Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
476 commits
Select commit Hold shift + click to select a range
b43f33b
fix(variables): #1250 - refuse a dict or list for a bool-typed variab…
dkackman Nov 7, 2026
471694c
fix(plan): #1224 - chained-segments fixed_minutes 10, set from drill …
dkackman Nov 7, 2026
6080930
Merge remote-tracking branch 'origin/develop' into develop
dkackman Nov 7, 2026
1688c06
feat(chain): #1251 - seam look_hold log line carries measured/applied…
dkackman Nov 7, 2026
ac053ec
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 7, 2026
ac399eb
fix(chain): #1232 - keystone on t2va warns it does nothing; keystone …
dkackman Nov 7, 2026
ec80f7b
fix(chain): #1232 - match_audio/guide_backoff references not read tru…
dkackman Nov 7, 2026
d8add94
fix(h3): #1232 - audio pass treats a source within 0.5% of 24 fps as …
dkackman Nov 7, 2026
7ca8eae
fix(h3): #1232 - schema not-clause no longer refuses a null/reference…
dkackman Nov 7, 2026
093952b
fix(mcp): #1237 - serve the server, security and remote guides so lin…
dkackman Nov 7, 2026
1e985e6
feat(server): #1245 - a job a server restart dropped is listed as can…
dkackman Nov 7, 2026
508c129
feat(server): #1245 - server liveness lock and sweep live beside the …
dkackman Nov 7, 2026
6db8d96
Merge feat/1220-A-restart-tombstones (#1245)
dkackman Nov 7, 2026
57bc4cb
feat(templates): #1252 - expose check/retries on templates/minimax/ch…
dkackman Nov 7, 2026
4c497de
style(tests): #1252 - ruff format
dkackman Nov 7, 2026
92dfffe
feat(server): #1254 - get_gallery_metadata summarizes an .otioz timel…
dkackman Nov 7, 2026
0c3d1b9
fix(mcp): #1207 - a take copied away from its run (shots, no segment …
dkackman Nov 7, 2026
5d2f852
chore(api): #1254 - regenerate the OpenAPI document
dkackman Nov 7, 2026
51dcc42
chore(ui): regenerate the API contract after #1254 (blocked the #1207…
dkackman Nov 7, 2026
ac94b0b
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 7, 2026
6d2aaf6
fix(h3): #1208 - the too-long message takes its ceiling from the decl…
dkackman Nov 7, 2026
7d3467a
feat(chain): #1246 - an audio-level floor on chain segments (wip engine)
dkackman Nov 7, 2026
88e3558
fix(plan): #1224 - one owner for the cost entry choice; fixed_minutes…
dkackman Nov 7, 2026
0d0409c
feat(chain): #1246 - audio-level floor: tests, docs and skill quotes
dkackman Nov 7, 2026
a4b6d99
Merge feat/1221-A-audio-floor: #1246 audio-level floor on chain segments
dkackman Nov 7, 2026
415becf
fix(h3): #1232 - capstone refused at validate (flag off, from_end pas…
dkackman Nov 7, 2026
647beb9
fix(catalog): #1232 - cost_drivers and lora workflow_bases read the d…
dkackman Nov 7, 2026
e5b87cc
test(chain): #1246 - patch.object for the floor's warning patches
dkackman Nov 7, 2026
3d40ff5
Merge feat/1221-A-audio-floor: #1246 patch.object
dkackman Nov 7, 2026
377eeb7
fix(memory): #1232 - host-memory history rows without an explicit val…
dkackman Nov 7, 2026
de61535
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 7, 2026
0258d35
fix(guides): #1237 - serve the docs the server guide links; the link …
dkackman Nov 7, 2026
0de7497
fix(chain): #1242 - near-greyscale fed chroma below a floor is not le…
dkackman Nov 7, 2026
d945fe4
feat(variables): #1253 - opt-in required, no-default variable; refere…
dkackman Nov 7, 2026
7622d80
refactor(variables): #1253 - keep constraint_errors under the complex…
dkackman Nov 7, 2026
c599fb1
refactor(chain): #1246 - one owner of held audio, the floor and segme…
dkackman Nov 7, 2026
ab29dfd
Merge feat/1221-A-audio-floor: #1246 one owner of segment 1's audio c…
dkackman Nov 7, 2026
abc787e
fix(h3): #1256 - decode refuses latents too short to decode; T=1-2 an…
dkackman Nov 7, 2026
be71e68
fix(h3): #1256 - refine windows refuse a gated capstone row; drop unu…
dkackman Nov 7, 2026
2ac5fe0
fix(memory): #1256 - host-memory history row that chose another prese…
dkackman Nov 7, 2026
40b3e33
style: #1256 - ruff format host-memory test
dkackman Nov 7, 2026
7d6f03b
fix(guides): #1257 - section match ignores a heading's trailing paren…
dkackman Nov 7, 2026
7e185e5
fix(chain): #1246 - a literal audio floor reads 'below 0'; observed a…
dkackman Nov 7, 2026
ca4d640
Merge feat/1221-A-audio-floor: #1246 friendly literal floor refusal, …
dkackman Nov 7, 2026
fac9c10
feat(plugin): #1255 - static/moving class column for the camera moves…
dkackman Nov 7, 2026
b8b40ad
