potocolom

Model runtime on 16 GB VRAM

Curated prompts, fixed parameter variants, one model loaded at a time. GPU milliseconds are measured on the worker; wall time includes queue and I/O.

No benchmark data yet

Run make benchmark on a GPU worker, then make benchmark-publish to copy results.json into the site.

Model specifications

Hardware and license requirements from worker manifests. VRAM is for full GPU residency; lower-VRAM cards may use CPU offload at reduced speed.

ModelArchitectureParametersMin VRAMResolutionsStepsCapabilitiesLicenseCommercialIn studio

SDXL (default)

sdxl-base

SDXL~3.5B10 GB768, 102410-50text to image, image to imageCreativeML Open RAIL++-MUnrestricted (RAIL use policy)Yes

SDXL Fast

sdxl-fast

SDXL + Lightning LoRA~3.5B10 GB10244-8text to imageRAIL++-M + ByteDance LightningUnrestricted (RAIL use policy)Yes

SSD-1B

ssd-1b

SDXL distilled~1.3B8 GB768, 102410-40text to imageApache 2.0UnrestrictedYes

DreamShaper 8 LCM

dreamshaper-lcm

SD 1.5 LCM~860M6 GB512, 7682-15text to imageCreativeML Open RAIL-MUnrestricted (RAIL use policy)Reference only

SD Turbo

sd-turbo

SD 1.5 distilled~860M8 GB5121-4text to image, image to image, realtimeStability AI Community≤ $1M revenue / yearYes

SDXL Turbo

sdxl-turbo

SDXL distilled~3.5B10 GB5121-4text to image, image to image, realtimeStability AI Community≤ $1M revenue / yearYes

SDXL Hyper-SD

sdxl-hypersd

SDXL + Hyper-SD LoRA~3.5B10 GB10244-8text to imageRAIL++-M base; LoRA: no declared licenseUnclear - LoRA license undeclaredYes

Segmind VegaRT

vega-rt

Segmind-Vega + VegaRT LCM LoRA~1.0B8 GB512, 10242-8text to image, image to image, realtimeApache 2.0UnrestrictedYes

SSD-1B + Lightning

ssd-1b-lightning

SSD-1B + Lightning LoRA~1.3B8 GB10244-8text to imageApache 2.0 base + Open RAIL++-M LoRARAIL use policy appliesYes

Stable Diffusion 3.5 Medium

sd35-medium

MMDiT + CLIP-L + OpenCLIP-G + T5-XXL~2.5B14 GB768, 102420-50text to imageStability AI Community<= $1M revenue / yearYes