Skip to content

fix(website): make product-photo-to-video's combined video run on Cloud - #19284

Closed
benjcooley wants to merge 2 commits into
mainfrom
benjcooley/workshop-product-video-light-stitch
Closed

benjcooley wants to merge 2 commits into
mainfrom
benjcooley/workshop-product-video-light-stitch

Conversation

@benjcooley

@benjcooley benjcooley commented Sep 28, 2026 •

Copy link
Copy Markdown
Collaborator

Summary

product-photo-to-video crashed the Cloud worker (ServiceError: RIP to the server your workflow was running on) on every recent run. Its combined video decoded all three MiniMax clips into frames and batched them, which exceeds the 10 GiB RAM limit for CPU-only Cloud jobs (BE-6855). The combined video now comes from Cloud's core ConcatenateVideo node, which joins VIDEO objects without decoding them.

Changes

  • What:
    • workshop-workflows.jsonl: the Combine Video subgraph (27:*: three GetVideoComponents, two ImageBatch, CreateVideo) is replaced by 27 ConcatenateVideo (videos.video0..2 = the three MiniMax clips, codec auto). SaveVideo 30 saves its output as before, and all four outputs are unchanged.
    • public/workflow-graphs/product-photo-to-video.json and .svg: node 27 becomes the same ConcatenateVideo node with its original links (33, 34, 35 → 41), and the unused subgraph definition is removed. The file keeps its original formatting.

Review Focus

Production Cloud runs with the example input, through workflow_render:

Graph Result
Original (Jim's sweep, 3 repeats) worker died at about 700s
Original (my runs) worker died at about 200–230s twice
Half-size downscale before the stitch died at 199s and 187s
MiniMax at 768p plus the stitch died at 141s
No combined video succeeded in 167s
This PR (ConcatenateVideo) succeeded in 157s, all 4 outputs

Building the combined video decodes MiniMax's clips into frames, and
that step takes the Cloud worker down ("RIP to the server") on every
run, even at 768p with a quarter-size stitch. The three clips alone
complete (167s, all three delivered), so the workflow now outputs just
those. The downloadable graph and preview drop the same Combine Video
subgraph and save node so they stay consistent with the Cloud request.
@benjcooley
benjcooley requested review from a team September 28, 2026 19:59
@vercel

vercel Bot commented Sep 28, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

2 Skipped Deployments
Project Deployment Actions Updated
billing-web Ignored Ignored Sep 28, 2026 8:32pm UTC
billing-web-test Skipped Skipped Sep 28, 2026 8:32pm UTC

Request Review

@coderabbitai

coderabbitai Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

Review in Change Stack →

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Repository: Comfy-Org/ComfyUI_frontend/.coderabbit.yaml

Review profile: ASSERTIVE

Plan: Advanced

Run ID: a5982553-c58d-48cb-b892-ab8239abd2e1

📥 Commits

Reviewing files that changed from the base of the PR and between 7478254 and 7c3f730.

⛔ Files ignored due to path filters (1)
  • apps/website/public/workflow-graphs/product-photo-to-video.svg is excluded by !**/*.svg
📒 Files selected for processing (2)
  • apps/website/public/workflow-graphs/product-photo-to-video.json
  • apps/website/src/content/workshop-workflows.jsonl

Included review availability: This review used your included allowance. Your plan provides up to 8 included reviews per hour; 0 remain after this review.


📝 Walkthrough

Walkthrough

The product photo-to-video workflow now combines the rotation, texture-zoom, and cinematic-pan clips with a ConcatenateVideo node. It retains the combined video and the three individual clip outputs.

Changes

Product photo-to-video workflow

Layer / File(s) Summary
Combine generated video clips
apps/website/public/workflow-graphs/product-photo-to-video.json, apps/website/src/content/workshop-workflows.jsonl
The workflow uses ConcatenateVideo to combine the rotation, texture-zoom, and cinematic-pan clips. The combined video and individual clip outputs remain connected.

Priority: ➖ Normal

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Suggested reviewers: anupreet-comfy

Merge Risk: ⚪ Minimal · up to 7c3f7

The workflow retains its combined and individual video outputs, with no identified issue preventing merge.

Architecture Summary

Architecture risk: 🔵 Low · up to 7c3f7

The change affects 1 system.

Changed systems: apps/website

Architecture concerns
No architecture-level concerns identified.

Review details

Systems and components

  • observed — apps/website (service) was modified; 2 changed files map to changed impact.