Merge feat/1200-A-move-class: #1255 move class column
dkackman Nov 7, 2026
89d5650
fix(validate): #1259 - refuse a chain with segments plus match_audio …
dkackman Nov 7, 2026
331603c
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 7, 2026
4a69ebe
docs(proposals): #1200 close-out - move-class design record, todo entry
dkackman Nov 7, 2026
8226abe
Merge docs/1200-closeout-move-class: #1200 close-out
dkackman Nov 7, 2026
8109691
fix(templates): #1261 - relevel-clip exposes limit (default true); ca…
dkackman Nov 7, 2026
6b4e31b
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 7, 2026
e7f284e
fix(chain): #1259 - the run asks match_audio_source_problem; map rows…
dkackman Nov 7, 2026
7ad6b6f
Enhance documentation and workflows for improved clarity and function…
dkackman Nov 8, 2026
5a45bcb
docs(workflow-guide): #1264 - the catalog index is list_workflows; th…
dkackman Nov 8, 2026
3432ba4
fix(worker): #1265 - a release_models task frees the warm worker's id…
dkackman Nov 8, 2026
8a257e2
fix(worker): #1265 - a release_models workflow step (the flux3 upsamp…
dkackman Nov 8, 2026
b7a3f3c
docs(chain): #1246 - an audio_dbfs re-roll is not guaranteed to bring…
dkackman Nov 8, 2026
6dc4a5f
Merge feat/1221-a-reroll-caveat (#1246)
dkackman Nov 8, 2026
5f89b2a
feat(arguments): #1247 - from_pretrained torch_dtype accepts diffuser…
dkackman Nov 8, 2026
64748cc
test(arguments): #1247 - format
dkackman Nov 8, 2026
e514431
docs(acceleration): #1247 - per-component dtype example loads audio_v…
dkackman Nov 8, 2026
08c6ebf
refactor(type_references): #1247 - detect a variable reference with r…
dkackman Nov 8, 2026
8d31fd8
Merge feat/1221-b-dict-torch-dtype (#1247)
dkackman Nov 8, 2026
4ccc021
refactor(type_references): #1247 - validate-time escape check calls a…
dkackman Nov 8, 2026
3743424
Merge feat/1221-b-dict-torch-dtype (#1247): arch-review fix
dkackman Nov 8, 2026
ed838aa
docs(bench): #1267 - LTX audio dropout Cause: the conditioning, not t…
dkackman Nov 8, 2026
98dc656
Merge feat/1221-b2-ltx-audio-cause (#1267)
dkackman Nov 8, 2026
bffa0b4
fix(host-ram): #1227 review - private claims folder, cold-worker clai…
dkackman Nov 8, 2026
07f6a56
fix(plan): #1224 review - observed history holds the entry's fixed share
dkackman Nov 8, 2026
6f79c07
fix(engine): #1265 review - a child's release_models spares its paren…
dkackman Nov 8, 2026
06e57c8
fix(tasks): review nits - per-gain look samples, pruned-segment error…
dkackman Nov 8, 2026
4521625
Merge fix/ultrareview-2026-11-08: ultrareview findings on develop vs …
dkackman Nov 8, 2026
bd668eb
fix(chain): review of #1199 pipeline_processors
dkackman Nov 8, 2026
6ffeda8
Merge fix/review-1199-pipeline-processors: #1199 review, pipeline_pro…
dkackman Nov 8, 2026
c3dd604
fix(tasks): review of #1199 dw/tasks
dkackman Nov 8, 2026
c5ca5fb
Merge fix/review-1199-tasks: #1199 review, dw/tasks
dkackman Nov 8, 2026
2cc8e1f
fix(engine): review of #1199 dw (server, presets, cost, memory)
dkackman Nov 8, 2026
b9313f8
Merge fix/review-1199-dw-rest: #1199 review, rest of dw
dkackman Nov 8, 2026
46fcb7e
test: review of #1199 tests - make the weak ones able to fail
dkackman Nov 8, 2026
fad034e
Merge fix/review-1199-tests: #1199 review, tests
dkackman Nov 8, 2026
5e87cd0
docs(templates): review of #1199 workflows and plugins
dkackman Nov 8, 2026
a2d9265
Merge fix/review-1199-workflows-plugins: #1199 review, workflows and …
dkackman Nov 8, 2026
ceb144d
refactor(arch): bring the ratchet back to baseline
dkackman Nov 8, 2026
52a0543
Merge fix/arch-ratchet-2026-11-08: architecture ratchet back to baseline
dkackman Nov 8, 2026
ce304f8
fix(chain): derive a length-planned chain's frame_snap from its num_f…
dkackman Nov 8, 2026
c2e86cf
fix(host_memory): check the claims folder's mode on posix only
dkackman Nov 8, 2026
2ae10c6
fix(chain_repair): name the extra segment files rather than "it has n…
dkackman Nov 8, 2026
afcbdc6
fix(media_readers): a ForwardFrameReader read after close() restarts …
dkackman Nov 8, 2026
7585b7e
test(notes_api): leave the host-claims folder out of the traversal sn…
dkackman Nov 8, 2026
f3d1a38
refactor(chain,server): four review cleanups, no behaviour change
dkackman Nov 8, 2026
b9458de
Merge fix/release-review-0-20-0: the 0.20.0 readiness review defects
dkackman Nov 8, 2026
c023968
feat(ui,tasks): the job page draws an .otioz timeline; clips carry th…
dkackman Nov 8, 2026
05d30cc
