Runflow
Stability AI

SDXL Inpainting API

SDXL's mask-based inpainting variant — selectively edit regions of an image while preserving the rest. Full LoRA support with adjustable strength and guidance.

sdxl-inpainting3.5B paramsUNet-based latent diffusion with dual CLIP text encoders (inpainting variant)
Updated March 2026·Early access
Price
~$0.003/image
Usage-based, per compute second
Latency
~2–4s
25 steps at 1024×1024
Cold Start
5–10s
Not billed · handled by Runflow
Category
Image Editing
inpainting

SDXL family by Stability AI

Explore all variants in the SDXL family available on Runflow.

SDXL

Standard

Standard 3.5B text-to-image generation

SDXL Image-to-Image

Image-to-Image

Image-guided generation with strength control

SDXL Inpainting

Inpainting

Mask-based region editing

SDXL Lightning

Fast

1–4 step generation, sub-second latency

SDXL ControlNet Union

ControlNet

10+ ControlNet types in one model

SDXL Fooocus

Enhanced

Enhanced quality with automated prompt optimization

Benchmarked for your use case

Every model on Runflow is scored per vertical — not generic leaderboard numbers. Scores come from domain-expert evaluation on real production prompts.

Use CaseAssessmentNotes
PortraitPendingPortrait inpainting quality
FashionPendingFashion inpainting quality
ProductPendingProduct inpainting quality
CreativePendingCreative inpainting quality

API Parameters

Full parameter reference for the sdxl-inpainting model endpoint.

ParameterTypeDescription
promptstringText description of what to generate in the masked region.
image_urlstringURL of the base image to inpaint. Accepts JPEG, PNG, WebP, GIF, AVIF.
mask_urlstringURL of the mask image. White regions are edited, black regions are preserved.
strengthfloatModification intensity in masked region. 0.01 = subtle; 1.0 = full generation. Default: 0.95.
negative_promptstringDescribe what to exclude from the inpainted region.
image_sizeobject | stringOutput dimensions. Presets: square_hd, square, landscape_4_3, landscape_16_9, portrait_4_3, portrait_16_9.
num_imagesintegerNumber of inpainted variations. Range: 1–8. Default: 1.
num_inference_stepsintegerDenoising steps. 20–25 for standard, 30–50 for max fidelity. Default: 25.
guidance_scalenumberClassifier Free Guidance. 7–9 recommended. Default: 7.5.
seedintegerReproducibility seed. Same seed + same parameters = identical output.
lorasarray[object]Array of LoRA weights to apply during inpainting.
embeddingsarray[object]Textual inversion embeddings to apply during generation.
expand_promptbooleanAuto-expand prompt with additional detail via LLM. Default: false.
formatstringOutput image format. Options: "jpeg", "png". Default: "jpeg".
enable_safety_checkerbooleanEnable content safety filter. Default: true.
sync_modebooleanIf true, returns the image as a data URI.

Technical Specifications

ArchitectureUNet-based latent diffusion (3.5B, inpainting variant)
Parameters3.5B (base); ~6.6B with refiner
Input FormatsJPEG, PNG, WebP, GIF, AVIF
Max Resolution1024 × 1024 native
Mask FormatBinary mask image (white = edit, black = preserve)
Strength Range0.01–1.0 (controls generation intensity in masked region)
VRAM7–8 GB (FP16)
Output FormatJPEG, PNG
LoRA SupportYes — 50,000+ community LoRAs on Civitai
LicenseCreativeML Open RAIL++-M (no revenue restrictions)

Why run SDXL Inpainting on the Runflow API

Flux AI models are available on many platforms. Here's what makes the Runflow image generation API different.

Precise region control

Binary mask defines exactly which pixels to regenerate — edit specific objects, backgrounds, or details while keeping everything else untouched.

Seamless blending

The inpainting model matches lighting, perspective, and style of surrounding unmasked areas for natural-looking edits with no visible seams.

Full LoRA support

Apply any of 50,000+ community LoRAs during inpainting for style-consistent region replacement — match any visual aesthetic.

Effective negative prompts

SDXL's highly responsive negative prompt system lets you precisely control what to avoid in the inpainted region.

Batch variations

Generate up to 8 inpainted variations per request with different seeds — quickly explore options for the edited region.

Explore other models

Same API, different strengths. Switch models with one parameter change.

Frequently asked questions

Resources

Start generating images with SDXL Inpainting

Get API access in minutes. No GPU setup, no infrastructure to manage. Pair with Sentinel to control the quality of every AI-generated image you ship.