[mvanhorn]

Summary

FLUX.2 reference-image generation can fail with a CUDA/CPU device mismatch when the VAE runs on CPU. Generation without a reference image can still succeed. Reference-image encoding selected the global default device instead of the VAE's compute device. This could place image tensors on CUDA while the VAE weights remained on CPU, and use the wrong device for tiling and memory estimates.

Reference-image encoding now uses vae_info.compute_device. Regression tests cover a CPU VAE with a mocked CUDA default, checking latent output and device selection for memory estimates and conditional tiling. Another test checks that encoding failures preserve the shared VAE's tiling settings and propagate the original exception.

Related Issues / Discussions

Closes #10204

QA Instructions

./scripts/family_e2e could not run in this environment for the InvokeAI encode_reference_image change (./scripts/family_e2e is not in this checkout).
uv run --frozen --project . pytest -q tests/backend/flux2/test_reference_encode_tiling.py failed during conftest loading because torch was unavailable; no tests ran.
./scripts/family_e2e could not run because the script was absent from the checkout.
A run with the InvokeAI encode_reference_image change reverted gave no clear result, so no test is shown to catch this bug.
The changed test tests/backend/flux2/test_reference_encode_tiling.py did not run here: the torch module is not installed.

Review

Not applicable to this change.

Compatibility / Rollout

Not applicable to this change.

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Meaningful regression coverage added / updated where needed; obsolete tests/code removed
• Persisted-state and API changes include required migrations / compatibility validation
• Relevant performance/efficiency opportunities considered; material claims have evidence
• Material review findings resolved and relevant checks rerun
• Documentation added / updated (if applicable)
• Updated What's New copy (if doing a release after this PR)

---

^{Open in [code]smith · Need help on this PR? Tag @codesmith-bot with what you need.}