Merge feat/rr-f: the job page draws an .otioz timeline
dkackman Nov 8, 2026
8a08588
feat(audio): match_levels lufs - a BS.1770 delivery target on the join
dkackman Nov 8, 2026
e245b48
feat(server,mcp): rerun_job takes argument overrides - promote a draf…
dkackman Nov 8, 2026
9a7ee9e
feat(tasks): check_narration - the script against the speech, per take
dkackman Nov 8, 2026
c5fd773
fix(audio): lufs matching passes a silent-video input through
dkackman Nov 8, 2026
7a5210a
Merge feat/rr-d: match_levels lufs, a BS.1770 delivery target on the …
dkackman Nov 8, 2026
ef415a1
fix(tasks): check_narration blames a misheard last word on its own take
dkackman Nov 8, 2026
8d2530b
fix(server): a rerun's overrides survive in the job history
dkackman Nov 8, 2026
0a05c56
Merge feat/rr-e: check_narration, the script against the speech per take
dkackman Nov 8, 2026
85881f9
Merge feat/rr-a: rerun_job takes argument overrides
dkackman Nov 8, 2026
fc1433f
chore(api): regenerate the OpenAPI document for rerun overrides
dkackman Nov 8, 2026
04cbc48
feat(chain): resume_from reuses a dead chain's kept segments
dkackman Nov 8, 2026
f45925d
fix(chain): resume_from reads this iteration's own segment files
dkackman Nov 8, 2026
3f33723
Merge feat/rr-b: chain.resume_from reuses a dead chain's kept segments
dkackman Nov 8, 2026
12b0ad7
docs(releasing): draft the 0.20.0 notes; the release-review feature d…
dkackman Nov 8, 2026
202120f
fix(narration): charge a repeat at a take boundary to the take it rep…
dkackman Nov 8, 2026
421206d
fix(joins): a lufs match with no soundtrack to measure needs no sampl…
dkackman Nov 8, 2026
e6d6a18
fix(rerun): a rerun with overrides keeps the seed the original drew
dkackman Nov 8, 2026
0343528
docs(long-form): delivered loudness is assemble-and-score's target_lufs
dkackman Nov 8, 2026
e36b089
docs(architecture): seam rows for rerun overrides, LUFS match, take a…
dkackman Nov 8, 2026
028ce65
refactor(narration): decide a boundary insertion's take once per run
dkackman Nov 8, 2026
06bbc3e
chore(arch): modules baseline 259 -> 261 for narration_check and chai…
dkackman Nov 8, 2026
4b32254
docs(plan): the promote-a-draft example is tier standard
dkackman Nov 8, 2026
39720f8
Merge fix/rr-final: the whole-branch review of the 0.20.0 features
dkackman Nov 8, 2026
7b5d1c8
fix(mcp): #1283 - listings clip a note to its first line
dkackman Nov 9, 2026
fc11440
fix(voice): #1293 - seed htdemucs random shift so separation is repea…
dkackman Nov 9, 2026
ccb3280
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
978f5db
style(tests): #1293 - ruff format
dkackman Nov 9, 2026
299bd34
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
5137fb1
fix(worker): #1296 - report a memory reading at job start so mid-run …
dkackman Nov 9, 2026
126ea1e
fix(tasks): #1287 - whole-number refusals for text and non-finite val…
dkackman Nov 9, 2026
860d12b
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
8f5c5ba
fix(security): refuse a task weight filename outside the model repo
dkackman Nov 9, 2026
b0d339d
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
031eb14
fix(server): #1289 - clear_memory refusal with every card busy names …
dkackman Nov 9, 2026
4f97e8b
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
5b125a7
fix(worker): #1288 - a no-growth run records a null job peak, not the…
dkackman Nov 9, 2026
4773bf4
fix(server): #1278 - pre-device history rows match a plan's card by t…
dkackman Nov 9, 2026
1c94c77
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
6586d4f
docs(arch): #1278 - map names the renamed observed-cost test
dkackman Nov 9, 2026
0e8166c
feat(chain): #1297 - dialogue_text, number-word normalising and check…
dkackman Nov 9, 2026
e36b99b
feat(chain): #1297 - words.md spread and CPU cost, fold lone number w…
dkackman Nov 9, 2026
4cdf6ff
Merge branch 'feat/1262-A-words-measure' into develop (#1297)
dkackman Nov 9, 2026
c98fd4a
fix(server): #1278 - default card's name through ObservedCosts.card, …
dkackman Nov 9, 2026
543b755
feat(chain): #1297 - seam head on the script-check owners' fold and a…
dkackman Nov 9, 2026
c114322
Merge branch 'feat/1262-A-words-measure-arch' into develop (#1297)
dkackman Nov 9, 2026
84d5f28
fix(check-script): #1297 - digits written apart are not one number, h…
dkackman Nov 9, 2026
07a3a23
Merge branch 'feat/1262-A-words-measure-verify' into develop (#1297)
dkackman Nov 9, 2026
dd0cf57
