Google AI Studio
Multimedia AI Benchmarks
Multimedia models with prompts matched to their tasks. Results, speed, and cost appear after each model run is completed and verified.
Results appear after each model's media and test record are verified.
BotConnector Benchmark: Matcha Mug & 3 Lemons
Transparent glass mug filled to brim with surface tension + exactly 3 whole yellow lemons on white marble
Why we chose this prompt: Tests extreme surface tension meniscus against coffee/tea prior biases and exact numeric counting (precisely 3 lemons).
STANDARDIZED TEST PROMPT
A transparent glass mug completely filled to the absolute brim with iced matcha latte, dark green liquid flush with the top edge and surface tension visible. Exactly three whole fresh yellow lemons sit directly to the right of the mug on a white marble tabletop. Crisp straight-on eye-level macro photo, neutral lighting, seamless soft grey studio background, no other objects in frame.
Showing 16 multimedia models for this prompt
Google AI Studio
Google AI Studio
FLUX
![FLUX.2 [klein] 4B](/img/benchmarks/results/flux-2-klein-4b/result.jpg)
FLUX
Alibaba Cloud
Alibaba Cloud
OpenAI
OpenAI
Google AI Studio
Stability AI
Stability AI
Recraft
Krea AI
Meta
Ming AI
Agnes AI
BotConnector Benchmark: Two-Liquid Density Boundary
Static laboratory shot showing a stable density boundary for at least five seconds
Why we chose this prompt: Tests physical law consistency, fluid mechanics boundary stability without morphing, and camera lock without cuts.
STANDARDIZED TEST PROMPT
A continuous locked static tripod shot of a clear glass measuring beaker on a lab desk. A blue liquid slowly drains out while simultaneously being filled with bright golden orange oil from a pipette above, creating two distinct separated liquid layers with a sharp horizontal boundary. Static camera, zero zoom or pan, steady clean lighting, with no cuts; maintain the scene for at least five seconds.
Showing 8 multimedia models for this prompt
Google Veo
Google Veo
Alibaba Wan
Alibaba Wan
Kling AI
Past test result. This model is no longer available.
Kling AI
Past test result. This model is no longer available.
Alibaba Wan
Past test result. This model is no longer available.
MiniMax
BotConnector Benchmark: Indonesian Rupiah & Billing Cadence
Indonesian Rupiah currency pronunciation (Rp), thousand separators, and decimal cadence test
Why we chose this prompt: Tests Indonesian locale token expansion, thousand point grouping, and technical voice naturalness in payment automation.
STANDARDIZED TEST PROMPT
Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik.
Showing 6 multimedia models for this prompt
Google DeepMind
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
Alibaba Cloud
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
Alibaba Cloud
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
Xiaomi
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
Kokoro
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
Chatterbox
Target Transcript:
"Halo, layanan BotConnector Cloud aktif dua puluh empat jam. Transaksi nomor 4.820 senilai Rp150.000 telah terverifikasi dalam 1,2 detik."
BotConnector Benchmark: Music Generation
Indonesian music from a text prompt
Why we chose this prompt: Lyria 3.5 generates music, so its output needs a music task instead of a speech pronunciation test.
STANDARDIZED TEST PROMPT
Buat lagu pendek berbahasa Indonesia tentang layanan BotConnector Cloud yang aktif sepanjang hari. Musik pop akustik bertempo sedang, vokal jelas, dan instrumen gitar serta piano. Sebutkan nama BotConnector Cloud dalam lirik.
Showing 1 multimedia models for this prompt
Google DeepMind
The same source image is enlarged by every AI upscaler through the BotConnector API (22 Oct 2026). Cost = the amount actually charged; time = end to end, including upload and download.
- Topaz Wonder 3.5 lands exactly on the target size (3840 and 7680 px) with the cleanest detail, but costs the most.
- Real-ESRGAN is the cheapest and crisp on text and screenshots, but only 2x.
- P-Image is rounded by the provider slightly below the target (e.g. 3830 px for 4K).
- Creative modes (P-Image Kreatif, Clarity) add new texture; Clarity also shifts the composition slightly, so its crop is not perfectly aligned.
Product photo 1024 px → 4K
AIs that can do 4x. Source: our benchmark image (Gemini 3 Pro Image).
