# Multi-Reference Image Generation on OmniRef-Bench

Source: [Multi-Reference Image Generation on OmniRef-Bench](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench>)

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

Data updated: 2026-06-26T08:47:57.263Z

Evaluation of Multi-Reference Image Generation models on the OmniRef-Bench dataset, assessed using objective and MLLM-based metrics covering style, background, lighting, and subject fidelity.

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-Reference Image Generation |
| Dataset (including any stated split) | OmniRef-Bench |
| Metric | Objective Subject Score |
| Unit | SCALAR |
| Direction | Higher is better |
| Results loaded | 9 of 9 |
| 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

- [Objective Subject Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Subject%20Score>): Higher is better; unit: SCALAR.

- [Objective Style Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Style%20Score>): Higher is better; unit: SCALAR.

- [Objective Background Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Background%20Score>): Higher is better; unit: SCALAR.

- [Objective Lighting Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Lighting%20Score>): Higher is better; unit: SCALAR.

- [Objective Pose Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Pose%20Score>): Higher is better; unit: SCALAR.

- [Objective Average Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=Objective%20Average%20Score>): Higher is better; unit: SCALAR.

- [MLLM Subject Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Subject%20Score>): Higher is better; unit: SCALAR.

- [MLLM Background Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Background%20Score>): Higher is better; unit: SCALAR.

- [MLLM Style Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Style%20Score>): Higher is better; unit: SCALAR.

- [MLLM Lighting Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Lighting%20Score>): Higher is better; unit: SCALAR.

- [MLLM Aesthetic Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Aesthetic%20Score>): Higher is better; unit: SCALAR.

- [MLLM IF Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20IF%20Score>): Higher is better; unit: SCALAR.

- [MLLM Average Score](<https://www.sota2.com/research/sota/multi-reference-image-generation-on-omniref-bench.md?metric=MLLM%20Average%20Score>): Higher is better; unit: SCALAR.

## Recorded results

| Method | Objective Subject Score | Result date | Report paper |
| --- | --- | --- | --- |
| Seedream4.5 | 0.65 | Not recorded | 6a3e0791206542251d7cb496 |
| Nano Banana Pro | 0.64 | Not recorded | 6a3e0791206542251d7cb496 |
| FLUX.2 \[klein\] | 0.63 | Not recorded | 6a3e0791206542251d7cb496 |
| FLUX.2 \[klein\] + DyRef | 0.63 | Not recorded | 6a3e0791206542251d7cb496 |
| Qwen-2511 + DyRef | 0.62 | Not recorded | 6a3e0791206542251d7cb496 |
| OmniGen2 | 0.59 | Not recorded | 6a3e0791206542251d7cb496 |
| BAGEL | 0.58 | Not recorded | 6a3e0791206542251d7cb496 |
| DreamOmni2 | 0.57 | Not recorded | 6a3e0791206542251d7cb496 |
| Qwen-2511 | 0.55 | Not recorded | 6a3e0791206542251d7cb496 |

### Seedream4.5

Objective Subject Score: 0.65. Result ID: 6a3e3cbafbc373592c607bf5.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ba51e209131cea89983.md>)

Recorded evaluation configuration: {"Model Category":"Closed-Source Models","Evaluation MLLM":"Gemini 3 Flash"}. This record does not assert independent reproduction.

### Nano Banana Pro

Objective Subject Score: 0.64. Result ID: 6a3e3cbafbc373592c607bf6.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

Recorded evaluation configuration: {"Model Category":"Closed-Source Models","Evaluation MLLM":"Gemini 3 Flash"}. This record does not assert independent reproduction.

### FLUX.2 \[klein\]

Objective Subject Score: 0.63. Result ID: 6a3e3cbafbc373592c607bfc.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash","Backbone":"FLUX.2 \[klein\] 9B Base"}. This record does not assert independent reproduction.

### FLUX.2 \[klein\] + DyRef

Objective Subject Score: 0.63. Result ID: 6a3e3cbafbc373592c607bfd.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash","Optimization":"Dynamic Reward Optimization (DAR)","Backbone":"FLUX.2 \[klein\] 9B Base"}. This record does not assert independent reproduction.

### Qwen-2511 + DyRef

Objective Subject Score: 0.62. Result ID: 6a3e3cbafbc373592c607bfb.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash","Optimization":"Dynamic Reward Optimization (DAR)","Backbone":"Qwen-Image-Edit-2511"}. This record does not assert independent reproduction.

### OmniGen2

Objective Subject Score: 0.59. Result ID: 6a3e3cbafbc373592c607bf7.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ba31e209131cea89864.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash"}. This record does not assert independent reproduction.

### BAGEL

Objective Subject Score: 0.58. Result ID: 6a3e3cbafbc373592c607bf9.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a311e209131cea81aad.md>)

[Implementation](<https://bagel-ai.org/>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash"}. This record does not assert independent reproduction.

### DreamOmni2

Objective Subject Score: 0.57. Result ID: 6a3e3cbafbc373592c607bf8.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7bec1e209131cea8b53f.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash"}. This record does not assert independent reproduction.

### Qwen-2511

Objective Subject Score: 0.55. Result ID: 6a3e3cbafbc373592c607bfa.

[Reporting paper](<https://www.sota2.com/research/paper/6a3e0791206542251d7cb496.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7a6a1e209131cea83137.md>)

Recorded evaluation configuration: {"Model Category":"Open-Source Models","Evaluation MLLM":"Gemini 3 Flash","Backbone":"Qwen-Image-Edit-2511"}. This record does not assert independent reproduction.

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