feat(chain): #1298 - words check: words_similarity_min (null default)…
dkackman Nov 9, 2026
9dce522
test(chain): #1298 - words check: measure, re-roll, warning, lines ru…
dkackman Nov 9, 2026
aa92ca7
Merge branch 'feat/1262-B-words-chain' into develop (#1298)
dkackman Nov 9, 2026
38486c8
fix(chain): #1298 - words check on the observed/inherited estimate, l…
dkackman Nov 9, 2026
f37ac82
Merge branch 'feat/1262-B-words-chain-bounce1' into develop (#1298)
dkackman Nov 9, 2026
1ffa7ac
fix(chain): #1298 - words failure names its findings and what was hea…
dkackman Nov 9, 2026
be4a7b5
Merge branch 'feat/1262-B-words-chain-bounce2' into develop (#1298)
dkackman Nov 9, 2026
18fabc5
fix(cost): #1298 - _is_literal_text asks references.is_ref for the de…
dkackman Nov 9, 2026
ee4e18c
Merge branch 'feat/1262-B-words-chain-bounce3' into develop (#1298)
dkackman Nov 9, 2026
cf25d56
fix(tests): #1277 - shrink test waits for the final size instead of a…
dkackman Nov 9, 2026
9cd859e
fix(chain): #1275 - frame_snap derivation notes, grid-less advisory w…
dkackman Nov 9, 2026
daa3500
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
fe31eeb
fix(ui): #1274 - timeline strip minors: single otioz probe, rounded g…
dkackman Nov 9, 2026
4dec7e8
fix(chain): #1268 #1269 #1270 - resume_from follow-ups
dkackman Nov 9, 2026
1d26be2
Merge fix/1268-1269-1270-resume into develop (#1268 #1269 #1270)
dkackman Nov 9, 2026
249a940
fix(chain): #1270 - keep run_chain under the function-length ratchet
dkackman Nov 9, 2026
1c429a6
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
f7d4e19
fix(joins): #1271 - warn on inert/refused match_levels_lufs at validate
dkackman Nov 9, 2026
2ffb2eb
fix(ui): #1274 - link a bundle member only when its clip's recorded o…
dkackman Nov 9, 2026
a109be5
fix(narration): #1272 - module-level ASR seam, audio required, head-m…
dkackman Nov 9, 2026
0d3fbfd
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
2e3de64
fix(rerun): #1273 - null seed override draws fresh, seed_unknown word…
dkackman Nov 9, 2026
8e8ce07
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
c8823e4
fix(validate): #1302 - export_timeline result content_type must be ap…
dkackman Nov 9, 2026
9f859c8
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
0f8c850
fix(validate): #1302 - export_timeline declares its result content_ty…
dkackman Nov 9, 2026
32ab548
fix(validate): #1271 - inert match_levels warning asks joins._match_t…
dkackman Nov 9, 2026
c32f491
docs: #1272 - take attribution row and Checking a Narration name the …
dkackman Nov 9, 2026
63432ad
fix(validate): #1303 - match_levels pairing warnings resolve variable…
dkackman Nov 9, 2026
959087e
fix(chain): #1304 - realize_args leaves a chain's resume_from output:…
dkackman Nov 9, 2026
e0633d4
docs(arch): #1304 - ARCHITECTURE rows name the resume_from key-name e…
dkackman Nov 9, 2026
f4f07a4
fix(chain): #1268 - decide fan-out from the chain step's own argument…
dkackman Nov 9, 2026
569f69a
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
82f272f
fix(tests): use process time for performance measurement in TestNorma…
dkackman Nov 9, 2026
f9981cb
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
d58d67a
feat(chain): #1300 - line_clipped_at_head names a line opened inside …
dkackman Nov 9, 2026
7afbce8
fix(plan): #1305 - an inherited chain figure scales by segments when …
dkackman Nov 9, 2026
aa9e816
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
adb33c1
docs(tasks): #1301 - drop the AAC-overshoot caveat on match_levels_lu…
dkackman Nov 9, 2026
7e6a917
feat(plugin): #1299 - chained-segments takes lines; long-form and h3 …
dkackman Nov 9, 2026
63ae4a8
fix(chain): #1304 - hold any variable a resume_from names out of the …
dkackman Nov 9, 2026
6bf8983
Merge branch 'fix/1304-resume-any-variable' into develop
dkackman Nov 9, 2026
abdef93
refactor(cost): #1305 - one public chain_segment_scaling rule for cat…
dkackman Nov 9, 2026
7b57ea2
fix(cost): #1306 - a resume keeping every planned segment says nothin…
dkackman Nov 9, 2026
68c189f
docs(arch): #1300 - chain check row names line_clipped_at_head and wh…
dkackman Nov 9, 2026
ac6e482
fix(chain): #1307 - a fan-out's finished iteration renders in full wh…
dkackman Nov 9, 2026
785c99f
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
9b1eaaf
fix(chain): #1270 - hint on a resume load failure after copying; anch…
dkackman Nov 9, 2026
607c2f9
