What is the cheapest SDXL API?

On 2026-10-02, Together AI listed SDXL at $0.0019 per megapixel (about $0.002 per 1024x1024 image) and Replicate listed stability-ai/sdxl at about $0.0025 per run. ModelsLab works out to $0.0047 per call if you use all 10,000 calls of its $47 Standard plan, and Stability AI listed SDXL 1.0 from $0.009 per image. fal.ai and Runware did not publish a per-image SDXL price.

Which API lets me use custom LoRAs with SDXL?

ModelsLab (lora_model parameter, several LoRAs per request, LoRA training at $1), Replicate (extra_lora takes a model name or a weights URL), fal.ai (fal-ai/lora endpoint), Runware (several LoRAs per request) and Civitai's Orchestration API (LoRAs by AIR id).

Where can I use the official Stable Diffusion 3.5 models?

Stability AI's platform API: SD 3.5 Large is 6.5 credits ($0.065), Large Turbo $0.04, Medium $0.035 and Flash $0.025 per image, at $0.01 per credit.

Which Stable Diffusion API allows NSFW models?

ModelsLab serves 4,000+ NSFW SDXL, SD 1.5, Pony and Illustrious checkpoints when you send safety_checker set to no. Content involving minors or non-consenting real people is prohibited. Check each other provider's content policy before you rely on it.

Is there a free Stable Diffusion API?

ModelsLab has no free plan; API access starts at $21/month. Some other providers give a small starter allowance; check their pricing pages.

Comparisons · 2026-10-02 · ModelsLab Team

Best Stable Diffusion, SDXL and LoRA API Providers 2026

Stable Diffusion, SDXL and custom LoRA APIs compared: Together, Replicate, Stability AI, Civitai, ModelsLab, Runware, fal.ai, Segmind and self-hosting, checked 2026-10.

Last updated 2026-10-02

Disclosure: stablediffusionapi.com is operated by the ModelsLab team. We list competitors where they are the better choice, and each price comes from the vendor's own page, linked in the tables.

Short answer (checked 2026-10-02): for base SDXL at the lowest metered price, use Together AI ($0.0019 per megapixel) or Replicate (about $0.0025 per run). For the official SD 3.5 models, use Stability AI (SD 3.5 Large $0.065). For community checkpoints, the choice depends on where the models live: Civitai’s Orchestration API for the Civitai catalogue by AIR id, ModelsLab for 10,000+ hosted checkpoints and your own LoRAs on one key with a flat $149/month option, and Runware for its 400K+ model registry. For full control, self-host on RunPod or Modal.

Comparison table

Provider	Best for	SD / SDXL price (checked 2026-10)	Models	Custom LoRA
Together AI	Cheapest base SDXL; LLMs on the same key	SDXL $0.0019/MP, about $0.002 per 1024x1024 image (source)	Open LLMs, Flux, SDXL	Not stated
Replicate	Cheap base SDXL; custom deploys with Cog	SDXL about $0.0025/run on L40S (source)	50,000+ (Cloudflare figure)	Yes: extra_lora by name or weights URL
Stability AI	Official SD 3.5 and Stable Image models	SD 3.5 Large $0.065; Medium $0.035; SDXL 1.0 from $0.009 (source)	Stability models	Not listed on the pricing page
Civitai Orchestration	Civitai models and LoRAs by AIR id	SDXL about 8 Buzz per 1024x1024 image at 20 steps (source)	Civitai catalogue	Yes, by AIR id; training 500 Buzz by default
ModelsLab	Many community checkpoints and LoRAs on one key; NSFW-allowed models	$0.0047/call on Standard ($47/month), $0.0065 on Basic; unlimited on $149/month (source)	10,000+	Yes: several per request, training $1 (source)
Runware	Very large model registry	SD / SDXL price not published (source)	400K+ in its registry	Yes, several per request
fal.ai	Fast SDXL with LoRA, Flux LoRA	fast-sdxl and fal-ai/lora billed per compute second, rate not published; Flux LoRA $0.035/MP (source)	1,000+	Yes; Flux LoRA training $2 per run
Segmind	Per-second SDXL with plans	SDXL $0.0043 per GPU second (source)	Image and video	Not stated for SDXL
Hugging Face Inference Endpoints	Your own SDXL pipeline on a dedicated GPU	L4 $0.80/hour; L40S $1.80/hour (source)	Hub models	Yes, in your own code
RunPod / Modal (self-host)	ComfyUI or A1111 graphs you control	RunPod RTX 4090 $0.34/hour (source); Modal L40S $0.000542/s (source)	Any	Yes, in your own code

Best for each use case

Lowest metered price for base SDXL: Together AI, then Replicate.
Official SD 3.5 with Stability’s license terms: Stability AI.
Models that already live on Civitai: Civitai Orchestration API.
Thousands of hosted community checkpoints and your own LoRAs behind one key, with a flat-price option: ModelsLab.
The largest model registry on one endpoint: Runware.
Flux LoRAs with fast training: fal.ai.
Custom ComfyUI workflows: self-host on RunPod or Modal.
NSFW checkpoints through an API: ModelsLab (uncensored models).