Before / after behavior

  • observed — Modified behavior in apps/website/public/workflow-graphs/product-photo-to-video.json: The former subgraph wrapper is now a ConcatenateVideo node with three video inputs connected to the generated videos and a fourth unconnected optional input; its output remains wired to the combined-video saver.
  • observed — Modified behavior in apps/website/public/workflow-graphs/product-photo-to-video.json: The node metadata now identifies the wrapper as ConcatenateVideo, with an auto widget value and the title Concatenate Video, replacing the prior subgraph metadata and empty widget values.
  • observed — Modified behavior in apps/website/src/content/workshop-workflows.jsonl: The combined-video generation now uses ConcatenateVideo with the rotation, texture-zoom, and cinematic-pan clips as inputs, replacing the old chain that extracted their video components, batched their frames, and created a video. The combined output and three individual clip outputs remain.
🚥 Pre-merge checks | ✅ 7
✅ Passed checks (7 passed)
Check name Status Explanation
Linked Issues check ✅ Passed Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check ✅ Passed Check skipped because no linked issues were found for this pull request.
End-To-End Regression Coverage For Fixes ✅ Passed The title and commit subjects contain bug-fix language, but all changed files are under apps/website. The custom check explicitly excludes changes that are exclusively in apps/website, so the `bro…
Website End-To-End Regression Coverage ✅ Passed PASS. The PR changes workflow content and downloadable graph assets, not website application code or an interactive website route. The description gives a concrete reason for omitting a Playwright tes…
Adr Compliance For Entity/Litegraph Changes ✅ Passed PASS. The PR changes only workflow graph/content JSON files. The described diff changes graph-node wiring and video concatenation data; it does not add entity methods or properties, spatial mutations,…
Title check ✅ Passed The title clearly identifies the main change: updating the product-photo-to-video workflow so its combined video runs on Cloud.
Description check ✅ Passed The description is complete and on-topic. It explains the Cloud worker failure, the replacement with ConcatenateVideo, the affected files, review focus, and verification results. Screenshots are not n…
✨ Finishing Touches 💡 1
🧪 Generate unit tests (beta)
  • Commit to this branch
  • Create a new PR
🛠️ Fix failing CI checks 💡
  • Commit to this branch
  • Create a new PR

Comment @coderabbitai help to get the list of available commands.

@benjcooley benjcooley left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👨‍💻 BILL — review of #19284 at 74782547e2.

Verdict: accepted. No code findings. One merge-order point below.

Severity Count
BLOCKER 0
SHOULD FIX 0
TRIVIAL 0

Downloadable graph: well-formed. I checked every link reference in both the old and new files.

  • Root nodes 27 and 30, links 33/34/35/41 and the Combine Video subgraph definition are gone, and nothing references the removed subgraph type.
  • Nodes 5, 18 and 20 each output only to their own SaveVideo (links 16, 18 and 20). No node input or output points at a missing link, and every link's endpoints exist.
  • Keeping last_link_id/last_node_id at 41/30 is correct: ids aren't reused.

Cloud graph and outputs: consistent.

  • The graph goes from 16 to 9 nodes, with no 27:* or 30 left and no dangling [nodeId, slot] references. The input bindings are untouched.
  • outputs is exactly video-2/video-3/video-4 → SaveVideo 16/17/19, and outputs is the only non-graph field that changed.
  • The page matches: the description and example promise "a sequence of showcase clips" / "several complementary showcase shots", never a combined video. Output lookups go through the returned outputId, so nothing hard-codes the dropped video id.

SVG: regenerating with workflowPreviewSvg from the new JSON is byte-identical to the committed file.

Merge order: #19282 (disabling this same workflow) is still OPEN. If it merges after this PR, it switches off the workflow this PR just fixed. Close #19282 when this lands. If #19282 merges first, this PR also needs to remove the workflows/product-photo-to-video availability entry and restore the 30/6 counts.

Verification: 181 tests pass locally on this head (prepare-workflow-previews, workflow catalog, workflow content). CI was still running when I posted.

@benjcooley

Copy link
Copy Markdown
Collaborator Author

Live verification on prod against 7478254, from Jim's sweep with 3 concurrent repeats: 3/3 passed, and every output downloaded.

  • f75e9846-ad72-4f84-8c1d-3c3dfd96bef2: 158s
  • e3b8f4bf-778e-49d4-99d9-c13bbe9456cb: 174s
  • 0698d332-d275-4249-8581-0ecd207050fc: 172s

The free page, graph and media checks passed 5/5. #19282 (the temporary disable) is closed, so the page stays enabled once this merges.