Merge branch 'develop' of https://github.com/dkackman/diffusers-workf…
dkackman Nov 9, 2026
7106d4e
fix(chain): #1308 - cue a carried dialogue line past the guide span; …
dkackman Nov 9, 2026
167fd6b
feat(plugin): #1299 - chained-segments check takes the words ceilings…
dkackman Nov 9, 2026
ef79c6e
fix(mcp): #1309 - inherited-cost warning gives the resume notes their…
dkackman Nov 9, 2026
1afe221
fix(chain): #1270 - a kept file load refuses leaves the spill; resume…
dkackman Nov 9, 2026
b6ee309
fix(mcp): #1309 - inherited-cost warning sorts differs entries by sha…
dkackman Nov 9, 2026
9159f7b
style: #1309 - ruff format test
dkackman Nov 9, 2026
58861bb
docs: #1262 - words check design record and todo entry
dkackman Nov 9, 2026
1f72513
Merge branch 'docs/1262-closeout-chain-words-check' into develop (#1262)
dkackman Nov 9, 2026
1f966b8
docs(proposal): narrated long take - design and plan
dkackman Nov 9, 2026
4d07c27
fix(chain): #1308 - a passed segment missing its opening words re-rol…
dkackman Nov 9, 2026
f55b3ea
feat(narration): retime_narration and narration_lines - pauses by pun…
dkackman Nov 9, 2026
40e23a7
fix(narration): boundary margin tolerance; prompt_template is plain p…
dkackman Nov 9, 2026
585c315
fix(narration): prompt_template length is bounded before substitution
dkackman Nov 9, 2026
e45829c
feat(chain): prompts and lines may be one previous_result list
dkackman Nov 9, 2026
3435be8
feat(narration): narration-retime template
dkackman Nov 9, 2026
eded3f7
fix(narration): narration-retime requires its script
dkackman Nov 9, 2026
a4b9d61
test(narration): step-cache keys row for narration-retime
dkackman Nov 9, 2026
57ac0ca
fix(narration): a segment boundary may sit at any word junction; line…
dkackman Nov 9, 2026
c11f78e
fix(narration): mid-word boundaries are advisory; the take's end is n…
dkackman Nov 9, 2026
e6b4acb
feat(chain): a guide chain held to a narration carries no guide audio…
dkackman Nov 9, 2026
4c6e43b
feat(chain): stop_after renders a probe; resume_from continues it
dkackman Nov 9, 2026
8e67c09
fix(chain): a probe is quoted for the segments it renders; stop_after…
dkackman Nov 9, 2026
e354062
feat(chain): face_drop and lip_sync segment measures, warn-only; meas…
dkackman Nov 9, 2026
29545a6
fix(chain): lip sync leaves the segment check after calibration; face…
dkackman Nov 9, 2026
72eba85
feat(minimax): narrated-take template
dkackman Nov 9, 2026
3fb6d62
fix(cost): a matched chain's observed history is priced per segment
dkackman Nov 9, 2026
0e36a3e
feat(chain): latent carry under a held narration, latent keystone, an…
dkackman Nov 9, 2026
036cb27
Merge branch 'feat/narrated-long-take' into develop: narrated long ta…
dkackman Nov 9, 2026
0a7fe24
feat(chain): repair a narrated take - held master slice, take tier, l…
dkackman Nov 9, 2026
29ec563
docs(chain): repair-segment says a narrated take's tier cannot be low…
dkackman Nov 9, 2026
77b393b
bench(narrated): CP1 control vs candidate, seed 1
dkackman Nov 10, 2026
11916e5
docs(narrated): amendment 1 - the moving keystone (CP1 verdict)
dkackman Nov 10, 2026
cd1a6ee
feat(chain): keystone_images - a per-segment last image (moving keyst…
dkackman Nov 10, 2026
12e60a4
Merge branch 'feat/narrated-long-take' into develop: repair a narrate…
dkackman Nov 10, 2026
233fe0c
bench(narrated): moving-keystone keyframes v2 and the 11c arm
dkackman Nov 10, 2026
7cb871f
bench(narrated): CP1 moving keystone, seed 1
dkackman Nov 11, 2026
d778679
bench(narrated): CP1 moving keystone, seed 2
dkackman Nov 11, 2026
030be5c
bench(narrated): arc arms - camera arc keyframes, mid-word mouths, gr…
dkackman Nov 11, 2026
d230a3f
bench(narrated): arc results, keyframes v4 (mouth-only blend), path a…
dkackman Nov 12, 2026
58cef9b
bench(narrated): Kandinsky S2V Lite-distill spike - fails on identity…
dkackman Nov 12, 2026
c4e662f
bench(narrated): arc v4 + latent carry, seed 1
dkackman Nov 12, 2026
a5f993a
bench(narrated): arc v4 arm without latent carry
dkackman Nov 12, 2026
eb0c1c8
docs(long-form): the narrated take - standing preset, skill rule, pro…
dkackman Nov 12, 2026
f94e414
docs(long-form): narrated take - review fixes (wall time, seed count,…
dkackman Nov 12, 2026
be197a3
docs(proposals): narrated long take - verdicts
dkackman Nov 12, 2026
f4db3db
docs(proposals): narrated long take - follow-up issue numbers
dkackman Nov 12, 2026
f2099f0
Merge feat/narrated-long-take: narrated take close-out - standing pre…
