#602 stage B: rewire upscale-clip and refine-clip through fit/restore

Stage B of #602. Section Stages → Stage B: rewire upscale-clip and refine-clip of plan v4; the plan comment stays authoritative, including The design, Templates (stage B), Risks and Decisions. Needs stage A (#631, verified).

Filed by the lead agent, model claude-opus-5-5 via anthropic. Reshaped for plan v4 (reconciled 2026-10-06 by the same model): v2's ref_width/ref_height breaking change is gone (Q2′). upscale-clip keeps width/height as the output size, and the stage now also changes fit_to_model itself: a downscale argument, and video as an fps-carrying float32 ndarray. No breaking-change.

What it builds

In dw/tasks/fit.py:
- fit_to_model gains downscale: an optional positive integer, default 1. It fits into width/downscale × height/downscale and records that as the model size. It refuses, naming both values, when width or height isn't divisible by it. Its get_task description says so.
- fit_to_model's video becomes a float32 numpy.ndarray subclass carrying .fps (in dw/media_types.py, alongside AudioVideo), in place of an AudioVideo. Then previous_result:fit.video feeds a pipeline argument or a condition's frames directly.
- Unit tests for both. One realizes previous_result:fit.video into a pipeline's arguments and a condition's frames through the real argument path, not a mock. Write that one first. If a condition's frames still won't take it, stop and re-plan (plan Risks); don't bend the engine.
In both upscale-clip and refine-clip, the same step sequence: fit (fit_to_model, mode from a new fit variable, default letterbox) → the pipeline fed previous_result:fit.video → restore (restore_to_source, fit: previous_result:fit.fit) → the existing pair_audio.
upscale-clip: width/height stay the output size (default 960×544, 32n) and go to the pipeline as today. fit gets width: variable:width, height: variable:height, downscale: 2, so the reference is 480×272, and diffusers' crop becomes a no-op. The IC reference is previous_result:fit.video in place of variable:source_video.
refine-clip: fit to width×height. loop_frames is replaced, and the upsampler's stretch becomes a no-op.
If the GPU runs show an edge halo at the letterbox edge, flip the default to stretch (Q1).

What it owes

Rewritten descriptions and summaries in both templates, within the COMPACT_BUDGET (tests/test_catalog_structure.py:411). They state the changed behaviour: a different aspect is letterboxed and restored, a short source is held and trimmed, and the output is exactly 2× the source at the source's length. They also say which size each template's variables name: the output for upscale-clip, the working size for refine-clip.
workflows/templates/ltx2/README.md, plugins/dw/skills/ltx-2.5/SKILL.md with a plugin version bump, and docs/RECIPES_24GB.md:186.
Updated tests that pin today's shape: tests/test_ltx2_ic_loras.py (:212, :174-185 for upscale-clip only, :216), tests/test_catalog_structure.py:566-645 (TestLtxRefineClip), and the tests/test_step_cache.py:706-726 hashes if the configuration changes.

Deploy

Server, plus a plugin release.

Passes when

Its own repro works, per the plan's stage B acceptance intent:

both templates validate clean on a 640×480, 50-frame, 24 fps asset with a soundtrack;
final/ is 1280×960, 50 frames, 24 fps, with audio at the source's length, no edge content lost, no bars, and no lapped tail;
intermediate/ shows the model frame with bars;
a source already at the working size gives the same output size as before;
downscale: get_task shows it, width=512, height=288, downscale=2 gives a 256×144 fit with that model size in the record, and downscale=3 with width=512 is refused naming both;
the existing refusals are unchanged;

and every pending: #632 case the tester writes passes.

After it lands (lead files to Don)

The Q3 follow-up (adopt the pair in restore-deblur / restore-decompression), once letterbox holds up in these runs.
Removal of loop_frames as a proposal, if no template consumer is left.

Stage B stopped before any code: the approved plan (v2) doesn't hold against the code. The approval on #602 is withdrawn, and plan v3 goes to Don for review. No branch was pushed, because nothing was built.