@github-actions

github-actions Bot commented Sep 28, 2026 •

Copy link
Copy Markdown
Contributor

🌾 Fallow: ✅ Passed

Fallow 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 e4229433c30c410a096594c064ba2298c7013398

Full docs: ADR-DEVEX-LINT-0015

🌐 Website E2E

Caution

Some tests failed.

Status ❌ Failed
Report View Report
Screenshot diffs 0
Other failures 1

Warning

1 non-screenshot failure — these require manual review.

🧩 Custom Nodes Ecosystem Matrix: 🟡 Withheld — could not measure

Executes the frontend JS of every registry pack that ships any, against this change, in both renderers. The eight ecosystem-matrix shard checks are a harness self-test; the pack verdict is matrix-verdict.

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

🔗 Website Preview

Website Preview: https://comfy-website-preview-pr-19284.vercel.app

Models is built and gated by PostHog (workshop-enabled), using staging Cloud.

This commit: https://website-frontend-k8g7o23di-comfyui.vercel.app

Last updated: 2026-09-28T20:36:31Z for 7c3f730

@coderabbitai coderabbitai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pre-merge checks failed. Please resolve the failing checks before merging.

@codecov

codecov Bot commented Sep 28, 2026 •

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.

@@            Coverage Diff             @@
##             main   #19284      +/-   ##
==========================================
- Coverage   84.20%   84.17%   -0.04%     
==========================================
  Files        3195     3170      -25     
  Lines      155999   155607     -392     
  Branches    40488    40416      -72     
==========================================
- Hits       131364   130987     -377     
+ Misses      22992    22977      -15     
  Partials     1643     1643              
Flag Coverage Δ
billing-web-unit ?
website-unit 78.22% <ø> (+0.12%) ⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 75 files with indirect coverage changes

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

Cloud's core ConcatenateVideo node joins VIDEO objects without decoding
them into frames (the BE-6855 fix), so the combined video no longer
needs the decode-and-batch stitch that exceeded the 10 GiB CPU-job
limit. The three MiniMax clips feed ConcatenateVideo, and its output is
saved as before. The downloadable graph swaps the Combine Video subgraph
for the same node and keeps its links. Verified on prod: 157s, all four
outputs.
@vercel
vercel Bot temporarily deployed to Preview – billing-web-test September 28, 2026 20:32 Inactive
@benjcooley benjcooley changed the title fix(website): ship product-photo-to-video without the combined video fix(website): make product-photo-to-video's combined video run on Cloud Sep 28, 2026

@benjcooley benjcooley left a comment

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

👨‍💻 BILL — re-review of #19284 at 7c3f7304e4 (the ConcatenateVideo approach, one commit on 74782547e2).

Verdict: accepted. No findings. The manifest and count edits come after #19288, as agreed.

Severity Count
BLOCKER 0
SHOULD FIX 0
TRIVIAL 0

Cloud graph

  • Against main, only two things change: the six 27:* subgraph nodes become one 27 ConcatenateVideo (videos.video0/1/2 = ["5",0]/["18",0]/["20",0], codec: "auto"), and SaveVideo 30 is re-pointed from ["27:27",0] to ["27",0].
  • No references dangle, and outputs is identical to main's (video → 30, video-2/3/4 → 16/17/19).

Downloadable graph

  • Against main, node 27 is the only node that changed, and the links array is identical.
  • Links 33/34/35 target node 27 at slots 0/1/2, which match videos.video0..2. Link 41 (27 → 30) is unchanged.
  • The Combine Video subgraph definition is removed, and no link id dangles in either direction.
  • The autogrow shape matches the frontend. autogrowOrdinalToName (src/core/graph/widgets/dynamicWidgets.ts) builds ${group}.${prefix}${ordinal} for a single-spec template, which gives videos.video0. The prod Cloud run accepted those exact keys, which confirms the group and prefix. Autogrow also keeps one empty slot after the last connected one, so three linked slots plus an empty videos.video3 is what ComfyUI itself serializes. widgets_values: ["auto"] carries the codec.

SVG: regenerating with workflowPreviewSvg is byte-identical (7,296 bytes).

Verification: 181 tests pass locally on this head (prepare-workflow-previews, workflow catalog, workflow content). Prod job 6080c332 delivered 4 outputs in 157s, and JIM's 3× repeat is pending. CI was 5 passed, 27 pending when I posted.