dkackman Nov 12, 2026
e77bdc7
docs(bench): narrated take - arc v4 without latent carry isolates the…
dkackman Nov 13, 2026
732444c
docs(bench): narrated take - masked handoff probe removes over-sharpe…
dkackman Nov 13, 2026
d296716
feat(chain): levelled latent carry - each carried tail pulled back to…
dkackman Nov 13, 2026
837c6e9
docs(bench): narrated take - levelled latent carry removes the burn, …
dkackman Nov 13, 2026
9212fb3
docs(bench): narrated take - levelled seam peak range (refs #1322)
dkackman Nov 13, 2026
31bb303
docs(proposals): narrated long take - second close, masked and levell…
dkackman Nov 13, 2026
28e914f
fix(h3): a refine schedule is set verbatim under diffusers 0.44 - shi…
dkackman Nov 14, 2026
b7403a1
fix(server): a shot's file is named as files are, in manifests and st…
dkackman Nov 14, 2026
dd73eb7
refactor(arch): back under the ratchet - rerun seed rule to job_recor…
dkackman Nov 14, 2026
ba7102f
fix(review): 0.20.0 review - narration lines count frames as the chai…
dkackman Nov 14, 2026
4484385
test(review): the run-time model_name test patches the module object,…
dkackman Nov 14, 2026
80f691e
docs(release): 0.20.0 notes - words check, narrated take, probes, sec…
dkackman Nov 14, 2026
595cee1
fix(minimax): narrated-take passes no chain lines - a held chain read…
dkackman Nov 14, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
4 changes: 2 additions & 2 deletions .claude-plugin/marketplace.json
Original file line number Diff line number Diff line change
@@ -1,15 +1,15 @@
{
"$schema": "https://anthropic.com/claude-code/marketplace.schema.json",
"name": "diffusers-workflow",
"description": "Skills for driving a diffusers-workflow server from Claude Code: a composition skill per model family, one for cutting a series from them, one camera-and-look vocabulary compiled per family, and storyboards with a timed animatic.",
"description": "Skills for driving a diffusers-workflow server from Claude Code: a composition skill per model family, one for cutting a series from them, one camera-and-look vocabulary compiled per family, storyboards with a timed animatic, and a playbook for pieces longer than one clip.",
"owner": {
"name": "Don Kackman"
},
"plugins": [
{
"name": "dw",
"source": "./plugins/dw",
"description": "Compose MiniMax H3 video, MiniMax Music 3, LTX-2.5, Kandinsky 6.0 and FLUX 3 Klein workflows over a dw MCP server, cut a multi-episode series from them, direct camera and look in each family's caption format, and storyboard a scene before it is filmed: which template fits which shape, the hard rules, cost, and how to judge the output. Prompt format comes from the vendors' own guides."
"description": "Compose MiniMax H3 video, MiniMax Music 3, LTX-2.5, Kandinsky 6.0 and FLUX 3 Klein workflows over a dw MCP server, cut a multi-episode series from them, direct camera and look in each family's caption format, storyboard a scene before it is filmed, and plan a piece longer than one clip from measured levers: which template fits which shape, the hard rules, cost, and how to judge the output. Prompt format comes from the vendors' own guides."
}
]
}
8 changes: 4 additions & 4 deletions CLAUDE.md
Original file line number Diff line number Diff line change
Expand Up @@ -74,10 +74,10 @@ The guides in `docs/`, by topic:
composition skill per model family (`minimax-h3`, `minimax-music3`, `ltx-2.5`,
`kandinsky-6`, `flux3-klein`) that chooses a template for a request's shape and states the
family's hard rules, plus cross-cutting composition skills (`script-to-video`,
`series-episodes`, `cinematography`, `storyboard`) - shapes above the families that
orchestrate the decision trees, cast consistency across multiple generations, one
camera-and-look vocabulary compiled per family, and boards and a timed animatic before
video. Every skill the directory holds is named in `plugins/dw/README.md` and here, pinned
`series-episodes`, `cinematography`, `storyboard`, `long-form`) - shapes above the
families that orchestrate the decision trees, cast consistency across multiple
generations, one camera-and-look vocabulary compiled per family, boards and a timed
animatic before video, and the measured playbook for pieces longer than one clip. Every skill the directory holds is named in `plugins/dw/README.md` and here, pinned
by `tests/test_plugin_skills.py`; that README says how the rest is pinned and versioned.

