[WIP][CI][Accuracy] Add HunyuanImage3 pixel accuracy test and nightly CI

[BLANKETusers]

Summary

• Add assert_images_pixel_close helper for full-image pixel-level comparison
  with mean/p99 absolute channel difference metrics and detailed diagnostics
• Add test_hunyuan_image3_pixel_accuracy that generates images via offline
  end2end.py and compares output against a pre-saved baseline image
• Add nightly CI step (4× H100) in the Diffusion X2I group to gate pixel
  accuracy regressions
• Rename diffusers_image → baseline_image across accuracy helper APIs
  (assert_similarity, assert_image_sequence_similarity)

Files changed

File	Change
.buildkite/test-nightly.yml	+38 lines: new CI step vllm-omni · HunyuanImage3 · Accuracy Test
tests/assets/hunyuan/hunyuan_baseline.png	Baseline reference image (1024×1024)
tests/e2e/accuracy/helpers.py	+68 lines: assert_images_pixel_close; rename params
tests/e2e/accuracy/test_hunyuan_image3_pixel_accuracy.py	+142 lines: new test

Test plan

HUNYUAN_IMAGE3_DEPLOY_CONFIG=../hunyuan_image3_dit_copy.yaml pytest -s -v tests/e2e/accuracy/test_hunyuan_image3_pixel_accuracy.py --run-level

Test Result

1 passed, 18 warnings in 101.49s (0:01:41)

Pixel Metrics

tencent/HunyuanImage-3.0-Instruct — (offline vs baseline)

Metric	Value	Threshold	Status
mean_abs_diff	0.000000	≤ 0.02	✓
p99_abs_diff	0.000000	≤ 0.10	✓
p50	0.000000	—	—
p90	0.000000	—	—
p95	0.000000	—	—
p99	0.000000	—	—
p99.9	0.000000	—	—

Mismatch ratios (pixel_ratio / channel_ratio)

Threshold (1/255)	0	1	2	4	8	16	32	64	128
pixel_ratio	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
channel_ratio	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Credits must be used to enable repository wide code reviews.