@benjcooley

Copy link
Copy Markdown
Collaborator Author

Live verification on prod against 7c3f730 (ConcatenateVideo), from Jim's sweep with 3 repeats: 3/3 passed. Each run delivered all 4 outputs, including the combined video (206 video/mp4).

  • 56635afd-1456-4838-a1a0-dcd37e12eec4: 191s
  • daff6b57-7d3b-42bf-be72-f3dd45737582: 184s
  • 98babfa5-459c-4e38-a5a2-347b979ba090: 213s

The free page, graph and media checks passed 5/5. My earlier run, 6080c332-410c-4409-98bc-5772276338ac, also succeeded in 157s.

@datadog-official

Copy link
Copy Markdown

Pipelines

❌ Errors

Your PR has failed checks. Please review the issues below and take necessary action before merging.

🚦 2 Pipeline jobs failed

CI: Website E2E | website-e2e-shard (3, 4) — 🔧 Needs a code fix, caused by this PR

View in Datadog · View in GitHub Actions

CI: Website E2E | website-e2e

View in Datadog · View in GitHub Actions

Useful? React with 👍 / 👎

This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 7c3f730 | Docs | View more details | Give us feedback!

@benjcooley

Copy link
Copy Markdown
Collaborator Author

Superseded by #19291, which carries both commits here unchanged (graph, download and preview byte-identical to 7c3f730) together with #19272's fixes, as one PR at Ben's request. The live evidence here (6080c332 plus Jim's 3/3: 56635afd, daff6b57, 98babfa5) applies to #19291's product-photo-to-video graph.

@benjcooley benjcooley closed this Sep 28, 2026
benjcooley added a commit that referenced this pull request Sep 28, 2026
## Summary

Makes every published Cloud workflow run. Combines the remaining
workflow fixes into one PR: #19284 (product-photo-to-video) and #19272
(remove-object, virtual-try-on). #19288's temporary disables are no
longer needed because all three are fixed here. Together with #19251
(already merged), all 30 published workflows pass live on production.

## Changes

- **product-photo-to-video** (Bob's commits from #19284): Cloud lost the
worker at ~700 s decoding the three MiniMax videos to stitch them. The
combined video now uses core `ConcatenateVideo` (joins VIDEO objects
without decoding); the downloadable graph and SVG match.
- **remove-object**: the example now includes a website-owned 1024×1024
mask over the apple
(`public/workflow-inputs/remove-object-apple-mask.png`).
- **virtual-try-on**: the example now includes the template's own
subject (pinned workflow_templates revision).
- **Tests**: every published workflow's example must validate in the
shared form; a browser test opens remove-object and virtual-try-on and
asserts their inputs are pre-filled.

## Review Focus

Live evidence on production (`test:workflow-pages --execute`):
- product-photo-to-video at `7c3f7304e4`: 3/3 passed, all 4 outputs
incl. the combined video — runs `56635afd…` (191 s), `daff6b57…` (184
s), `98babfa5…` (213 s).
- remove-object with this mask: run
`1de167fa-7317-4963-b3aa-24e3535a016b`; virtual-try-on with this
subject: run `33160fe7-ea1e-46a4-b4cf-e8531cc335f9`.
- Every other published workflow passed in the full sweep after #19251.

Post-deploy: the mask is served from comfy.org once this ships; run
remove-object once through the site.

Previously accepted by Bill as #19272 and #19284 (same commits). CI's
`/models/` JavaScript-budget failure is fixed by #19256, which merges
first.

Validation: 7,377 website unit tests, website typecheck, 15 workflow
browser tests on the CI-equivalent build.


## Why product-photo-to-video has no new browser test

The change is inside the Cloud graph (`ConcatenateVideo` replacing the
decode-and-stitch subgraph) and only has an effect when Cloud executes
it. The website's Playwright suites stub Cloud's `/api/prompt` and
`/api/jobs` (`e2e/workshop-workflow-runs.spec.ts`), so no browser test
can observe whether Cloud keeps the worker alive or produces the
combined video. The page, its form and its submission path are unchanged
and remain covered by the existing workflow browser tests; the graph
change is verified where it runs — three live production runs above (all
four outputs, combined video included) — and by the scheduled live sweep
in #19274.

This branch was previously deployed

1 inactive deployment
Preview – billing-web-test — 7c3f7304 Deployed Sep 28, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

risk:high PR risk grade (advisory shadow check; grader-owned)

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant