Text-to-Image Generation on GenEval 2 (Efficiency/Cost Metrics)
34.16Prompt LengthFLUX.2-Klein-9B
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FLUX.2-Klein-9Bmode=single-pass, MLLM decoding temperature=0.152026.05 | 34.16 | — | 79.06 | 1 | — | — | — | — | — | |
| T2I-CopilotInference-time scaling=FLUX.2-Klein-9B, MLLM decoding temperature=0.15, Cost Evaluation Model=Qwen3-VL-32B2026.05 | 35.51 | 1.35 | 80.57 | 1.22 | 4.43 | 15 | 1.11 | 0.3 | 0.09 | |
| BoN+NVILAInference-time scaling=FLUX.2-Klein-9B, MLLM decoding temperature=0.152026.05 | 41.6 | 7.44 | 82.38 | 32 | — | — | 0.23 | — | — | |
| RAISEInference-time scaling=FLUX.2-Klein-9B, MLLM decoding temperature=0.15, Cost Evaluation Model=Qwen3-VL-32B2026.05 | 52.81 | 18.65 | 86.42 | 19.9 | 9.13 | 49.9 | 0.94 | 2.04 | 0.37 | |
| EPICInference-time scaling=FLUX.2-Klein-9B, MLLM decoding temperature=0.15, Cost Evaluation Model=Qwen3-VL-32B2026.05 | 68.79 | 34.63 | 89.38 | 13.9 | 2.48 | 9.1 | 2.49 | 13.96 | 3.79 |