ImageMusic

Model developer guide

Stability AI

7 website model guides

Stability's image and audio models solve different creative problems. Begin with a visual composition or a sound brief, then read the exact model's contract. The current app evidence is narrower than the provider's editing tools, and both Stable Audio routes have important timing or identity questions that should remain visible.

Image generationAudio generationOpen ecosystem

Practical guidance

Where should I start?

Editorial starting points, not a performance ranking. Check current app access before choosing a model.

Develop a detailed visual brief

Stability positions SD 3.5 Large for image quality and prompt adherence at approximately one-megapixel resolution. Evaluate it with a clear subject, composition and material palette rather than relying on a long list of unrelated style labels.

Tradeoff: Provider positioning is not a measured advantage inside the app. The audited request is prompt-only, so do not plan an image-edit workflow around this selection.

SD 3.5 Large guide

Compare another image workflow

Core has its own generation endpoint in Stability's API, separate from Ultra and the SD3.5 family. Use a consistent brief to compare composition and usable details across the image choices listed below.

Tradeoff: A different endpoint does not imply the same seed, parameter behavior or customer charge. Judge results against the task rather than assuming identical output from similar names.

Stable Image Core guide

Plan an understated music bed

Replicate documents text-to-audio for music and sounds. It is a useful reference for a simple instrumental brief, with an explicit warning that the app's recorded timing field differs from the current gateway schema.

Tradeoff: Do not promise a timed deliverable before checking the returned file. The website explains the mismatch; it does not repair or verify the application's request.

Stable Audio 2.5 guide

Technical details

What does the provider document?

Official documentation describes the named model or gateway—not every feature available in this app.

Image endpoints have different model selectors

Stability's API separates Stable Image Ultra, Core and Stable Diffusion 3.5 generation. SD3.5 variants use the shared endpoint's model selection. The provider exposes options such as aspect ratio and seed, but those fields are not automatically available in an integration that sends only its defaults.

Stable Audio 3 is a family, not one duration

Stability's official repository lists Small-Music and Small-SFX up to 120 seconds and Medium/Large up to 380 seconds. It describes text-conditioned audio, editing and continuation workflows. Those family capabilities do not identify which variant is behind the audit's stability-ai/stable-audio-3/music gateway label.

Replicate's time field is duration

The current Stable Audio 2.5 schema on Replicate uses duration, accepts 1–190 seconds and defaults to 190. It also lists seed, steps and guidance controls. The owner audit instead reports seconds_total = 60. Those names must not be treated as equivalent without verifying the final handler and gateway response.

Complete website coverage

Explore all 7 model guides

Every existing catalog entry for this provider is linked below. Status labels come from the individual website guides; catalog coverage is not a guarantee of current app access. Historical and snapshot pages remain references, not extra active models.

Practical guidance

From a brief to a useful result

Original prompt examples, not generated-output demonstrations. They have not been generation-tested.

Workflow 1

Create an uncluttered materials study

Use a square composition matching the audited app default. Describe three objects and how light should reveal their surfaces; do not request an unavailable aspect-ratio or reference-image control.

Create an original square studio still life of a frosted glass tumbler, a folded burgundy linen cloth and one smooth pale stone. Place the objects on matte charcoal with clear spacing. Light from the upper left to distinguish glass, fabric and stone textures. Keep the scene physically plausible, with no writing, logos, extra objects or glossy plastic appearance.

Check the result: Count objects, inspect shadows and look for believable material differences. Confirm the result format and dimensions instead of assuming the prompt changed output settings.

Read the SD 3.5 Large guide

Workflow 2

Brief a narration-friendly instrumental

Use the reported prompt-only music interface. Keep the brief below 500 characters and plan to inspect timing because the audit's seconds_total field does not match Replicate's duration schema.

Create an original quiet instrumental for a repair tutorial. Use a soft electric piano pattern, muted bass and light brushed percussion. Keep the energy even and leave space for spoken instructions. Avoid singing, spoken words, large drum fills and dramatic transitions. Finish with a gentle resolved phrase rather than a sudden stop.

Check the result: Check for unwanted vocals, masking of speech and abrupt endings. Measure the actual file length before editing; a requested minute has not been validated for this route.

Read the Stable Audio 2.5 guide

Technical details

What changes inside EZ Ai Assist?

Official Stability and Replicate documentation was checked. The October 8 image audit and October 9 music audit report app behavior, not independently tested generation. Stable Audio 3's exact WaveSpeed variant and the Audio 2.5 duration-field conflict remain unresolved.

Owner image audit · October 8

Image settings in the audited app

The reported flow sends a prompt with default square aspect ratio and PNG output. It does not expose seed, negative-prompt or aspect-ratio controls and does not forward images for editing. SD3.5 uses the shared generation endpoint; Ultra and Core retain separate paths.

Owner music audit · October 9

Audio 3's route is not fully established

The app choice is stability-ai/stable-audio-3/music via WaveSpeed. The audit reports prompt only: no duration, input audio, seed or format override. The exact gateway variant and default output length remain unverified; the existing guide records an unavailable exact route page, not a substituted endpoint.

Audit versus current gateway schema

Audio 2.5's minute is not a selector

The music client requests durationSeconds: 60 with no user-selectable duration. The Audio 2.5 wrapper reportedly sends seconds_total = 60, while Replicate documents duration. Audio 3 does not forward duration at all. Neither route is certified here to return exactly 60 seconds.

Before you rely on a result

Limits, costs and tradeoffs

  • Provider audio editing, continuation and fine-tuning features do not establish an app upload flow. Plan around documented app controls and mark everything else as provider-only capability.
  • The missing Audio 3 gateway identity and Audio 2.5 field conflict are material uncertainties. A family maximum must not become an app promise, and the website does not silently fix a backend route.
  • Credits, per-generation API tariffs, picker estimates and customer charges are different quantities. No app price, quality ranking or turnaround time is inferred from the provider's reference material.

Provider references and picker estimates are not EZ Ai Assist customer charges. Read the website pricing explanation.

Compare by task

Other paths to explore

FLUX

Compare image-focused alternatives on the same composition, with route-specific editing and control checks.

Sonilo

Explore another music gateway contract and its reported app timing behavior.

Google Lyria

Compare music versions with explicit vocal and duration distinctions rather than a single family-wide assumption.

Common questions

Stability AI, without the assumptions

Can I edit an uploaded image with these app choices?

The existing image audit reports prompt-only requests and no forwarded image-edit input for the Stability routes. The provider's wider editing API is separate and must not be advertised as an app feature on that evidence.

Does Stable Audio 3 always support 380 seconds?

No. Stability lists different limits for Small versus Medium/Large, and the exact WaveSpeed variant behind the audited music route is unresolved. The app has no duration selector and sends only a prompt on that route.

Why might Stable Audio 2.5 ignore the minute request?

The audit names seconds_total, but Replicate's current schema names duration. The resulting behavior has not been tested here, so request acceptance and returned length remain unverified rather than presumed correct.

New to the terminology?

Official documentation

Sources & review notes

Highlighted provider contracts reviewed . Individual model guides carry their own review dates. Sources can change; these pages do not perform live health checks.

Official Stability and Replicate documentation was checked. The October 8 image audit and October 9 music audit report app behavior, not independently tested generation. Stable Audio 3's exact WaveSpeed variant and the Audio 2.5 duration-field conflict remain unresolved.