fix(vace): stop firstlastframe log flood with config validation flag

[livepeer-tessa]

Problem

VaceEncodingBlock in firstlastframe mode raises a ValueError on every processed chunk when num_frame_per_block=1. Since ValueError is treated as recoverable by pipeline_processor, the pipeline keeps running and re-fires the same error every ~50ms — producing ~700 identical ERROR log lines in a 22-second session window.

Root: the check lived deep inside _build_extension_frames_and_masks, which is called per-chunk with no gate to short-circuit once a config error is known.

Fixes #685

Changes

1. Early validation in _encode_extension_mode (before the tensor-build path)

• Checks num_frame_per_block >= 2 for firstlastframe mode at the start of each encoding call
• Emits a clear, actionable message naming the config knob and suggesting alternatives (firstframe/lastframe with a single ref image)

2. _config_validation_error flag on the block instance

• Set to the error message string when validation fails on the first chunk
• __call__ checks the flag at entry; if set, returns early with vace_context=None (silent, DEBUG log)
• Only the first chunk ever reaches the log as an ERROR — subsequent chunks no-op
• Reset by clear_encoder_caches() so a corrected config works normally in a new session

3. Redundant ValueError → assert

• The original deep check in _build_extension_frames_and_masks is converted to an assert; it is now dead code for callers going through __call__, but remains a safety net for direct callers.

Behaviour After Fix

	Before	After
First chunk	ERROR logged	ERROR logged (same)
Subsequent chunks	ERROR × N per session	silent (DEBUG only)
Corrected config / new session	works	works (flag reset by clear_encoder_caches)

Related

• VaceEncodingBlock temporal kernel underflow in streamdiffusionv2: (2 x 64 x 64) input < (3 x 1 x 1) kernel #673 — same per-chunk flood pattern (temporal kernel underflow)
• VaceEncodingBlock resolution mismatch: input 384x384 vs target 512x512 #601 — same cascading error pattern (resolution mismatch)

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 33487a6e-7b00-4173-b40f-958aa4611866

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/vace-firstlastframe-log-flood

📝 Coding Plan

• Generate coding plan for human review comments

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Tip

You can disable the changed files summary in the walkthrough.

Disable the reviews.changed_files_summary setting to disable the changed files summary in the walkthrough.

[github-actions]

🚀 fal.ai Preview Deployment

App ID	daydream/scope-pr-687--preview
WebSocket	wss://fal.run/daydream/scope-pr-687--preview/ws
Commit	f756405

Testing

Connect to this preview deployment by running this on your branch:

uv run build && SCOPE_CLOUD_APP_ID="daydream/scope-pr-687--preview/ws" uv run daydream-scope

🧪 E2E tests will run automatically against this deployment.

[github-actions]

✅ E2E Tests passed

Status	passed
fal App	daydream/scope-pr-687--preview
Run	View logs

Test Artifacts

Check the workflow run for screenshots.