# Prompt Stealing on MS COCO 50 random prompts Source: [Prompt Stealing on MS COCO 50 random prompts]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:38:32.632Z Evaluation of model performance on the Prompt Stealing task using a subset of 50 random prompts derived from the MS COCO dataset, measured using CLIP, LPIPS, and SBERT similarity scores. 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 | Prompt Stealing | | Dataset (including any stated split) | MS COCO 50 random prompts | | Metric | CLIP Similarity | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 28 of 28 | | 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 Similarity](): Higher is better; unit: SCALAR. - [LPIPS](): Lower is better; unit: SCALAR. - [SBERT Similarity](): Higher is better; unit: SCALAR. ## Recorded results | Method | CLIP Similarity | Result date | Report paper | | --- | --- | --- | --- | | PROMPTMINER | 0.958 | Not recorded | 692d2108d6b0dd3342931618 | | PROMPTMINER | 0.953 | Not recorded | 692d2108d6b0dd3342931618 | | PROMPTMINER | 0.934 | Not recorded | 692d2108d6b0dd3342931618 | | PROMPTMINER | 0.933 | Not recorded | 692d2108d6b0dd3342931618 | | VLM-as-expert | 0.923 | Not recorded | 692d2108d6b0dd3342931618 | | VLM-as-expert | 0.913 | Not recorded | 692d2108d6b0dd3342931618 | | PromptStealer | 0.898 | Not recorded | 692d2108d6b0dd3342931618 | | CLIP-IG | 0.887 | Not recorded | 692d2108d6b0dd3342931618 | | BLIP | 0.885 | Not recorded | 692d2108d6b0dd3342931618 | | CLIP-IG | 0.883 | Not recorded | 692d2108d6b0dd3342931618 | | BLIP | 0.881 | Not recorded | 692d2108d6b0dd3342931618 | | CLIP-IG | 0.874 | Not recorded | 692d2108d6b0dd3342931618 | | VLM-as-expert | 0.866 | Not recorded | 692d2108d6b0dd3342931618 | | PromptStealer | 0.866 | Not recorded | 692d2108d6b0dd3342931618 | | CLIP-IG | 0.862 | Not recorded | 692d2108d6b0dd3342931618 | | VLM-as-expert | 0.861 | Not recorded | 692d2108d6b0dd3342931618 | | PromptStealer | 0.861 | Not recorded | 692d2108d6b0dd3342931618 | | PromptStealer | 0.861 | Not recorded | 692d2108d6b0dd3342931618 | | BLIP | 0.858 | Not recorded | 692d2108d6b0dd3342931618 | | BLIP | 0.855 | Not recorded | 692d2108d6b0dd3342931618 | | VGD | 0.816 | Not recorded | 692d2108d6b0dd3342931618 | | VGD | 0.802 | Not recorded | 692d2108d6b0dd3342931618 | | VGD | 0.791 | Not recorded | 692d2108d6b0dd3342931618 | | VGD | 0.785 | Not recorded | 692d2108d6b0dd3342931618 | | PH2P | 0.673 | Not recorded | 692d2108d6b0dd3342931618 | | PH2P | 0.645 | Not recorded | 692d2108d6b0dd3342931618 | | PH2P | 0.642 | Not recorded | 692d2108d6b0dd3342931618 | | PH2P | 0.641 | Not recorded | 692d2108d6b0dd3342931618 | ### PROMPTMINER CLIP Similarity: 0.958. Result ID: 6995e73c3badfaffe6ef7473. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### PROMPTMINER CLIP Similarity: 0.953. Result ID: 6995e73c3badfaffe6ef7474. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### PROMPTMINER CLIP Similarity: 0.934. Result ID: 6995e73c3badfaffe6ef7472. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### PROMPTMINER CLIP Similarity: 0.933. Result ID: 6995e73c3badfaffe6ef7471. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### VLM-as-expert CLIP Similarity: 0.923. Result ID: 6995e73c3badfaffe6ef7463. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### VLM-as-expert CLIP Similarity: 0.913. Result ID: 6995e73c3badfaffe6ef7464. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### PromptStealer CLIP Similarity: 0.898. Result ID: 6995e73c3badfaffe6ef746b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### CLIP-IG CLIP Similarity: 0.887. Result ID: 6995e73c3badfaffe6ef745f. [Reporting paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### BLIP CLIP Similarity: 0.885. Result ID: 6995e73c3badfaffe6ef745b. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### CLIP-IG CLIP Similarity: 0.883. Result ID: 6995e73c3badfaffe6ef7460. [Reporting paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### BLIP CLIP Similarity: 0.881. Result ID: 6995e73c3badfaffe6ef745c. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### CLIP-IG CLIP Similarity: 0.874. Result ID: 6995e73c3badfaffe6ef745d. [Reporting paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### VLM-as-expert CLIP Similarity: 0.866. Result ID: 6995e73c3badfaffe6ef7462. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### PromptStealer CLIP Similarity: 0.866. Result ID: 6995e73c3badfaffe6ef746c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### CLIP-IG CLIP Similarity: 0.862. Result ID: 6995e73c3badfaffe6ef745e. [Reporting paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### VLM-as-expert CLIP Similarity: 0.861. Result ID: 6995e73c3badfaffe6ef7461. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### PromptStealer CLIP Similarity: 0.861. Result ID: 6995e73c3badfaffe6ef7469. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### PromptStealer CLIP Similarity: 0.861. Result ID: 6995e73c3badfaffe6ef746a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### BLIP CLIP Similarity: 0.858. Result ID: 6995e73c3badfaffe6ef7459. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### BLIP CLIP Similarity: 0.855. Result ID: 6995e73c3badfaffe6ef745a. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### VGD CLIP Similarity: 0.816. Result ID: 6995e73c3badfaffe6ef746d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### VGD CLIP Similarity: 0.802. Result ID: 6995e73c3badfaffe6ef746e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### VGD CLIP Similarity: 0.791. Result ID: 6995e73c3badfaffe6ef746f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. ### VGD CLIP Similarity: 0.785. Result ID: 6995e73c3badfaffe6ef7470. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### PH2P CLIP Similarity: 0.673. Result ID: 6995e73c3badfaffe6ef7465. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion v1.5"}. This record does not assert independent reproduction. ### PH2P CLIP Similarity: 0.645. Result ID: 6995e73c3badfaffe6ef7466. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"SDXL-Turbo"}. This record does not assert independent reproduction. ### PH2P CLIP Similarity: 0.642. Result ID: 6995e73c3badfaffe6ef7468. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"Stable Diffusion 3.5 Medium"}. This record does not assert independent reproduction. ### PH2P CLIP Similarity: 0.641. Result ID: 6995e73c3badfaffe6ef7467. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Target Generative Model":"FLUX.1 dev"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.