create-photo-flipbook-ui: a Codex skill that builds photo flipbook sites

What create-photo-flipbook-ui actually produces

This is a Codex skill, not an application you open in a browser tab. The README describes it as a way to turn photographs into expressive photobooks using a bundled 2D Book runtime kept in assets/html/ inside the skill folder. The deliverable of a run is a working local URL showing the finished book. The skill itself defines outcomes and supplies tools; the README is explicit that the agent chooses the process.

The audience is narrow in a useful way. You need Codex with its skill installer, a folder of photographs, and a willingness to let an agent make art-direction decisions. If you want to upload images to a website and drag pages around, this is the wrong shape of tool: there is no hosted editor here, and the repository ships a starter with no sample photographs.

Three stages: understand the photos, define the style, build the UI

The README lays out one workflow in three stages. First the agent examines the photos for form, content and themes, with semantic search and contact sheets available as aids. Second it defines style and design, and the README lists Pinterest, the photo-skill catalogue and image generation as options. Third it builds the interface with the built-in 2D HTML runtime and hands back a local URL.

The default path matters more than the optional one. When the user has no preferred style, the skill preserves the original photographs and applies the bundled default photobook style, documented in references/default-style.md. That style is described as concise text guidance plus Source Serif 4, paper and cloth textures, and reusable page styles in the runtime. So the fallback is conservative: your images are not regenerated, and the visual identity comes from typography and texture rather than from a model.

Art direction and sequencing live in workflow instructions rather than in code, which is a deliberate split. The reusable pieces shipped with the skill are the photo-library adapter, an ordered contact-sheet renderer, a Pinterest helper and the 2D book runtime. Model weights and photo libraries are not part of the package.

Installing the skill and running a first photobook

The README gives a single install command that pulls the skill folder from GitHub into the Codex skills system. It calls the installer script that ships with Codex, so the path below has to exist on your machine before the command will do anything.

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
  --repo HaichaoLihc/create-photo-flipbook-ui \
  --path skills/create-photo-flipbook-ui \
  --ref main

After that, invocation is a text instruction to the agent rather than a CLI subcommand. The README's example is a single line naming the skill and the task.

Use $create-photo-flipbook-ui to turn these photographs into photobooks.

What you should see at the end is a working local URL for the book. The repository also ships a validation script for its own contents, which is useful after a fresh clone or before you report a problem.

python3 tests/validate_repo.py

One dependency is declared rather than installed for you: the photo-search MCP, used for library search, ranked contact sheets and larger previews. The README notes the MCP itself is read-only. The bundled scripts/photo_library.py adapter adds folder-specific cached indexes and semantic searches by reusing an installed photo-search engine's index.py and PhotoSearchEngine, and it does not replace the connected MCP's global index. To use it you install the engine and its Python requirements separately, set PHOTO_SEARCH_ENGINE_DIR to its location, and use that engine's Python interpreter. Index caches are kept outside the skill and source folders, and new or changed images are embedded incrementally while unchanged ones are reused.

The optional UI collections are not what the skill installs

The repository contains more than the skill, and the README is careful about the distinction. The templates in ui-collections/ sit outside the skill folder and are described as optional resources for people to copy and customize. They are not installed with the skill. The skill uses its own bundled 2D runtime; the library, the 3D readers, the film archive and the spatial galleries are separate alternatives.

That matters when you evaluate the screenshots. The Library template is a reorderable shelf with three empty mock books. 3D Book 1 is a React and WebGL reader with curved pages on a dark stage, and the README states the adapted code in ui-collections/3d-book-1/ is excluded from the MIT grant unless expressly stated otherwise. 3D Book 2 uses Three.js and Quick FlipBook with a light stage and soft shadows. Card Gallery, Image Atlas and Photo Ring are vanilla HTML, CSS and JavaScript with local SVG placeholders and fictional sample content, no personal photographs and no external image requests, and they need no install or build step.