Two things break:

Q2's ref_width/ref_height shape can't be wired. LTX2InContextPipeline takes width/height at the output size and derives the reference size itself: ref_width = width // reference_downscale_factor (pipeline_ltx2_ic_lora.py:1403-1404). With only ref_width/ref_height variables, nothing can pass it 2 × ref_width:
- workflow JSON has no expressions;
- no task does arithmetic;
- variable_constraints has no rule relating two variables (SNAP_FIELDS, dw/variable_constraints.py:57).
Setting reference_downscale_factor: 1 instead would give a 1× output from a LoRA trained for 2×.
previous_result:fit.video can't feed a pipeline.
- It resolves to the AudioVideo that fit_to_model returns (dw/tasks/fit.py:183).
- Nothing unwraps an AudioVideo for a diffusers argument or for a condition's frames. generative-upscale and window_video consumers read .frames off a bare AudioVideo, but here it sits inside a dict, and get_artifact_properties doesn't walk video.frames (dw/result.py:217).
- The plan's named fallback (save fit.video as a hidden intermediate and reference it) doesn't work either. output: only names earlier runs, and an h264 round trip would lose detail before a 2× re-render.

Plan v3 proposes a fix for each on #602. This stage waits for Don's decision on it.

Lead agent, model claude-opus-5-5 via anthropic.

Reshaped for plan v4 (#602). The body now says the stage also changes fit_to_model: a downscale argument, and video as an fps-carrying float32 ndarray. upscale-clip keeps width/height as the output size, with downscale: 2 on fit. v2's ref_width/ref_height is gone, so the breaking-change label is removed. A downscale acceptance item is added. — lead agent, claude-opus-5-5 via anthropic

test: #632 - pin the fit step in both 2x templates, downscale and Fit…

feat(catalog): #632 - 2x template descriptions, ltx2 README, skill, T…

Merge branch 'feat/602-B-fit-restore-templates' into develop (#632)

Merge branch 'feat/602-B-downscale-names-side' into develop (#632)

Bounce fix shipped: develop @ abae46d4 (merge of feat/602-B-downscale-names-side). The deploy log ended deployed develop @ abae46d4, and the server came back healthy ("version":"0.10.0-alpha.1" mcp=true).

C-F320 arms 1 and 2: the downscale refusal now names only the side or sides that downscale doesn't divide. The same sentence (fit_downscale_problem, dw/task_domains.py) is used at validate and at run time:

512×288, downscale 3: fit_to_model needs 'width' 512 to be divisible by 'downscale' 3
515×288, downscale 5: fit_to_model needs 'height' 288 to be divisible by 'downscale' 5
both sides failing, e.g. 512×290, downscale 3, names both: 'width' 512 and 'height' 290.

The refusal is still reported at steps[N].task.arguments.downscale.

Tests: tests/test_fit_to_model.py now pins each arm at run time and at validate. The tests check that the side that divides isn't named. Full suite: 9369 passed. ruff is clean.

Not changed: the stray double period on the "two" refusal (arm 3, which you called cosmetic). It comes from the shared domain message, not from this stage. Nothing else changed since the last hand-off, and there are no deviations from the plan.

The GPU cases C-F246 and C-F247 passed last verify. C-F321 and C-F322 are still to run.

— lead agent, claude-opus-5-5 via anthropic

Architecture review queued by the loop driver: this hand-off changed dw/ or dw_mcp/ on develop (2cb94ba530..abae46d4ad: dw/task_domains.py). The reviewer checks it against docs/ARCHITECTURE.md before the tester verifies.

Architecture review: pass. Reviewer agent, model claude-opus-5-5 via anthropic.

I read git diff 2cb94ba5..abae46d4 -- dw/ dw_mcp/. The only file it touches is dw/task_domains.py (commit b1b5944), and I checked it against docs/ARCHITECTURE.md and stage-c item 3.

Second owner: none. Only the message text of fit_downscale_problem changed: it now lists the failing side or sides instead of the WxH pair. That function is still the one place the sentence is built. fit_to_model_errors (dw/task_domains.py:1140) calls it at validate and fit_to_model (dw/tasks/fit.py:148) calls it at run time, which follows the Task argument domains row. No other copy of the rule or the message exists under dw/.
Build vs buy: none. It's a list comprehension and a join, so no dependency applies.
Map: no missing row. No module was added, and no rule moved between modules.
Context: clean. The docstring now says why only the failing side is named. No CLAUDE.md changed in the range.

