Image GenerationActive

Stable Diffusion 3.5 Large

Stability AI's 8.1B-parameter MMDiT text-to-image model from the Stable Diffusion 3.5 release, built for 1 megapixel images in styles that range from photography and 3D to painting and line art.

Text to ImageFp32

Model ID

stable-diffusion-3.5-large

Provider

stabilityai

Updated

Nov 7, 2026, 04:25

Support Wiro on G2

wiro playground—stabilityai/stable-diffusion-3.5-large

Prompt*

A strikingly symbolic surreal composition portraying a single tree split into two contrasting halves, forming the profile of a human face, where one side is barren and lifeless while the other thrives with lush greenery. The left half of the image presents a bleak dystopian landscape, filled with towering smokestacks belching thick, dark clouds into the sky, a sea of overflowing garbage bags piled beneath, and a cracked, ashen road stretching endlessly. The skeletal branches of the tree mirror the decay, devoid of leaves, twisted and lifeless, blending into the smog-filled atmosphere. On the right side, a vibrant utopian paradise emerges, with rolling green fields stretching toward lush forested mountains, illuminated by a soft, golden glow. The tree here is full of life, its rich green foliage thriving under a bright blue sky, where a radiant rainbow arcs gracefully, casting a hopeful aura over the pristine natural landscape. The stark contrast between industrial destruction and environmental harmony conveys a profound visual metaphor of human impact, nature’s resilience, and the choice between devastation and renewal in a hyper-detailed, thought-provoking surrealist art style.

Tell us about any details you want to generate

Negative Prompt

Specify things to not see in the output

Your request will cost $0.0006 per second.

Running this model on Wiro costs approximately $0.074803 in total.

(Total cost varies depending on the request's execution time.)

Sample outputs

Updated Nov 7, 2026, 04:25

Overview

Stable Diffusion 3.5 Large is an 8.1-billion-parameter MMDiT text-to-image model from Stability AI's 3.5 release. Its transformer blocks add Query-Key Normalization, which Stability AI introduced to stabilize training and make fine-tuning easier. Two CLIP encoders (OpenCLIP-ViT/G and CLIP-ViT/L) read up to 77 tokens each, and a T5-xxl encoder was trained with 77 and 256 tokens at different stages.

Stability AI aims the model at professional work at 1 megapixel. Its release post lists 3D, photography, painting and line art among the styles it covers, and says the model shows people with varied skin tones and features without extra prompting.

Choosing a variant: Stable Diffusion 3 Medium is the earlier 2B model of the SD3 series. Choose 3.5 Large when a single image has to follow a detailed prompt closely, since Stability AI reports it leads on prompt adherence. The default here is that kind of prompt: a tree shaped like a human profile, polluted on one half and green on the other. Go to Medium for batches of up to 20 images per run; this page has no output-count setting.

What you can build

Technical-drawing posters like the mechanical cat schematic in the samples
Aerial views of invented cities, such as the cliffside city sample
Fantasy scenes like the sailing ship under a nebula or the frozen Statue of Liberty
Editorial and concept art from paragraph-length briefs like the default prompt

Inputs

You write a prompt, which is required, and can describe unwanted traits in an optional negative prompt (prefilled with "bad quality, lowres"). Six advanced settings sit behind it. Steps default to 30 (max 500) and guidance to 7.0 (1 to 20). The seed begins at 123456, capped at 9,999,999. Width and height open at 1024 and stop at 2048 pixels. FlowMatchEulerDiscreteScheduler is the default, with seven other samplers such as DDIM and DPM-Solver Multistep available.

Outputs

Images come back at the width and height you set. All four samples here are square 1024 by 1024 renders.

Recommended settings

Size: stay near 1 megapixel, the resolution Stability AI built the model around. 1024 by 1024 fits.
Guidance: Stability AI's example pairs 28 steps with guidance 3.5, and its 4-bit example uses 4.5. Both are well below the 7.0 default here, so try that range first.
Prompt order: the CLIP encoders stop at 77 tokens, so put the subject and style early in a long prompt.
Seeds: try several. Stability AI says results for one prompt vary more across seeds by design, to keep a wide range of styles.

Limitations

Prompts that leave details open give less predictable images, and Stability AI says aesthetic quality may vary with them.
Stability AI's card rules out the model for accurate portrayals of real people or events.

Safety & compliance

The Stability AI Community License covers SD3.5 Large. Individuals and organizations under its annual revenue limit may use the model commercially as well as for research; anyone above that limit needs an Enterprise License. Stability AI states that users own the images they generate, and all use falls under its Acceptable Use Policy. The card adds that training-data filtering does not guarantee every harmful output is gone and asks deployers to build their own guardrails.

Example prompts

Great starting points for stable-diffusion-3.5-large.

Poster of a mechanical cat, techical Schematics viewed from front.Image Generation

Bright scene, aerial view,ancient city, fantasy, gorgeous light, mirror reflection, high detail, wide angle lens.Image Generation

Pirate ship trapped in a cosmic maelstrom nebulaImage Generation

a dramatic and apocalyptic style, featuring a frozen landscape. The central subject is the partially submerged and ice-covered Statue of Liberty, with only the torch-bearing arm and head visible, symbolizing a catastrophic event. The sky is overcast with dark, swirling clouds, enhancing the ominous atmosphere. The overall composition conveys a sense of urgency and disasterImage Generation

API quick start

Run stable-diffusion-3.5-large with a single API call.

POST https://api.wiro.ai/v1/Run/stabilityai/stable-diffusion-3-5-large

{
  "prompt": "A strikingly symbolic surreal composition…",
  "negativePrompt": "bad quality, lowres",
  "steps": 30,
  "scale": 7.0
}

curl

curl -X POST "https://api.wiro.ai/v1/Run/stabilityai/stable-diffusion-3-5-large" \
  -H "Content-Type: application/json" \
  -H "x-api-key: YOUR_WIRO_API_KEY" \
  --data-binary @- <<'JSON'
{
  "prompt": "A strikingly symbolic surreal composition…",
  "negativePrompt": "bad quality, lowres",
  "steps": 30,
  "scale": 7.0
}
JSON