Repository navigation
feat(website): index the Models pages and describe them to search engines - #18480
benjcooley wants to merge 1 commit into
Conversation
…ines The model pages were noindex and kept out of the sitemap as the pre-launch gate, and nothing turned that off at launch. Every canonical /models/<slug> is now indexable and in the sitemap; the alias stubs, showcase and checkout routes stay out. Each model page declares a SoftwareApplication with its provider, the breadcrumb to the catalogue, and an Offer priced from the estimate the page shows at the published credit rate, only when that estimate is one number. The catalogue declares an ItemList of every model. The markdown twin now drops status regions, so a loading frame never reads as page content, and the model hero is a section rather than a header so the twin keeps it. Refs FE-2770, closes FE-2026
|
The latest updates on your projects. Learn more about Vercel for GitHub. 1 Skipped Deployment
|
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
🌐 Website E2ETip All tests passed.
🧩 Custom Nodes Ecosystem Matrix: 🟡 Withheld — could not measureExecutes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight This candidate changes nothing the matrix measures, so every expensive stage was skipped. Reproduce locally# build one shard's fixture, then run it
python3 scripts/registry-census/build_matrix.py --shard 1/4
MATRIX_RENDERER=vue MATRIX_VUE=1 \
npx vitest run --config vitest.matrix.config.mts
Full docs: scripts/registry-census/README.md 📦 Bundle Size
🌾 Fallow: ✅ PassedFallow audits the diff for duplication, complexity and dead code. It gates on new findings only, so when the audit completes, a red is something this PR introduced or moved — never inherited backlog. A red can also mean the audit itself did not finish; the details below say which happened. No new findings in the changed files. Reproduce locally# the base the gate compared against
npx fallow audit --root . --quiet --format json \
--changed-since 1c989a236d7f13b54a3566962af63c1bdf2be351
Full docs: ADR-DEVEX-LINT-0015 🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-18480.vercel.app Models is built and gated by PostHog (workshop-enabled), using staging Cloud. This commit: https://website-frontend-fm68wqjgb-comfyui.vercel.app Last updated: 2026-09-23T00:01:18Z for |
Codecov Report✅ All modified and coverable lines are covered by tests. @@ Coverage Diff @@
## bencooley/fe-2770-models-static-content #18480 +/- ##
========================================================================
Coverage 83.35% 83.35%
========================================================================
Files 2957 2957
Lines 146822 146828 +6
Branches 39796 39799 +3
========================================================================
+ Hits 122380 122386 +6
Misses 23201 23201
Partials 1241 1241
Flags with carried forward coverage won't be shown. Click here to find out more.
🚀 New features to boost your workflow:
|
| if (!estimate) return undefined | ||
| const match = /^~?\s*(\d+(?:\.\d+)?)\s*credits(?:\/|$)/.exec(estimate) | ||
| if (!match) return undefined | ||
| return { price: creditsToUsd(Number(match[1])), description: estimate } |
There was a problem hiding this comment.
issue: creditsToUsd rounds to whole cents, so anything under about 1.06 credits prices at $0. Two pages ship that today: luma--photon-flash-1-image-generation--generate-images and luma--photon-flash-1-image-modify--edit-images show "0.57 credits/Run" and publish "offers":{"price":0,...}, which Google shows as Free. Eleven of the 32 offers are also off by more than 1%, up to 11%: bria--remove-image-background at 3.8 credits publishes $0.02 for a true $0.018, and the three seedream-5-pro pages publish $0.05 for ~9.5 credits ($0.045). The '~9.5 credits/Image' → 0.05 row in price-label.test.ts encodes that rounding as correct. schema.org price takes decimals, so could offerForEstimate divide by CREDITS_PER_USD and round to 4 places instead, with a '0.57 credits/Run' row asserting a non-zero price?
| title="Models - Comfy" | ||
| title={title} | ||
| description="Run the world's leading AI models in ComfyUI. Browse every supported model with community workflow templates ready to run." | ||
| pageType="CollectionPage" |
There was a problem hiding this comment.
issue: this turns /models/ into a CollectionPage with an ItemList of all 148 models, but its markdown twin /models.md (and its section of llms-full.txt) is still built from the showcase: it's headed "Grok Imagine in ComfyUI" and links no models. The cause is ALTERNATE_TWIN_SOURCES in integrations/markdown-twins.ts:39 (['models', 'models/showcase']), which predates this PR and made sense while /models/ was the showcase. Now an answer engine reading the twin gets a different page from the one we describe to Google. Could that entry go, so /models.md is built from /models/ itself?
| expect(graph.find((node) => node['@type'] === 'SoftwareApplication')).toEqual( | ||
| expect.objectContaining({ | ||
| name, | ||
| offers: expect.objectContaining({ priceCurrency: 'USD' }) | ||
| }) | ||
| ) |
There was a problem hiding this comment.
