fix: cast block_seed to int before manual_seed()

[livepeer-tessa]

Summary

Cast block_seed to int before passing to torch.Generator.manual_seed() in both PrepareVideoLatentsBlock and PrepareLatentsBlock.

Problem

manual_seed() requires a long/int, but base_seed can arrive as a float when deserialized from JSON, causing:

RuntimeError: manual_seed expected a long, but got float

Fix

One-line change in each block: int(base_seed + block_state.current_start_frame)

Fixes #618

Summary by CodeRabbit

• Bug Fixes
  • Fixed seed calculation in video latent generation to ensure proper type handling, improving reliability and consistency in random number generation for video processing workflows.

[coderabbitai]

📝 Walkthrough

Walkthrough

Two pipeline blocks in the WAN2.1 pipeline are updated to explicitly cast block_seed to an integer before passing it to torch.Generator.manual_seed(). The seed is computed by adding base_seed and block_state.current_start_frame, with the result now wrapped in int() to ensure type consistency.

Changes

Cohort / File(s)	Summary
Block Seed Type Casting src/scope/core/pipelines/wan2_1/blocks/prepare_latents.py, src/scope/core/pipelines/wan2_1/blocks/prepare_video_latents.py	Cast block_seed to int() when computed from base_seed and current_start_frame. Ensures torch.Generator.manual_seed() receives an integer instead of float, resolving RuntimeError: manual_seed expected a long, but got float.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Poem

🐰 A seed cast true, no float in sight,
Where torch.Generator finds delight,
An int, not float, the fix takes wing,
Now latents prepare without a sting! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately summarizes the main change: casting block_seed to int before calling manual_seed(), which directly addresses the core issue.
Linked Issues check	✅ Passed	The PR fully addresses issue #618 by casting block_seed to int in both PrepareVideoLatentsBlock and PrepareLatentsBlock before calling manual_seed().
Out of Scope Changes check	✅ Passed	All changes are directly related to fixing the block_seed float casting issue in the two affected blocks; no out-of-scope modifications detected.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

• 📝 Generate docstrings (stacked PR)
• 📝 Generate docstrings (commit on current branch)

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch fix/manual-seed-float-cast

---

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

[coderabbitai]

🧹 Nitpick comments (1)

src/scope/core/pipelines/wan2_1/blocks/prepare_latents.py (1)

93-93: Validate base_seed before truncating it.

int(base_seed + block_state.current_start_frame) fixes 42.0, but it also silently turns values like 42.7 into 42. Since base_seed is declared as an integer input, it would be safer to normalize or reject non-integral floats once when the state/config is built, then keep integer arithmetic here.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@src/scope/core/pipelines/wan2_1/blocks/prepare_latents.py` at line 93,
base_seed is being truncated silently by int(...) in the prepare_latents block
(block_seed = int(base_seed + block_state.current_start_frame)); validate and
normalize base_seed earlier (when the config/state is constructed) so it is
guaranteed to be an integer or explicitly reject non-integral floats, then use
integer arithmetic here without int() casting; update the config/state
validation routine that sets base_seed to either raise on non-integer floats or
coerce using a clear normalization (e.g., round or floor) and add a unit test
asserting prepare_latents sees only integer base_seed.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@src/scope/core/pipelines/wan2_1/blocks/prepare_latents.py`:
- Line 93: base_seed is being truncated silently by int(...) in the
prepare_latents block (block_seed = int(base_seed +
block_state.current_start_frame)); validate and normalize base_seed earlier
(when the config/state is constructed) so it is guaranteed to be an integer or
explicitly reject non-integral floats, then use integer arithmetic here without
int() casting; update the config/state validation routine that sets base_seed to
either raise on non-integer floats or coerce using a clear normalization (e.g.,
round or floor) and add a unit test asserting prepare_latents sees only integer
base_seed.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 40d98101-86a0-416d-a32b-58b1d26b257f

📥 Commits

Reviewing files that changed from the base of the PR and between 0a52f7c and 56cfb8a.

📒 Files selected for processing (2)

• src/scope/core/pipelines/wan2_1/blocks/prepare_latents.py
• src/scope/core/pipelines/wan2_1/blocks/prepare_video_latents.py

[github-actions]

🚀 fal.ai Preview Deployment

App ID	daydream/scope-pr-620--preview
WebSocket	wss://fal.run/daydream/scope-pr-620--preview/ws
Commit	56cfb8a

Testing

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

uv run build && SCOPE_CLOUD_APP_ID="daydream/scope-pr-620--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-620--preview
Run	View logs

Test Artifacts

Check the workflow run for screenshots.