Negative Sleeves, the film archive UI, is separate again: photo-free, with page-corner previews, draggable film strips, reversible single-frame inspection and optional continuously paginated semantic search. Importing a local photo folder uses its Python builder, and generated books and photographs stay Git-ignored.

Where the workflow breaks down

The clearest limitation is the dependency split. The skill declares the photo-search MCP, but the README does not install it for you, and the bundled adapter expects a separately installed engine, a PHOTO_SEARCH_ENGINE_DIR pointing at it, and that engine's own Python interpreter. If you skip all of that, semantic search, ranked contact sheets and larger previews are unavailable, and the agent falls back to whatever it can infer from the images themselves.

Style control is the second constraint. The default style is intentionally conservative: original photos preserved, Source Serif 4, paper and cloth textures. If you want a heavily designed result, you are relying on the agent's choices, and the README's own framing is that the skill defines outcomes while the agent chooses the process. That is a different contract from a template with fixed slots.

The repository is also not a self-contained product. The starter contains no sample photographs, model weights and photo libraries are excluded from the package, and the validation script checks the repository rather than your generated book. And the licence boundary is uneven: third-party components keep their own licences, and the adapted code in ui-collections/3d-book-1/ plus all photographs, videos and other media are excluded from MIT unless expressly stated otherwise.

How this differs from a browser flipbook service

FlipAnim and similar web tools are the obvious comparison, and the difference is architectural rather than cosmetic. A hosted flipbook service keeps your frames on its servers, gives you a drawing or upload interface, and returns a page on its domain. create-photo-flipbook-ui does the opposite: the skill runs inside Codex, the 2D runtime is bundled in the repository, the output is a local URL, and the photographs never need to leave your machine. The photo-search MCP is described as read-only, which fits that model.

A second alternative is the repository's own ui-collections templates. If you already know the layout you want, copying the 3D Book 2 folder and editing its HTML gives you a fixed reader with no agent in the loop, at the cost of doing the sequencing yourself. The skill's value is the middle stage: reading the photos, proposing a coherent order and style, then building the book. If you do not want an agent making those decisions, the templates are the cheaper path.

Maintenance, licence and upgrade cost

The last push to the default branch was on 2026-09-16, one day before this writing, and the repository is not archived. Three releases are listed: v0.1.0 on 2026-08-24, v0.2.0 on 2026-08-25 and v0.3.0 on 2026-09-08, the last described as the photobook workflow and default style. The cadence across those three releases is roughly two weeks, though nothing in the README commits the project to keeping it.

Upgrade cost is low if you treat the skill as the unit. The install command takes a --ref argument, so pulling a newer revision means re-running it against main or a tag. The parts you should expect to touch yourself are the separately installed photo-search engine and its PHOTO_SEARCH_ENGINE_DIR setting, which the skill does not manage. Cached indexes live outside the skill and source folders, so they survive a reinstall.

Licensing is MIT for the original project code and the installable skill, with third-party components keeping their own licences. The exclusions are the ones worth reading twice: the adapted code in ui-collections/3d-book-1/ and all photographs, videos and other media, unless expressly stated otherwise. This is a description of the licence file, not legal advice; if you plan to redistribute a generated book or the 3D Book 1 code, read LICENSE and the individual template READMEs.

Editorial conclusion

Adopt it if you already run Codex skills and want a local, MIT-licensed flipbook built from your own images, and if you are willing to install the separate photo-search engine for semantic library features. Do not adopt it if you want a hosted drag-and-drop editor, since the skill produces files and a local URL rather than a service. Before you start, verify that install-skill-from-github.py exists at the path the README gives, and check whether the photo-search MCP is connected if you plan to use contact sheets and ranked search.

Frequently asked questions

Official sources

1. HaichaoLihc/create-photo-flipbook-ui on GitHub
2. Issues
3. License: MIT
4. README
5. Releases