nitpick (non-blocking): offers: expect.objectContaining({ priceCurrency: 'USD' }) still passes with price: 0, the exact case above, and the toEqual on the five @types (line 59) fails whenever a node is added even if nothing regressed. Could this assert offers.price is above 0, and use arrayContaining for the types?
|
Tested on a build with #18462 and current main merged in, on this PR's Vercel preview, and on a local build with #18484 on top. The probe walks every model in the build's own catalogue.json, not just the URLs in the description. The indexing side is solid. All 148 model pages are indexable with the right canonical, the sitemap holds exactly those 148 plus both indexes (no alias, showcase, checkout, .json or .md, and nothing prod serves went missing), all 212 alias stubs stay noindex and point at the right page, and no hreflang points at a 404. JSON-LD parses on every page with correct breadcrumbs and SoftwareApplication fields, the ItemList has all 148, and a </script> in a model name can't break out of the script tag.
Not tied to a line: apps/website/README.md:206 still says these pages stay out of sitemaps and markdown exports. And once these pages are indexed, a WORKSHOP_IN_BUILD=0 build would 404 all 148 (the release gate refuses any /models/ entry but index.html). Prod builds with 1, so it's not live, but FE-2770 says a kill switch never unpublishes. Worth changing the gate, or is that switch retired? |
dante01yoon
left a comment
There was a problem hiding this comment.
Two crawler-visible outputs do not match the intended indexed model catalogue yet. I confirmed these against the current head and the catalogue examples in the PR description.
| if (!estimate) return undefined | ||
| const match = /^~?\s*(\d+(?:\.\d+)?)\s*credits(?:\/|$)/.exec(estimate) | ||
| if (!match) return undefined | ||
| return { price: creditsToUsd(Number(match[1])), description: estimate } |
There was a problem hiding this comment.
issue: creditsToUsd rounds to cents before this Offer is emitted, so low-cost models publish price: 0 and are represented to search engines as free. For example, the catalogue's 0.57-credit runs convert below half a cent. Could the structured-data price retain enough precision to remain non-zero (and add that case to the test) rather than reuse the display-price rounding?
| title="Models - Comfy" | ||
| title={title} | ||
| description="Run the world's leading AI models in ComfyUI. Browse every supported model with community workflow templates ready to run." | ||
| pageType="CollectionPage" |
There was a problem hiding this comment.
issue: this describes /models/ as a 148-item catalogue, but the existing ALTERNATE_TWIN_SOURCES mapping still builds /models.md from /models/showcase. Answer engines therefore receive the Grok showcase instead of this catalogue. Could this change also remove/update that alternate twin mapping so the HTML, JSON-LD, and markdown representations agree?
Summary
Every canonical
/models/<slug>is now indexable, in the sitemap, and carries structured data; stacked on #18462 so the first crawl of each URL sees the static content, not the old showcase.Changes
routes/models/[slug].astro:noindexremoved. Adds breadcrumbs (Home › Models › name),mainEntityId, and aSoftwareApplicationnode with the provider as anOrganizationand anOfferpriced from the estimate the page shows, converted at the publishedCREDITS_PER_USDrate, with the credits label as its description. No offer when the estimate is not a single number (variable-priced models) — 32 of 148 pages get one today. The catalogue carries nopriceUsdFrom/creditsPerRunfor any model, so those fields were not usable.routes/models/index.astro:CollectionPagewith breadcrumbs and anItemListof all 148 models.config/indexing.ts: replaces the blanket/models/*sitemap exclusion with an explicit set of alias redirect stubs (routerModelSlugAliases);/models/showcaseand the checkout routes stay excluded viaNOINDEX_PATHNAMES.isWorkshopRoute()is gone with its only caller. Built sitemap: 148 canonical model pages +/models/+/zh-CN/models/; zero aliases, showcase, checkout or JSON routes. Alias stubs keep Astro's ownnoindex.utils/jsonLd.ts:softwareApplicationNodeacceptsproviderNameand a paidoffer.lib/workshop/price-label.ts:offerForEstimate()(pure, tested).lib/markdown-twin.ts: dropsrole="status"regions so the playground's loading frame never reads as page content; the model hero is a<section>rather than a<header>so the twin keeps it. Each model page now gets a.mdtwin with the hero, price, tags and related models.isWorkshopRouteremoved (no other callers).Review Focus
/modelsis already inLOCALE_INVARIANT_ROUTE_KEYS, so detail pages emit no zh-CN alternate — no new guard needed. (Pre-existing and out of scope: that also leaves/models/and/zh-CN/models/unclustered.)14.8 credits/Run→$0.07at 211 credits/USD. If that rate ever moves,creditsUtilis the single source and the offer follows it.indexing.tsnow importsworkshop-browse-content(loaded byastro.config.tsat config time). It is JSON + zod only; the build ran clean.Refs FE-2770 (https://linear.app/comfyorg/issue/FE-2770) · closes FE-2026 (https://linear.app/comfyorg/issue/FE-2026)
Screenshots (if applicable)
No visible change. Crawler-visible additions on
/models/bfl--flux-2-max--generate-images/: