

The software is free and self-hosted, but you pay for the text and image model APIs. A five-page post means at least five image calls, which is the main cost at volume.

Can I use it for commercial content?

Not without sorting out permission. The README marks it CC BY-NC-SA 4.0 (non-commercial), and GitHub does not recognize it as a standard open-source license. Personal non-commercial work is fine; brand accounts, agencies, and product embedding should confirm with the author.

Do I have to use Gemini and Nano Banana Pro?

No, but they are the defaults, and cover text quality depends heavily on Nano Banana Pro. Both text and images can point at OpenAI-compatible endpoints (v1.4.2 fixed connection tests for chat-based endpoints such as Doubao and Volcengine), so test typography quality after switching.

Does it work well for users in China?

Deployment does — the service runs locally with a Chinese UI. The variable is the model side: the official Google APIs require the right network and credentials, otherwise reconfigure to a locally reachable OpenAI-compatible service and expect to re-tune style and text rendering.

How long does one post take?

Outlines usually take 15–30 seconds; images are noticeably slower, and the project warns against leaving the page during generation. Sequential rendering is the default for rate-limited accounts, with up to 15 parallel images available if your API supports it.

What if one page comes out badly?

Regenerate just that page — no need to redo the set, and history reuses already-generated images so quota is not spent twice. For more control, write a per-page custom description at the outline stage before generating.

Bottom Line

RedInk’s value is compressing a scattered toolchain into one pipeline: a sentence goes in, and outline, cover, content pages, and captions come out — with Nano Banana Pro handling the genuinely hard part, readable on-image titles. The costs are equally clear: self-hosting, your own keys, image API spend, a non-commercial license, and a single maintainer. Validate it with a real topic: check whether the outline structure matches how you write, whether the cover text is publishable as-is, and whether five pages of API cost is acceptable. If all three hold, it removes the dullest part of your day. If you need commercial licensing or vendor support, a mature design platform is the safer choice.