Repository navigation
Conversation
…le instead of folding it to true Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ab7f54f73b2f and calibration run 8a669d31258e Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
… gains Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…frame is a copy
…thy; carry_pick window uses the snapped length
…24 (no needless retime, snapped soundtrack-length fallback)
…d duration beside match_audio; trim to arch ceilings
…ks in served guides resolve Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…celled with a server restart: reason Every admitted job has a history row from submit on (queued at submit, running at dispatch and at run_start, overwritten by _finish), stamped with a per-start server_instance. Each server holds servers/<instance>.lock beside jobs.sqlite for its life; a startup closes the open rows of every server whose lock is free as cancelled, error 'server restart: ...' (a running one names its run_dir), and leaves a live sibling's rows alone. History readers serve closed rows only. A failed write at submit or dispatch is logged and the job runs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…history they close jobs.py had gone over the 1,100-line ceiling; the lock and the startup sweep move to job_history.py as functions, the manager keeps the calls. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ained-segments so the (1 + retries) estimate is reachable Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ine (tracks, clips, members) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…files beside it) asks for chain_output Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…hand-off) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…ared num_frames constraint Drops h3_rules.H3_MAX_FRAMES, a second hand-kept copy of the templates' max_frames. Validate passes the workflow's declared constraint into media_problems; no declared ceiling, no limit sentence. The wrong "14 seconds" wording is gone. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… reads the entry _price quotes from Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
audio_dbfs re-rolls a segment (segment 1 too) whose generated audio reads under check audio_dbfs_min (-70); chain_segment_silent warns of a kept one with check on or off; the estimate quotes N(1+retries) for a chain that generates its audio. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…t num_frames); a VFR take clamped to its duration warns Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
…efault preset bundle Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ue read the default preset bundle Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…one) #1308 Amendment 1. A chain key `keystone_images` holds one image per segment; under keystone "segment_1" at keystone_slot "last", segment k's fl2va last_image is entry k-1 (0-based index i takes entry i) in place of segment 1's frame 0. Segment 1 takes entry 0 too. Off the anchor_every cadence (config.anchors(i), 0-based i % n == 0) a later segment gets no last image. Entry k stays segment k's on a probe and a resume_from. - chain_rules: chain_keystone_images_problem (keystone/slot, list shape, image entries, a step last_image, at the run an fl2va keyframe and no references) and keystone_images_count_problem (shorter than the full plan, naming both counts; a probe still needs the full plan's length). - chain_config: reads the list (or the resolved one), checks it before the keystone slot, counts it against the plan, loads paths through fetch_image (location policy); keystone_for(i, frame) picks each segment's keystone. - chain.py: per-segment placement, segment 1 included, and the job-log note names the entry (+4 lines; run_chain stays under 150). - previous_results: resolve_chain_lists resolves keystone_images - one previous_result: list of images (an image step's images, one per for_each entry, the first of each) or per-entry references. - guides: validate refuses it off H3 fl2va and runs the shared rule with the count when it can be told; locations checks its entries as media. - schema, WORKFLOW_GUIDE rows; narrated-take gains keystone, keystone_slot and keyframes variables and a shot preset walk_keyframed; the bench copy is re-synced and its pin test no longer treats keystone/keystone_slot as exposed-only keys. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d take, CP1 record, moving keystone (keystone_images) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Records static-long s1, the face-repair probe and Don's CP1 verdict; keyframes v1 (dropped) and v2 (plate crops + matted speaker, approved). The arm pads keyframes 13-43 with keyframe 12: the template refuses a list shorter than the full take, and the probe renders 12. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…aded plate; plus latent carry Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ssessment Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… at 86 s Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…cedure, guide section, attributions Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… routing) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…set, skill rule, guide, bench records, verdicts Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… over-sharpening (refs #1312) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ning, not colour drift (refs #1322) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… the first tail's channel mean and spread (refs #1322) experimental_latent_carry takes 'levelled' beside true. The first tail a run carries sets a per-channel level; every later tail's video latents are pulled back to it before the next segment reads them. Finished frames are untouched. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…softens late (refs #1322) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ed handoffs rejected (refs #1322) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ft_provided_sigmas held false for the call and put back Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ep_end (GHSA-5m72-m9g5-q8vp) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…d, levelled tail to h3_guides, resume hint to chain_resume, register_command's name rules split Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…n does, an empty script keeps Whisper's pauses, pauses capped at 60 s, ASR model_name checked at run time, surplus keystone_images never fetched Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… not a dotted dw. target Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…urity and diffusers 0.44 Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…s none, and every validate warned of it Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Release PR for 0.20.0: develop at 595cee1, CI green (run 43534282709; backend, ui and e2e all pass).
Notes are drafted in
docs/RELEASING.mdunder 0.20.0; paste them into the release body after the tag publishes.Quality check, 2026-11-13 (the 126 commits since the 2026-11-08 review)
refine_strength/refine_sigmasstep. dw now setsshift_provided_sigmasfalse for the call and puts it back.run_chainover 150 lines,register_commandover the complexity limit. Fixed by moves into existing modules; no baseline was raised.narration_linescounted frames by rounding up where the chain rounds, so about one narration in 300 was refused for a line-count mismatch.scriptcleared every pause inretime_narration; pauses are now capped at 60 s.templates/minimax/narrated-takedrew a "nothing reads 'lines'" warning on every validate.keystone_imageslist longer than the plan fetched its surplus.model_namearriving byprevious_result:was not checked at run time.Security
shots[].filecarried the server's absolute path inget_job,wait_for_jobandstep_end. Fixed in b7403a1.filename) is fixed in this range; publish after the tag.Local checks: 20,226 unit tests, 5 integration tests (MPS), both ratchets, ruff, and the UI's lint, type check, unit tests and build all pass. The UI e2e suite ran in CI only.
After merge:
scripts/release.sh 0.20.0 --next 0.21.0-alpha.1, paste the notes, restart the servers.🤖 Generated with Claude Code