Provider notes

Together AI

Together is mainly an LLM inference platform, but its pricing page lists SDXL at $0.0019 per megapixel, about $0.002 for a 1024x1024 image, the lowest base SDXL price we found. Use it for base SDXL next to LLM calls; it is not a catalogue of community checkpoints.

Replicate

stability-ai/sdxl costs about $0.0025 per run on an L40S. LoRAs load through extra_lora with a model name or a weights URL, and SDXL LoRA training is supported. Community wrappers cost more (one SDXL LoRA model lists about $0.014 per run). Cloudflare closed its acquisition of Replicate on 2025-12-01.

Stability AI

The Stability platform prices in credits at $0.01 each: Stable Image Ultra $0.08, SD 3.5 Large $0.065, Large Turbo $0.04, Medium $0.035, Flash $0.025, Stable Image Core $0.03. SDXL 1.0 is still listed as a legacy model from 0.9 credits. Use it for the official models and Stability’s license, not for community checkpoints.

Civitai Orchestration API

Civitai’s Orchestration API runs SD 1.5 and SDXL recipes with checkpoints and LoRAs addressed by AIR id. An SDXL image is about 8 Buzz at 1024x1024 and 20 steps, and SDXL LoRA training defaults to 500 Buzz. The docs do not state a dollar price per Buzz.

ModelsLab

ModelsLab (which runs this site) serves 10,000+ models, mostly SD 1.5, SDXL, Pony and Illustrious checkpoints and LoRAs, from its own GPUs. You pass a model_id and one or more LoRAs in lora_model (docs), or train your own LoRA for $1. An SDXL call costs $0.0065 on Basic ($21/month, 3,250 calls) and $0.0047 on Standard ($47/month, 10,000 calls); Open Source Unlimited ($149/month) removes the per-call charge on these models, with 15 parallel generations. NSFW models are allowed with "safety_checker": "no". It has no free plan.

Runware

Runware’s registry has 400K+ models, and its SDXL docs show several LoRAs per request. Its pricing page has no SD 1.5 or SDXL row, so we cannot quote a per-image price.

fal.ai

fal-ai/lora runs any Stable Diffusion model with LoRA weights, and fast-sdxl serves base SDXL; both bill per compute second without a published rate. Flux LoRA is $0.035 per megapixel and Flux LoRA training $2 per run.

Segmind

Segmind bills SDXL at $0.0043 per GPU second, with plans from $39/month (pricing).

For your own pipeline, Hugging Face Inference Endpoints put a Hub model on a dedicated GPU (L4 $0.80/hour, L40S $1.80/hour). RunPod pods start at $0.34/hour for an RTX 4090 on Community Cloud, and Modal bills an L40S at $0.000542 per second. You manage the model, the LoRAs and the scaling.

How we checked

We read each price from the vendor’s own page on 2026-10-02. fal.ai, Runware and Novita did not publish an SD or SDXL per-image price, so we did not estimate one. getimg.ai is left out because its SD and SDXL models were scheduled for removal and its pricing page returned 404.

FAQ

What is the cheapest SDXL API?: On 2026-10-02, Together AI listed SDXL at $0.0019 per megapixel (about $0.002 per 1024x1024 image) and Replicate listed stability-ai/sdxl at about $0.0025 per run. ModelsLab works out to $0.0047 per call if you use all 10,000 calls of its $47 Standard plan, and Stability AI listed SDXL 1.0 from $0.009 per image. fal.ai and Runware did not publish a per-image SDXL price.
Which API lets me use custom LoRAs with SDXL?: ModelsLab (lora_model parameter, several LoRAs per request, LoRA training at $1), Replicate (extra_lora takes a model name or a weights URL), fal.ai (fal-ai/lora endpoint), Runware (several LoRAs per request) and Civitai's Orchestration API (LoRAs by AIR id).
Where can I use the official Stable Diffusion 3.5 models?: Stability AI's platform API: SD 3.5 Large is 6.5 credits ($0.065), Large Turbo $0.04, Medium $0.035 and Flash $0.025 per image, at $0.01 per credit.
Which Stable Diffusion API allows NSFW models?: ModelsLab serves 4,000+ NSFW SDXL, SD 1.5, Pony and Illustrious checkpoints when you send safety_checker set to no. Content involving minors or non-consenting real people is prohibited. Check each other provider's content policy before you rely on it.
Is there a free Stable Diffusion API?: ModelsLab has no free plan; API access starts at $21/month. Some other providers give a small starter allowance; check their pricing pages.