This issue has no arch-approved label, and nothing here needed one. #632 goes back to the tester for verification.

Verified. Tester, claude-opus-5-5 via anthropic, against lem develop @ 88fb24d (contains abae46d4), workspace regression-complete.

(This replaces the text first posted here. That was a stale copy of the previous verify's bounce, posted by mistake.)

The bounce, C-F320. All arms checked with validate_workflow on an inline fit_to_model workflow. Each one was refused at steps[0].task.arguments.downscale.

Arm 1, 512×288 with downscale 3: fit_to_model needs 'width' 512 to be divisible by 'downscale' 3. Height isn't named.
Arm 2, 515×288 with downscale 5: fit_to_model needs 'height' 288 to be divisible by 'downscale' 5. Width isn't named.
Adjacent case, 512×290 with downscale 3: needs 'width' 512 and 'height' 290 to be divisible by 'downscale' 3.
Adjacent case, arm 2 with downscale: "variable:ds" and ds=5 passed as an argument: the same height-only refusal, also at validate.
Arm 3, downscale 0: needs 'downscale' above zero, got 0. Arm 3, downscale 1.5: needs 'downscale' as a whole number, got 1.5. The -2 and "two" arms passed last verify.

C-F321, both runs with source asset:qa-fit/src-512x288-121f.mp4 and seed 42.

refine-clip with defaults, job 6a7534f947be: succeeded.
- final/ is 1024×576, 121 frames, 24 fps, audio 5.0417 s.
- intermediate/ is 1024×576, 121 frames. Frame 0 shows the full picture with no bars.
upscale-clip at 1024×576 and 121 frames, job 1b2743f7a2bd: succeeded with no warnings.
- final/ is 1024×576, 121 frames, 24 fps, audio 5.0417 s.
- intermediate/ is 1024×576, 121 frames. Frame 0 shows the full picture with no bars.
Both runs deleted.

C-F322. I read the full get_workflow description of both templates, the list_workflows summaries, get_task("fit_to_model") and the dw:ltx-2-5 skill.

All of them describe letterbox by default (fit picks stretch or crop) with the padding cut back off, a short source held and trimmed, and output exactly 2× the source at the source's length.
upscale-clip: width/height are "the output size the model renders", and the source is fitted to half of it.
refine-clip: width/height are "the working size", and the model renders at twice that.
The skill gives the same split: "render size" for upscale-clip, "working size" for refine-clip.
None of these texts mentions ref_width/ref_height, asks the caller to match the source size, says centre-crop, stretch or lap, or calls anything breaking or renamed.

C-F246, C-F247, C-F249, C-F318 and C-F319 passed on the previous verify, and nothing outside fit_downscale_problem's message changed since.

Suite. I removed the pending: #632 lines from C-F246, C-F247, C-F249 and C-F318–C-F322. C-F245, C-F248 and C-F250 stay pending: they are plan-v2 specs, and their retirement is already requested in dkackman/harnest#60 and #65.

— tester, claude-opus-5-5 via anthropic

feat(tasks): dkackman#632 - fit_to_model downscale and a fitted video…

feat(tasks): dkackman#632 - fit_to_model's downscale refusal names on…

Navigation Menu

Repository navigation

#602 stage B: rewire upscale-clip and refine-clip through fit/restore #632

What it builds

What it owes

Deploy

Passes when

After it lands (lead files to Don)

10 remaining items

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

#602 stage B: rewire upscale-clip and refine-clip through fit/restore #632

Description

What it builds

What it owes

Deploy

Passes when

After it lands (lead files to Don)

Activity

dkackman commented on Oct 6, 2026

dkackman commented on Oct 6, 2026

10 remaining items

dkackman commented on Oct 6, 2026

dkackman commented on Oct 6, 2026

dkackman commented on Oct 6, 2026

dkackman commented on Oct 6, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions