Ethical Stack.

Stable Diffusion 3.5

Downloadable image-generation models that can run locally or through hosted services, under Stability’s version-specific license.

Score calculation

(4 + 3 + 8 + 7 + 1 + 8) ÷ 6 = 5.2

6 of six dimensions rated. Unrated dimensions are excluded from the average.

Full scoring method ↗
Setup details

Stable Diffusion 3.5: both access. Self-hosting is free under Community License below $1M annual revenue; over that threshold requires Enterprise terms. Ratings describe the stated product, company or configuration in each finding.

What fits your needs

Downloadable Stable Diffusion 3.5 model family; local deployment and third-party hosted routes have different hardware, privacy and costs.

Tasks

Image generation and fine-tuning can be run locally or through a hosted provider; several model variants and inference implementations are available.

See the evidence

Billing and license

Self-hosting is available under Stability's Community License below the stated revenue threshold; enterprise use has separate terms. Hosting compute is an additional cost.

See the evidence
Tradeoffs to expect

Local setup

Requires downloading weights, compatible GPU/VRAM, runtime and storage; hardware needs depend on the specific model variant and quantization, so no universal minimum is stated here.

See the evidence

Hosted routes

Using a third-party UI or API sends prompts and images to that provider; those routes do not inherit local-device data handling.

See the evidence
How to switch

Get started

Choose a specific SD 3.5 variant, review its license, then download from the official model listing and use a compatible local runtime or a separately assessed hosted provider.

See the evidence
Practical questions

Age and sensitive content

Stability’s AUP applies to self-hosted code/weights and hosted access: users must be 18 or the higher local age. It prohibits CSAM, NCII, illegal pornography, and impersonation without consent.

See the evidence

Consent and disclosure

The AUP prohibits privacy-rights violations, non-consensual intimate imagery, impersonation without consent, and misleading others about AI-generated output. Local operation does not add content moderation by itself.

See the evidence

Labels and reporting

C2PA standards are described for Stability hosted services; this doesn’t confirm automatic provenance labels for locally generated SD 3.5 files. Stability publishes an abuse report route.

See the evidence

Commercial use

Community License permits commercial use below the stated revenue threshold, subject to its terms and acceptable-use restrictions.

See the evidence

Energy practices

4/10↑

Stability AI documents open deployment routes, model sizes, a distilled four-step variant, and local hardware requirements. No model-specific electricity, emissions, or water measurements were found.

See the evidence

Stability AI documents open deployment routes, model sizes, a distilled four-step variant, and local hardware requirements. No model-specific electricity, emissions, or water measurements were found. The score describes documented measurement transparency and available controls, not environmental performance.

1 + 1 reporting + 0 electricity + 2 controls = 4/10. This grade covers documented practices and transparency; it does not measure the energy or emissions of a request.

Local users choose hardware and electricity, while hosted endpoints have separate operators; neither route has a footprint in the reviewed product sources. Model size, step count, and VRAM needs alone don’t show energy per image.

Reporting1/3
Stability AI’s model card and self-hosting docs identify the model and deployment boundary, with no quantified environmental reporting for the product.
Electricity sourcing0/3
The model supports user-chosen local, cloud VM, and third-party hosted deployments. No electricity source is established for any particular deployment, so no provider/company sourcing is transferred.
Resource controls2/3
The SD 3.5 family includes model-size/variant selection, output-size ranges, and a distilled Large Turbo variant documented to generate in four steps. Self-hosting docs also expose device choice. These controls may bound compute, but there are no measured physical-resource reductions.

Water disclosure

Water reporting has not been assessed for this choice.

Training data

3/10↑

The model card says SD 3.5 used synthetic and filtered publicly available data. Stability separately describes public web data, third-party access and synthetic data, but publishes no release-specific rights ledger.

See the evidence

The available disclosure provides broad provenance and general sourcing claims while leaving source datasets, licenses, and creator remedies unresolved. The low score measures the record that can be verified for SD 3.5, not a finding that its training was unlawful.

1 + 1 source transparency + 1 permission + 0 creator controls = 3/10. This grades documented training-data safeguards, not a legal ruling.

No named dataset inventory, license compatibility review, or model-specific creator exclusion/remedy is documented.

Source transparency1/3
The card identifies synthetic and filtered public data, while company reporting adds broad categories but no named inventory.
Documented permission1/4
General sourcing statements exist, but the model card does not connect material sources to licenses or permissions.
Creator controls0/2
No training-data rights-holder exclusion or remedy was found in the reviewed sources.
Data-rights rubric ↗

Your data & privacy

8/10↑

When weights are downloaded and run fully offline, prompts and images need not be sent to Stability; hosted deployments have separate privacy boundaries.

See the evidence

This score applies only to a locally operated, offline configuration. Hardware logs, plugins, custom interfaces, and any cloud-hosted setup can change the data flow.

Openness

7/10↑

SD 3.5 weights are downloadable and can be run locally, under Stability’s Community License with revenue and use restrictions.

See the evidence

Downloadable weights and self-hosting enable inspection and deployment control; the Community License is not an unrestricted OSI-style model license and includes conditions, so openness is not rated as fully permissive.

Worker protections

1/10↑

Stability AI’s model card and license explain model access, deployment, and commercial licensing, but do not report worker pay, exposure protections, grievance routes, supplier coverage, or independent labor outcomes for training-data and moderation work.

See the evidence

Stability AI’s model card and license explain model access, deployment, and commercial licensing, but do not report worker pay, exposure protections, grievance routes, supplier coverage, or independent labor outcomes for training-data and moderation work. Downloadable weights make downstream deployments distinct from the hosted service workforce.

1 + 0 pay & conditions + 0 worker voice & remedy + 0 evidence & follow-through = 1/10. This grades documented worker safeguards, not a measured rate of mistreatment.

Worker pay, exposure and health protections, paid versus volunteer roles, vendor/subcontractor coverage, grievance routes, and independent outcome checks are not established by the reviewed company license and a stability model card checked; neither is a workforce report, and open-weight downstream users are outside the hosted labor scope.

Pay & conditions0/3
No opened source reports worker pay, hours, exposure protections, contractor coverage, or worker-level outcomes for Stability AI. The reviewed material is limited to Company license and a Stability model card checked; neither is a workforce report, and open-weight downstream users are outside the hosted labor scope.; its silence is an information gap, not evidence of poor conditions.
Worker voice & remedy0/3
The reviewed material for Stability AI doesn’t confirm worker representation, grievance access, bargaining voice, or documented remedies. Public code contributions, customer controls, or creator rights do not demonstrate worker voice across employees and suppliers.
Evidence & follow-through0/3
No reviewed source provides an independent audit or corroborated worker-level outcomes for Stability AI’s relevant employees, contractors, and data-work suppliers. The available disclosures are company/project materials rather than independent labor verification.
Workers rubric ↗

Military ties

8/10Not established↑

Stability AI’s AUP bans illegal or regulated weapons development and multiple harmful surveillance practices, including for community-licensed self-hosted weights, but it does not ban military or intelligence use generally.

See the evidence

The July 2025 AUP applies to Stability products, APIs, models and materials, explicitly including self-hosting under Non-Commercial Research, Community or Enterprise licenses and third-party hosting. It bans development/manufacture of illegal or regulated weapons, real-time public biometric identification for law enforcement, facial-recognition database expansion without consent and specified harmful profiling. It does not state a blanket military/intelligence ban; this does not necessarily govern independently redistributed permissive releases.

7 base + 1 safeguard adjustment = 8/10 (4 of 9 safeguard points; every 3 adds 1). No military relationship was established in the cited sources. This ranks above known defense ties without claiming a verified clean record.

Unpublished contracts, reseller/subcontractor activity, customer-configured use and enforcement outcomes remain unknown. Hosted AUPs do not necessarily apply to independent local deployments or permissively licensed copies.

Use restrictions2/4
Specific weapons and harmful-surveillance safeguards exceed lawful-use-only rules, but no general military/intelligence prohibition exists; published license scope applies to specified Stability-licensed use, not every downstream copy.
Enforcement & oversight1/3
Stability may suspend or terminate access for AUP violations; no independent military-specific review or enforcement outcomes are disclosed.
Relationship disclosure1/2
A public AUP names some prohibited uses, but no defense customer or relationship is disclosed in reviewed material.

The bounded official review found no named military/intelligence contract or operational use for this company/project. This evidence limit is not a clean bill of health; third-party model-provider relationships and customers’ independent deployments are separate.

Military rubric ↗

BDS targets

Not listed↑

Categorical check of the BNC campaign guide and No Tech campaign page; absence from these named-target pages is not BDS approval.

See the evidence

No named target for this company was found in the two checked BNC sources as of 2026-10-24. This is a narrow list check, not BDS approval, and it does not capture unlisted relationships or partners.

More from Stability AI

All sources

Side by side