Stable Diffusion 3.5 Large guide to Bedrock access and API

Stable Diffusion 3.5 Large's documented Bedrock integration uses stability.sd3-5-large-v1:0 with the Bedrock Runtime API. AWS launched it on December 19, 2024, but its current supported-model catalog does not list SD3.5 Large; check account and region availability before using the request example below. The model-specific documentation still describes the JSON request and image response. Launch notice, current catalog, availability API, request reference

What are the key facts about SD3.5 Large on Bedrock?

What did AWS introduce with SD3.5 Large?

AWS introduced SD3.5 Large for image creation across styles and use cases such as concept art, campaign visuals, and product imagery. Its launch announcement described one-megapixel output. Launch announcement

The documented response format contains base64 image data and completion reasons. An integration should inspect both before presenting an image for review. Response contract

Start with a task your team already understands. A concept image of a reading chair beside a tall window provides clear requirements for the subject, setting, and lighting without requiring a complicated scene.

Evaluate the output against its intended placement. Check whether the chair is fully visible, whether the background leaves useful space, and whether the image still works in the final crop.

The Bedrock reference also documents image-to-image generation. Its request uses an input image and a strength setting, which creates a distinct workflow for testing variations of an existing composition. Model parameters

Keep text-to-image and image-to-image evaluations separate at first. That makes it easier to explain whether a result came from the prompt alone or from the composition supplied in a source image.

What access and response limits should you check?

The request reference caps the prompt at 10,000 characters and documents an explicit set of aspect ratios. Validate against that schema rather than forwarding arbitrary fields from another image provider. Input contract

Responses can contain filtering or inference-error reasons. A successful transport response is therefore not enough to assume that a usable image exists; inspect the documented response fields. Response contract

Write a separate application state for a filtered or failed request. Show a clear result to the operator and keep the error out of the normal image-saving path.

AWS model access also depends on account and permission prerequisites. Its guide explains the first-use subscription process for relevant third-party models and how missing prerequisites can produce access-denied errors. Model access

Resolve access problems before tuning prompts. A denied invocation is an account or permissions issue to investigate, not evidence that the visual description is unsuitable.

The current AWS catalog and regional availability table omit SD3.5 Large. Those pages do not provide a current region listing for a new deployment; use the account availability check before planning requests. Model catalog, regional table

How do you check access and call the documented Bedrock API?

1. Check stability.sd3-5-large-v1:0 with GetFoundationModelAvailability in the intended region. Inspect its authorization, entitlement, agreement, and region fields; continue only after establishing access. AWS documents these checks separately from the historical launch announcement. Availability API, access prerequisites

2. Configure AWS credentials for the runtime environment and permission to invoke the model. Use the Bedrock Runtime client and the exact model identifier from the model-specific documentation. Invocation example

3. Send a small text-to-image request and inspect its response before adding uploads, batch processing, or downstream editing.

After confirming model access, the following Python example follows AWS's documented boto3 request and Oregon region. It requests PNG output and writes the decoded image only when the response contains an image without a failure reason. Request and response fields

import base64
import json
from pathlib import Path
import boto3

client = boto3.client("bedrock-runtime", region_name="us-west-2")
response = client.invoke_model(
    modelId="stability.sd3-5-large-v1:0",
    contentType="application/json", accept="application/json",
    body=json.dumps({
        "prompt": "A reading chair beside a tall window, soft daylight",
        "mode": "text-to-image", "output_format": "png",
    }),
)
result = json.loads(response["body"].read())
if any(result.get("finish_reasons", [])) or not result.get("images"):
    raise RuntimeError(result.get("finish_reasons", "No image returned"))
Path("reading-chair.png").write_bytes(base64.b64decode(result["images"][0]))

Save the prompt, selected region, model identifier, and returned seed alongside the result. Those fields make an image easier to trace through your review process without implying that all future runs must be identical.

For an image-to-image request, set mode accordingly and supply the documented image and strength fields. AWS expects the input image as base64 data.

Follow that schema instead of reusing a multipart request from Stability's own API. Image-to-image reference

After the basic request works, add input validation and explicit handling for each error category your application encounters.

Keep sample successful and failed response shapes in development notes so the integration is understandable to another maintainer.

The current Bedrock pricing page lists other Stability image services but does not supply an SD3.5 Large rate. Obtain the applicable terms for an accessible deployment before estimating costs; another service's image-generation price does not establish this model's price. AWS pricing

Budget from the number of generated candidates that your review process needs. Track image generation separately from storage and other application services so the estimate has visible assumptions.

For batch work, start with a limited run and inspect the results before increasing volume. Record acceptance rate and observed request time, then adjust the workload around the account's actual limits.

How does the Bedrock route compare with self-hosting?

The ComfyUI pillar explains the graph-based self-hosting context. The cloud GPU pricing guide can help organize the infrastructure side of that comparison.

For Stability's own HTTP service, read the SD3 API sibling guide. Its authentication, request encoding, and model values differ from the AWS request shown here.

What else should you know about SD3.5 Large on Bedrock?

Can a new AWS account use SD3.5 Large on Bedrock?

AWS's current supported-model catalog does not list SD3.5 Large, although the historical launch and request documentation remain available. Check stability.sd3-5-large-v1:0 through GetFoundationModelAvailability to establish the account and region's access before attempting generation. Catalog, availability API

Does SD3.5 Large Bedrock access include downloadable weights?

The SD3.5 Large Bedrock integration invokes hosted inference through AWS. Stability publishes a separate weight repository with Community License conditions for self-hosted deployment. AWS invocation, weights

Why does an SD3.5 Large Bedrock response contain no image?

SD3.5 Large's Bedrock schema reports filtering and inference errors through finish_reasons. Inspect that field and the images list before decoding or saving an output. Response fields

Can I reuse a Stability API request for Bedrock?

The SD3.5 Large Bedrock integration uses the model ID stability.sd3-5-large-v1:0 and a JSON body sent through Bedrock Runtime. Follow the AWS schema, including base64 encoding for an image-to-image input. Request reference

Sources

• AWS SD3.5 Large launch announcement
• AWS Bedrock release notice
• Bedrock SD3.5 Large request and response reference
• Amazon Bedrock model access
• Current AWS supported-model catalog
• Current AWS regional model availability
• AWS model availability API
• Amazon Bedrock pricing
• Stability AI SD3.5 Large model card

Related guides on PromptZone

• Best SDXL Models in 2026
• ComfyUI 2026: The Complete Guide
• AI Model Releases Timeline