Repository navigation
docs(changelog): weekly digest 2026-10-26-2026-11-02 - #7032
Conversation
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
|
Preview deployment for your docs. Learn more about Mintlify Previews.
💡 Tip: Enable Automations to automatically generate PRs for you. |
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
|
Weekly video (49.6s, 1080x1080): https://static.heygen.ai/hyperframes/changelog-videos/weekly-changelog-oct26-nov02.mp4 It's embedded at the top of this week's entry. |
There was a problem hiding this comment.
Verdict: APPROVE · 0 blocker · 0 important · 2 nits
All 79 commit links resolve to the PR cited next to them. Each commit subject matches its PR title, and each sentence matches what that PR describes. The two nits are inline.
What I checked
- 78 PRs: commit subject compared with PR title and merge SHA. #6442 is linked to its single branch commit
b76b52937, which is under mergee34f9f01d. That is fine. - Specific claims checked against PR bodies or code: the #6970 upgrade steps and
PLAN_PROTOCOL_UNSUPPORTED(inreleases/v0.9.0.mdand the /changelog v0.9.0 notes), the #6985 omittedPlanProtocoldefault, the #6641 10 GB container getting 3 workers instead of 1, the #6444 flag names,~/.hyperframesfor #6687 (cli/src/auth/paths.ts), the #7006--voice Chineseexample, the #6884 rule that previews and drafts stay local, and the #6914 opt-in default. - Tags: v0.8.205 to v0.9.5 is 32 published releases with no v0.8.215, and it picks up where last week's v0.8.204 left off. No PR cited here also appears in the Oct 19 entry.
- The video URL returns 200
video/mp4(36.9 MB).DocsVideois already imported. - CI: every Actions run on this head is green.
|
|
||
| ## Faster renders | ||
|
|
||
| High quality MP4s encode with the x264 fast preset ([be05c8693](https://github.com/heygen-com/hyperframes/commit/be05c86932be215ec70621dc568af96415555fa0), [#6640](https://github.com/heygen-com/hyperframes/pull/6640)). Worker sizing uses measured memory budgets and reads a container's CPU quota, so a render in a 10 GB container captures with three workers instead of one ([cd003e201](https://github.com/heygen-com/hyperframes/commit/cd003e201051317234598bcf1c70cfed4770613f), [#6641](https://github.com/heygen-com/hyperframes/pull/6641)). A 14 core Mac with a hardware GPU captures with 8 workers instead of 4, and a 5 minute film renders 43% faster ([e6476c2cd](https://github.com/heygen-com/hyperframes/commit/e6476c2cd57d8e605ca4d25f6c8b4fef6fdb322e), [#6636](https://github.com/heygen-com/hyperframes/pull/6636)). `--gpu` renders encode while frames are still being captured ([6553b1f6b](https://github.com/heygen-com/hyperframes/commit/6553b1f6ba75582c0a2d0830cf74a84c434548a0), [#6631](https://github.com/heygen-com/hyperframes/pull/6631)). |
There was a problem hiding this comment.
nit: per #6636, the 8 workers come from the memory rule on a 24 GB Mac. A 14-core Mac with 16 GB gets 5, so a 16 GB reader would expect more than they get. Suggest "A 14 core, 24 GB Mac with a hardware GPU…"
|
|
||
| ## Colour grading | ||
|
|
||
| Grade groups grade a whole lane from one `<hf-color-group>`. Clips join it with `data-color-group`, so grading a camera cut into hundreds of clips means editing one attribute ([09b129047](https://github.com/heygen-com/hyperframes/commit/09b12904748b71ab1777d335f20026d45d6944ca), [#6678](https://github.com/heygen-com/hyperframes/pull/6678)). Core can bake a grade into a 3D LUT `.cube` file that matches the preview ([c9017f45f](https://github.com/heygen-com/hyperframes/commit/c9017f45f4b8c38e11396a70a3d05d3432c58b48), [#6679](https://github.com/heygen-com/hyperframes/pull/6679)). 65 point `.cube` LUTs, the size grading apps export, are accepted ([9284e69ff](https://github.com/heygen-com/hyperframes/commit/9284e69fff2af4928d41cc376f36d5cfbd789bfd), [#6921](https://github.com/heygen-com/hyperframes/pull/6921)). A graded clip no longer comes out blank on a real GPU ([9eaf9efb6](https://github.com/heygen-com/hyperframes/commit/9eaf9efb69f37de0d6d98324bda1c65c9a924f2f), [#6615](https://github.com/heygen-com/hyperframes/pull/6615)). |
There was a problem hiding this comment.
nit: "matches the preview" overstates #6679. The PR shows parity only at the cube's lattice nodes and says error between nodes is not bounded yet (up to 11 codes at size 33 with a secondary). Suggest "…a 3D LUT .cube file built from the same stages as the preview shader".
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
What
The weekly digest for October 26 to November 2, 2026, prepended to
docs/weekly-updates.mdx. It covers v0.8.205 through v0.9.5: 32 tags. v0.8.215 was cut but never published; its changes shipped in v0.8.216.The generator's draft was rewritten into ten themes: the v0.9.0 Plan v2 break with upgrade steps, render controls, faster renders, Safari and iPhone playback, the browser editor, the Studio app on Windows, check and the CLI, colour grading, skills, sign-in and security, and docs.
How it was checked
ceea23891(main re-checked before commit, nothing new). 13 were already cited in the Oct 19-26 entry and were removed by SHA set subtraction, leaving 513.origin/main, falls in this week's new set, is not a reverted commit, and carries(#N)for the PR it is cited beside. 79 commit links, 78 PRs. The one merge-commit PR (feat(cli): offer the HyperFrames app download on Windows #6442) was confirmed withgh api repos/heygen-com/hyperframes/pulls/6442/commits.HF_VIDEO_LAYER_PASSTHROUGHis described as opt-in and off by default. The host-only "copies ahead" message (feat(core): a heavy clip's lighter copy can be made ahead of time #6982) is left out because nothing turns it on yet.npx mint validatefromdocs/passes.Testing
npx mint validate(build passed, validation passed). Docs-only change, one file.🤖 Generated with Claude Code