## Security Rules
Expand Down
70 changes: 67 additions & 3 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,37 @@ variable, run it again, compare takes and keep the one you like.
<img alt="A workspace's overview in the web UI: recent outputs from an agent-made musical trailer, its jobs, and the workflows the agent wrote" src="docs/img/ui-overview.jpg">
</picture>

### Coming from ComfyUI?

The job is the same: load models, wire steps together, get media out. What changes is
who builds the graph and how it is checked before it runs.

| | ComfyUI | diffusers-workflow |
|---|---|---|
| A workflow is | a node graph you draw | a JSON document of named steps an agent writes, and you can edit as a form |
| A new model needs | a custom node pack | nothing, if diffusers has the pipeline: any pipeline class is reachable by name |
| Mistakes are found | links are type-checked on queue; the rest when it runs | at validation: schema, references and every argument against the real pipeline signatures, before any model loads |
| Before a long render | no estimate | `validate_workflow` quotes GPU minutes from this machine's own run history |
| Between runs | models stay loaded | models stay loaded (one persistent worker per GPU) |
| A result | an image with the graph embedded | the same, and a job that keeps the exact workflow it ran, drawn as a data-flow graph, re-runnable with a new seed |

It is not a ComfyUI replacement for hand-wiring a graph node by node. It is for
describing what you want and letting an agent compose, price, run and judge it, then
taking over in the browser where that is quicker.

### What it makes

| | Ready-made templates |
|---|---|
| **Images** | FLUX 3 Klein (text, edits, multi-reference, layout boxes), FLUX.1 / FLUX.2 / Krea, Qwen Image (and edit), Z-Image, SDXL and SD 1.5; ControlNet, IP-Adapter, LoRA, inpaint and outpaint |
| **Video with sound** | MiniMax H3 (dialogue, multi-shot cuts, music videos, reference subjects and voices), LTX-2.5 (image-to-video, keyframes, 2x upscale, restore and colourise, alpha mattes), Kandinsky 6 |
| **Music, speech, sound** | MiniMax Music 3 (sung lyrics or instrumental), Audio8 TTS voice cloning and MMS TTS, MMAudio sound effects from text or under a clip, Whisper transcription |
| **Longer than one clip** | chained segments that hold the look and audio level across seams, repair of one bad segment in place, music and narration of any length, storyboards and timed animatics, cut-and-score assembly, and an OpenTimelineIO export that DaVinci Resolve opens as one clip per shot |

Anything else diffusers can load runs from a workflow too; the templates are where the
agent starts, not a limit. The full list is in [workflows/](workflows/) and the
[dw plugin](plugins/dw/README.md).

## Getting started

**1. Install.** The script picks the right torch build for your platform, creates a
Expand Down Expand Up @@ -109,7 +140,8 @@ The [dw plugin](plugins/dw/README.md) adds what a general-purpose agent can't kn
on its own: one skill per model family (MiniMax H3 and Music 3, LTX-2.5, Kandinsky 6,
FLUX 3 Klein) that picks the template for a request and states the rules that bite.
Cross-cutting skills cover script-to-video, series with a consistent cast,
storyboards and animatics, and a shared camera-and-look vocabulary.
storyboards and animatics, pieces longer than one clip, and a shared
camera-and-look vocabulary.

![The same session hours later: shots rendering one at a time, with the agent reporting progress between them](docs/img/claude-generating.png)

Expand Down Expand Up @@ -182,15 +214,47 @@ you do, the [Workflow Guide](docs/WORKFLOW_GUIDE.md) is the reference, the
a command-line client, `python -m dw.run workflow.json prompt="a cat"`, which queues
a job on a running server (see [Server & Web UI](docs/SERVER.md#the-command-line)).

Here is the shape, trimmed: a still from FLUX 3 Klein, animated by LTX-2.5.

```json
{
"variables": { "prompt": "a lighthouse at dusk, waves breaking", "seed": null },
"seed": "variable:seed",
"steps": [
{
"name": "still",
"pipeline": {
"configuration": { "component_type": "ModularPipeline" },
"from_pretrained_arguments": { "model_name": "black-forest-labs/flux-3-image-klein" },
"arguments": { "prompt": "variable:prompt", "width": 1280, "height": 720 }
}
},
{
"name": "clip",
"pipeline": {
"configuration": { "component_type": "LTX2ImageToVideoPipeline" },
"from_pretrained_arguments": { "model_name": "Lightricks/LTX-2.5-Diffusers" },
"arguments": { "prompt": "variable:prompt", "image": "previous_result:still" }
},
"result": { "content_type": "video/mp4" }
}
]
}
```

`component_type` is a diffusers class name and `arguments` are its `__call__`
arguments, so the diffusers docs are the reference for what a step takes. The real
templates add the device placement, quantization and size rules each model needs.

The engine also covers:

- **Quantization:** BitsAndBytes, TorchAO, GGUF, SDNQ and optimum-quanto, set per
component.
- **Acceleration:** FirstBlockCache, FasterCache, MagCache, TaylorSeerCache,
`torch.compile` and attention backends.
- **Adapters and prompts:** LoRA, IP-Adapter and A1111-style prompt weighting.
- **Long video and caching:** long-video chaining with audio-driven length, and
step-output caching, so re-running a fixed-seed workflow finishes instantly.
- **Step-output caching:** re-running a fixed-seed workflow, or one that changed
only a later step, reuses the steps that did not change.
- **Utility tasks:** upscaling, face restoration, segmentation, captioning, frame
interpolation, audio slicing, cut-and-score assembly and more.

