Long Sequence Aesthetic Quality

long_sequence_aesthetic_quality measures mean no-reference aesthetic appeal over prediction frames. Each frame $I_t$ is encoded by a frozen CLIP ViT-L/14 backbone; a LAION-Aesthetics linear head produces per-frame scores normalized to $[0,1]$.

$$S_{\mathrm{aes}} = \frac{1}{T}\sum_{t=1}^{T} s_t^{\mathrm{aes}} \tag{1}$$

Higher is better. Long-rollout quality signal computed prediction-only via the long_sequence backend (aesthetic_quality dimension). See Long-Sequence Metrics for the shared runtime.

60-Second Same-Input Examples

Audit manifest: docs/assets/metric_examples/longseq_memory_60s_examples_20260909.json.

Input image

60s memory_loop · same benchmark case · gap 0.2110

memory-dolly_loop-memory_pool_190

Suite: memory_loopScene: stylized / outdoor / terrestrial_landscape

Benchmark prompt target

Move the camera forward through the scene, then back to where it started.

prompt source: memory_loop_prompt

The better example stays visually appealing over the 60s rollout; the weaker one becomes flat, noisy, or collapsed.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 0.6090 · official

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.3981 · official

Input image

60s memory_loop · same benchmark case · gap 0.2498

memory-yaw_loop_360-memory_pool_132

Suite: memory_loopScene: stylized / outdoor / terrestrial_landscape

Benchmark prompt target

Rotate the camera a full turn in place and come back to the starting view.

prompt source: memory_loop_prompt

The better example stays visually appealing over the 60s rollout; the weaker one becomes flat, noisy, or collapsed.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 0.6435 · official

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.3937 · official

Input image

60s memory_loop · same benchmark case · gap 0.2037

memory-yaw_loop_180-memory_pool_439

Suite: memory_loopScene: stylized / outdoor / aquatic_landscape

Benchmark prompt target

Turn the camera to the right, then turn back to the starting view.

prompt source: memory_loop_prompt

The better example stays visually appealing over the 60s rollout; the weaker one becomes flat, noisy, or collapsed.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 0.6080 · official

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.4043 · official

Input image

60s memory_loop · same benchmark case · gap 0.2006

memory-palindrome_loop-memory_pool_335

Suite: memory_loopScene: stylized / indoor / dining_spaces

Benchmark prompt target

Move forward, turn and continue, then retrace the whole path back.

prompt source: memory_loop_prompt

The better example stays visually appealing over the 60s rollout; the weaker one becomes flat, noisy, or collapsed.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 0.5995 · official

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.3989 · official

Input image

60s memory_loop · same benchmark case · gap 0.1957

memory-double_yaw_loop-memory_pool_414

Suite: memory_loopScene: photorealistic / outdoor / terrestrial_landscape

Benchmark prompt target

Turn away and back to the start twice in a row.

prompt source: memory_loop_prompt

The better example stays visually appealing over the 60s rollout; the weaker one becomes flat, noisy, or collapsed.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 0.5756 · official

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.3799 · official

References

LAION-Aesthetics Predictor (CLIP-based linear aesthetic regressor)

Used via pyiqa laion_aes; CLIP ViT-L/14 features with a linear head trained on human aesthetic ratings.