# Multi-Reward Alignment on Parti-Prompts

Source: [Multi-Reward Alignment on Parti-Prompts](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-14T08:46:41.236Z

Evaluation of image generation quality on the Parti-Prompts dataset using multiple reward signals, including CLIP, ICT, Aesthetic score, HPS, PickScore, ImgReward, and HP.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Multi-Reward Alignment |
| Dataset (including any stated split) | Parti-Prompts |
| Metric | CLIP Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 16 of 16 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [CLIP Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=CLIP%20Score>): Higher is better; unit: SCALAR.

- [ICT Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=ICT%20Score>): Higher is better; unit: SCALAR.

- [Aesthetic Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=Aesthetic%20Score>): Higher is better; unit: SCALAR.

- [HPS Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=HPS%20Score>): Higher is better; unit: SCALAR.

- [PickScore](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=PickScore>): Higher is better; unit: SCALAR.

- [ImgReward Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=ImgReward%20Score>): Higher is better; unit: SCALAR.

- [HP Score](<https://www.sota2.com/research/sota/multi-reward-alignment-on-parti-prompts.md?metric=HP%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | CLIP Score | Result date | Report paper |
| --- | --- | --- | --- |
| CLIP | 0.3554 | Not recorded | 6a055740dbc836b87d828e6e |
| Multi-Reward Joint Optimization | 0.3552 | Not recorded | 6a055740dbc836b87d828e6e |
| Separate-Cons | 0.3549 | Not recorded | 6a055740dbc836b87d828e6e |
| ICT | 0.3548 | Not recorded | 6a055740dbc836b87d828e6e |
| Weighted-Sum | 0.3546 | Not recorded | 6a055740dbc836b87d828e6e |
| Multi-Reward Joint Optimization | 0.3545 | Not recorded | 6a055740dbc836b87d828e6e |
| Multi-Reward Joint Optimization | 0.3544 | Not recorded | 6a055740dbc836b87d828e6e |
| Reward Soup | 0.3543 | Not recorded | 6a055740dbc836b87d828e6e |
| Multi-Reward Joint Optimization | 0.3541 | Not recorded | 6a055740dbc836b87d828e6e |
| Reward Soup | 0.3541 | Not recorded | 6a055740dbc836b87d828e6e |
| Reward Soup | 0.3539 | Not recorded | 6a055740dbc836b87d828e6e |
| HP | 0.3538 | Not recorded | 6a055740dbc836b87d828e6e |
| Reward Soup | 0.3537 | Not recorded | 6a055740dbc836b87d828e6e |
| PG-OT | 0.3534 | Not recorded | 6a055740dbc836b87d828e6e |
| HPS | 0.3501 | Not recorded | 6a055740dbc836b87d828e6e |
| SD3.5-Turbo | 0.3372 | Not recorded | 6a055740dbc836b87d828e6e |

### CLIP

CLIP Score: 0.3554. Result ID: 6a058bdc47f34ce924eca48d.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa69581e209131cea21ad0.md>)

[Implementation](<https://github.com/OpenAI/CLIP>)

Recorded evaluation configuration: {"Optimization Strategy":"Single-Reward Optimization"}. This record does not assert independent reproduction.

### Multi-Reward Joint Optimization

CLIP Score: 0.3552. Result ID: 6a058bdc47f34ce924eca491.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS ratios","Weight Ratios (ICT:HP:CLIP:HPS)":"2:2:3:3"}. This record does not assert independent reproduction.

### Separate-Cons

CLIP Score: 0.3549. Result ID: 6a058bdc47f34ce924eca498.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Multi-Reward under Heterogeneous Bounds (w/o OT)"}. This record does not assert independent reproduction.

### ICT

CLIP Score: 0.3548. Result ID: 6a058bdc47f34ce924eca48b.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a381e209131cea81e34.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Single-Reward Optimization"}. This record does not assert independent reproduction.

### Weighted-Sum

CLIP Score: 0.3546. Result ID: 6a058bdc47f34ce924eca497.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Multi-Reward under Heterogeneous Bounds (w/o OT)"}. This record does not assert independent reproduction.

### Multi-Reward Joint Optimization

CLIP Score: 0.3545. Result ID: 6a058bdc47f34ce924eca48f.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS ratios","Weight Ratios (ICT:HP:CLIP:HPS)":"1:1:1:1"}. This record does not assert independent reproduction.

### Multi-Reward Joint Optimization

CLIP Score: 0.3544. Result ID: 6a058bdc47f34ce924eca490.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS ratios","Weight Ratios (ICT:HP:CLIP:HPS)":"2:3:2:3"}. This record does not assert independent reproduction.

### Reward Soup

CLIP Score: 0.3543. Result ID: 6a058bdc47f34ce924eca493.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs","Weight Ratios (ICT:HP:CLIP:HPS)":"1:1:1:1"}. This record does not assert independent reproduction.

### Multi-Reward Joint Optimization

CLIP Score: 0.3541. Result ID: 6a058bdc47f34ce924eca492.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS ratios","Weight Ratios (ICT:HP:CLIP:HPS)":"4:4:1:1"}. This record does not assert independent reproduction.

### Reward Soup

CLIP Score: 0.3541. Result ID: 6a058bdc47f34ce924eca496.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs","Weight Ratios (ICT:HP:CLIP:HPS)":"3:2:0:0"}. This record does not assert independent reproduction.

### Reward Soup

CLIP Score: 0.3539. Result ID: 6a058bdc47f34ce924eca494.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs","Weight Ratios (ICT:HP:CLIP:HPS)":"1:1:4:4"}. This record does not assert independent reproduction.

### HP

CLIP Score: 0.3538. Result ID: 6a058bdc47f34ce924eca48c.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a381e209131cea81e34.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Single-Reward Optimization"}. This record does not assert independent reproduction.

### Reward Soup

CLIP Score: 0.3537. Result ID: 6a058bdc47f34ce924eca495.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Weighted ICT:HP:CLIP:HPS fusion of single-reward LoRAs","Weight Ratios (ICT:HP:CLIP:HPS)":"3:2:1:4"}. This record does not assert independent reproduction.

### PG-OT

CLIP Score: 0.3534. Result ID: 6a058bdc47f34ce924eca499.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Pareto-Frontier-Guided Optimal Transport"}. This record does not assert independent reproduction.

### HPS

CLIP Score: 0.3501. Result ID: 6a058bdc47f34ce924eca48e.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa70b41e209131cea4caeb.md>)

Recorded evaluation configuration: {"Optimization Strategy":"Single-Reward Optimization"}. This record does not assert independent reproduction.

### SD3.5-Turbo

CLIP Score: 0.3372. Result ID: 6a058bdc47f34ce924eca48a.

[Reporting paper](<https://www.sota2.com/research/paper/6a055740dbc836b87d828e6e.md>)

Recorded evaluation configuration: Not recorded. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