Expand Down
36 changes: 36 additions & 0 deletions bench/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,36 @@
# Long-form bench server

A second `dw.serve` on lem for benching long video takes, so the main server (the automated tester's, port 8765) is never disturbed.

| | |
|---|---|
| Port | 8766 (`http://lem:8766`) |
| Card | `cuda:1` (`BENCH_DEVICES` overrides); the main server runs on `cuda:0` only (systemd drop-in `~/.config/systemd/user/dw-serve.service.d/devices.conf`) |
| Workspace | `/home/don/bench-workspace` |
| Code | worktree `/home/don/dw-bench`, detached at `origin/<branch>`; the main checkout is never touched |
| Token | `BENCH_TOKEN`, else `/home/don/.dw-bench-token` (mode 600) |
| Log | `/home/don/dw-bench.log` |

## Use

```bash
# first time (the worktree does not exist yet): from any lem shell
ssh lem 'git -C ~/diffusers-workflow fetch origin feat/long-form-media && git -C ~/diffusers-workflow worktree add --detach ~/dw-bench origin/feat/long-form-media'
# afterwards
ssh lem 'bash ~/dw-bench/scripts/bench_server.sh deploy feat/long-form-media'
ssh lem 'bash ~/dw-bench/scripts/bench_server.sh start' # waits up to 120 s for /api/health
ssh lem 'bash ~/dw-bench/scripts/bench_server.sh status'
ssh lem 'bash ~/dw-bench/scripts/bench_server.sh stop' # SIGTERM, SIGKILL after 30 s
```

A code change needs `stop`, `deploy`, `start`: the server runs persistent workers.

## Host RAM rule

The two servers do not share the host-RAM hold. lem has 125 GB (122 GB available idle), but no recent H3 chain with a projected `host_ram_need` was on the main server to confirm that both servers' largest jobs fit together. Until that is measured, run the bench only while the main server is idle (`curl localhost:8765/api/jobs?status=running` on lem).

They share no step cache either: an arms file goes to one server (`bench/driver.py --server`) for its whole life, including measure and rerun, because another server's run is a cache miss.

## Reaching it from the Mac

lem's firewall passes 8765 but not 8766 (a rule needs sudo: `sudo ufw allow 8766/tcp`, if ufw is the filter). Until then, tunnel: `ssh -f -N -L 18766:127.0.0.1:8766 lem`, then use `http://127.0.0.1:18766`.
47 changes: 47 additions & 0 deletions bench/arms/audio-long-2.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,47 @@
[
{
"name": "s6-music-full-335s-s1166",
"scene": "s6-narration",
"workflow": "music-long-s6-full.json",
"arguments": {},
"seed": 1166
},
{
"name": "s6-music-full-335s-s1167",
"scene": "s6-narration",
"workflow": "music-long-s6-full.json",
"arguments": {},
"seed": 1167
},
{
"name": "s6-music-full-335s-xf3000-s1166",
"scene": "s6-narration",
"workflow": "music-long-s6-full.json",
"arguments": {
"crossfade_ms": 3000
},
"seed": 1166
},
{
"name": "s6-narr-long-t37-wps3-s1166",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {
"max_seconds": 37,
"words_per_second": 3.0
},
"seed": 1166,
"workflow_id_suffix": "-t37"
},
{
"name": "s6-narr-long-t37-wps3-s1167",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {
"max_seconds": 37,
"words_per_second": 3.0
},
"seed": 1167,
"workflow_id_suffix": "-t37"
}
]
9 changes: 9 additions & 0 deletions bench/arms/audio-long-prep.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,9 @@
[
{
"name": "s6-narrator-ref-s1166",
"scene": "s6-narration",
"workflow": "narrator-reference-s6.json",
"arguments": {},
"seed": 1166
}
]
57 changes: 57 additions & 0 deletions bench/arms/audio-long.json
Original file line number Diff line number Diff line change
@@ -0,0 +1,57 @@
[
{
"name": "s6-narr-long-240s-t60-s1166",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {},
"seed": 1166
},
{
"name": "s6-narr-long-240s-t60-s1167",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {},
"seed": 1167
},
{
"name": "s6-narr-long-240s-t45-s1166",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {
"max_seconds": 45
},
"seed": 1166
},
{
"name": "s6-narr-long-240s-gap005-s1166",
"scene": "s6-narration",
"workflow": "narration-long-s6.json",
"arguments": {
"gap_seconds": 0.05
},
"seed": 1166
},
{
"name": "s6-music-long-335s-s1166",
"scene": "s6-narration",
"workflow": "music-long-s6.json",
"arguments": {},
"seed": 1166
},
{
"name": "s6-music-long-335s-s1167",
"scene": "s6-narration",
"workflow": "music-long-s6.json",
"arguments": {},
"seed": 1167
},
{
"name": "s6-music-long-335s-xf3000-s1166",
"scene": "s6-narration",
"workflow": "music-long-s6.json",
"arguments": {
"crossfade_ms": 3000
},
"seed": 1166
}
]
Loading
Loading