Source 1024×1024
P-Image
Drag the slider to compare. Click a card to pick an AI. Card crops are 100% zoom at the output resolution; the first box is the source with plain (bicubic) enlargement.
No AI (bicubic)
Baseline
P-ImagePruna AI · 4x
- Output
- 3830×3830
- Cost
- $0.025 Rp449
- Time
- 13.9s
P-Image KreatifCreativePruna AI · 4x
- Output
- 3830×3830
- Cost
- $0.025 Rp449
- Time
- 13.1s
Topaz Wonder 3.5Topaz Labs · 4x
- Output
- 3840×3840
- Cost
- $0.161 Rp2.882
- Time
- 28.2s
ClarityCreativeClarity AI · 4x
- Output
- 3840×3840
- Cost
- $0.287 Rp5.162
- Time
- 93.3s
Product photo 1024 px → 2x (2048 px)
AIs that only do 2x, same source.
Source 1024×1024
Real-ESRGAN
Drag the slider to compare. Click a card to pick an AI. Card crops are 100% zoom at the output resolution; the first box is the source with plain (bicubic) enlargement.
No AI (bicubic)
Baseline
Real-ESRGANTencent ARC · 2x
- Output
- 2048×2048
- Cost
- $0.00075 Rp13
- Time
- 6.4s
SSwinIR
ETH Zürich (riset) · 2x
- Output
- 2048×2048
- Cost
- $0.00475 Rp85
- Time
- 9.6s
Phone text screenshot → 4K
botconnector.id Privacy Policy page, 884×1920 → 3840 px.
Source 884×1920
P-Image
Drag the slider to compare. Click a card to pick an AI. Card crops are 100% zoom at the output resolution; the first box is the source with plain (bicubic) enlargement.
No AI (bicubic)
Baseline
P-ImagePruna AI · 2x
- Output
- 1700×3694
- Cost
- $0.013 Rp224
- Time
- 11.1s
P-Image KreatifCreativePruna AI · 2x
- Output
- 1706×3682
- Cost
- $0.013 Rp224
- Time
- 11.0s
Real-ESRGANTencent ARC · 2x
- Output
- 1728×3840
- Cost
- $0.00075 Rp13
- Time
- 6.7s
SSwinIR
ETH Zürich (riset) · 2x
- Output
- 1728×3840
- Cost
- $0.00475 Rp85
- Time
- 15.0s
Topaz Wonder 3.5Topaz Labs · 2x
- Output
- 1768×3840
- Cost
- $0.107 Rp1.921
- Time
- 18.1s
ClarityCreativeClarity AI · 2x
- Output
- 1768×3840
- Cost
- $0.063 Rp1.122
- Time
- 25.4s
Phone text screenshot → 8K
Same source → 7680 px, only AIs that can do 4x.
Source 884×1920
P-Image
Drag the slider to compare. Click a card to pick an AI. Card crops are 100% zoom at the output resolution; the first box is the source with plain (bicubic) enlargement.
No AI (bicubic)
Baseline
P-ImagePruna AI · 4x
- Output
- 3542×7692
- Cost
- $0.05 Rp898
- Time
- 13.7s
P-Image KreatifCreativePruna AI · 4x
- Output
- 3554×7668
- Cost
- $0.05 Rp898
- Time
- 14.0s
Topaz Wonder 3.5Topaz Labs · 4x
- Output
- 3536×7680
- Cost
- $0.268 Rp4.803
- Time
- 26.1s
Video 720p → 4K & 8K
8-second clip from our video benchmark (Veo 3.1 Fast, 1280×720). Center crop at 100% zoom. Cost = estimated AI cost for this clip. Now available in Studio: open Your videos and choose Upscale 4K/8K.
- Output
- 4K 3840×2160
- Estimated cost (8 s)
- ≈$0.170 Rp3.047
- Time
- 113s
- Output
- 4K 3840×2160
- Estimated cost (8 s)
- ≈$0.257 Rp4.611
- Time
- 160s
- Output
- 8K 7680×4320
- Estimated cost (8 s)
- ≈$0.257 Rp4.611
- Time
- 320s
Compliance Evaluation Details
Evaluator Judges: