AI Image, Video, Voice and Music Models Hub: Status, Pricing and Provenance in 2026

Generative media models now launch, change price and retire within months, and the facts that matter to a buyer (status, price, licence and provenance marks) are scattered across dozens of provider pages. This hub gathers the main AI image, video, voice and music models in one place, using the same fields for each: owner, modality, status, access route, entry price, commercial terms and provenance. It is written for creative teams, marketing and legal reviewers, security and privacy teams, and anyone choosing or governing a media generation tool. Every fact comes from the provider's own page on the check date; where a provider publishes nothing, the table says so. Providers and models are listed alphabetically; order is never a ranking.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider.

Key takeaways

• Retirements are now routine. OpenAI shut down the DALL-E 2 and DALL-E 3 API on 12 May 2026 and the Sora Videos API on 24 September 2026, and lists gpt-image-1 as shut down on 23 October 2026.
• Video models are converging on sound plus picture. Seedance 2.0 generates stereo audio with video, Veo 3.1 supports sound generation, and Kling 4.0 is marketed around audio-visual performance.
• Two provenance systems run side by side. C2PA Content Credentials attach signed metadata; SynthID embeds an invisible watermark. Google applies SynthID to Omni videos and Lyria tracks, and Veo 3.1 on Google Cloud supports Content Credentials.
• Commercial rights usually follow the paid plan. Suno grants commercial rights only on paid plans, and ElevenLabs lists a commercial licence from its Starter plan.
• Open weights remain an image story. Black Forest Labs and Stability AI publish downloadable image model weights under their own licences; the video and music models on this page are reached through apps or APIs.

What counts as an AI media model?

An AI media model is a generative model whose main output is an image, a video clip, synthetic speech or music rather than text. Most are reached through three routes: a consumer app (for example the Gemini app or Suno), a creative platform that resells several models under one credit balance (Runway lists models from ByteDance, Google and Kling beside its own), and a developer API billed per image, per second or per token. A smaller group publishes weights that can be run on your own hardware.

As of 2026-11-15. Deeper single-model pages: AI Image Generators Compared and AI Video Generators Compared.

Which AI media models are active in 2026?

As of 2026-11-15. Black Forest Labs also lists FLUX 3 Max, FLUX 3 Video and FLUX 3 Image on its release post; details for those were not published on the page checked. Midjourney is active but its version history could not be confirmed from its official documentation on the check date, so it is covered separately in Midjourney Explained.

Which AI media models have been retired?

OpenAI publishes the most detailed public retirement schedule of the providers on this page. Fine-tuned versions follow the same shutdown dates as their base models.

As of 2026-11-15, from the OpenAI deprecations page. On Google Cloud, Veo 3.1 Generate and Veo 3.1 Fast Generate show a retirement date of "May 4, 2027 or later", and Google notes the models may remain reachable through the Gemini API after that date. For the full cross-provider list see the AI Media Model Retirements Tracker and OpenAI Sora Shutdown Explained.

How much do AI media models cost?

Pricing units differ by provider: monthly plans with credits, per-image or per-token API rates, or per-second video billing. Figures are in US dollars before tax as shown on the provider page on the check date.

As of 2026-11-15. ElevenLabs showed an introductory first-month price on Starter and Creator on the check date; the table shows the recurring price. Runway credits on Standard and Pro do not roll over; purchased top-up credits do not expire.

Can I use AI-generated media commercially?

Commercial use depends on the plan and, for open weights, the licence. The table records only what each provider's pricing or licence page states.

As of 2026-11-15. Plan terms are not legal advice on copyright ownership: read the provider's terms of service and your own legal position. For workplace rules on likeness, labelling and copyright see AI Image and Video Generation at Work and the AI Training Data Provenance and Copyright Exposure Guide.

Which AI media models carry provenance marks?

Provenance is the field most often missing from provider pages. C2PA Content Credentials are signed metadata describing where content came from and how it was edited (C2PA compares them to a nutrition label); SynthID is Google DeepMind's invisible watermark for images, video, audio and text. They answer different questions and can be used together. See SynthID vs C2PA Content Credentials and the AI Content Provenance and C2PA Guide.

As of 2026-11-15. On the verification side, Google says SynthID verification in the Gemini app has been used 50 million times and is expanding to Search and Chrome, and that C2PA verification began rolling out in the Gemini app on 20 May 2026. The SynthID Detector portal is open to everyone and covers content from Google and partners including OpenAI, NVIDIA and Kakao. The current C2PA specification linked from c2pa.org is version 2.3. Anthropic notes that a missing mark does not prove content is human-made, a caution that applies to every scheme on this page.

What safeguards cover real faces and voices?

• ByteDance Seed: the Seedance 2.0 launch post says using real human portraits as subject references requires identity verification or prior legal authorisation.
• ElevenLabs: instant voice cloning starts on Starter and professional voice cloning on Creator; Scale and Business include 3 and 10 professional clones.
• Google: Omni Flash in YouTube Shorts Remix and YouTube Create is limited to users aged 18 and over; audio references for Omni are limited to voice at launch.
• OpenAI: the Image API offers a moderation parameter (auto by default, or low) and recommends a hashed safety_identifier for apps serving individual end users; GPT Image models may require API Organisation Verification.
• Runway: Pro includes 1 custom voice clone and Max up to 3.

Voice cloning and face references are the features most often abused in fraud. See the Voice AI Agent Security and Fraud Guide, the Deepfake and AI Impersonation Guide and the Deepfake and Synthetic Media Laws Map.

What changed in 2026?

How do I choose an AI media model?

Factual criteria only; weigh them against your own needs.

• Lifecycle: check the provider's deprecation page and the notice period it has given before; OpenAI's image notices on this page ran about six months.
• Output limits: confirm clip length, resolution and aspect ratio on the model page (for example Veo 3.1 offers 4, 6 or 8 second clips at 24 FPS).
• Rights: confirm commercial use for the exact plan, and for open weights the revenue threshold in the licence.
• Provenance: if you must label AI content (for example under Article 50 of the EU AI Act), prefer a route that documents C2PA or watermark output. See the EU AI Act Article 50 Transparency Obligations Guide.
• Data residency and region: Veo 3.1 on Google Cloud is listed in us-central1 and accepts English prompts; check regions for any enterprise route.
• Likeness controls: read the provider's rules on real faces and voice cloning before staff use them.

Where identity and security fit

Media APIs are called with API keys and service accounts that need the same scoping, rotation and monitoring as any machine credential, and aggregator platforms such as Runway's MCP server add another path for agents to spend credits or generate content. Cloned voices and synthetic faces also defeat weak identity checks. See the Deepfake-Resistant Identity Verification Guide and the AI Model Provenance and Watermarking Guide.

Practitioner checklist

• Keep an inventory of every media model and plan in use, with its owner, API key holder and retirement date.
• Subscribe to each provider's deprecation or changelog page and diary the next shutdown (gpt-image-1-mini and gpt-image-1.5 on 1 December 2026).
• Record the commercial terms for the exact plan each team uses, and the revenue threshold for any open-weight licence.
• Decide which outputs must carry C2PA Content Credentials or a watermark, and test that your editing pipeline does not strip the metadata.
• Write a rule on real faces and voice cloning: consent, approval and record keeping before any clone is created.
• Store media API keys in a secrets manager with per-application keys and spending limits.
• Recheck prices and model names against the provider page before purchase; promotional rates on this page end on stated dates.

Standards and references

• C2PA: Coalition for Content Provenance and Authenticity
• Google DeepMind: SynthID
• OpenAI: Deprecations
• Google: I/O 2026 announcements
• Google Cloud: Veo 3.1
• Anthropic: How Claude marks AI-generated content
• ByteDance Seed: Official launch of Seedance 2.0
• Black Forest Labs: FLUX 3 Image Klein

Also checked on 2026-11-15: Google DeepMind: Lyria, OpenAI: Image generation, ElevenLabs: Pricing, Kling AI, Kuaishou investor relations: Kling AI, Runway: Pricing, ByteDance Seed: Seedance 2.0, Stability AI: Stable Image, Suno: Pricing.

Related NHI Mgmt Group resources: AI Video Generators Compared · AI Image Generators Compared · ElevenLabs Explained · Suno AI Explained · Runway AI Explained