# Competitors and capabilities

**Observed September 26, 2026.** This is a map of documented offerings, not an ordered quality leaderboard. Availability can vary by product, region, plan and API.

| Company / offering                    | Evidence from primary sources                                                                                                                                                                                                                  | What to test before choosing it                                                                                                                                                                                                             |
| ------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| Google — Gemini Omni Flash / Veo 3.1  | The API guide distinguishes conversational, multimodal generation/editing from Veo's frame and extension controls.                                                                                                                             | Whether multi-turn changes preserve untouched elements; API access, latency and quotas in the target region. [Docs](https://ai.google.dev/gemini-api/docs/video)                                                                            |
| Runway — Gen-4.5 and production tools | Its generation guide documents text-to-video and image-to-video; the developer platform offers multiple models and output options.                                                                                                             | Approved-shot rate and integration into an actual editing workflow. [Generation guide](https://help.runwayml.com/hc/en-us/articles/37425232841875-Getting-Started-with-Generative-Video)                                                    |
| ByteDance — Seedance 2.5              | The vendor describes up to 30 seconds per generation, extension, reference-video control and audiovisual editing.                                                                                                                              | Longer-scene consistency, reference adherence, endpoint availability and total input/output charges. [Model page](https://seed.bytedance.com/en/seedance2_5)                                                                                |
| Kuaishou — Kling 3.0                  | Its February launch describes native audio, multimodal references and up to 15-second video, with multi-shot storyboard controls in Video 3.0 Omni.                                                                                            | Product and character fidelity across cuts, dialogue handling, and access to the specific model variant. [Launch](https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-launches-30-model-ushering-era-where-everyone-can-be) |
| Luma — Ray family                     | Ray3.14's announcement documents native 1080p and video modification, while explicitly excluding character references and HDR/EXR in that variant's Modify mode. This is version-specific evidence, not a claim that 3.14 is the newest model. | Exact variant controls and whether transformations preserve the intended performance. [Ray3.14 announcement](https://lumalabs.ai/news/ray3_14)                                                                                              |
| Adobe Firefly                         | The workspace combines generation, editing and partner-model selection with Adobe's own offering.                                                                                                                                              | Handoff into the team's existing tools, comparison overhead, export fidelity and the terms for the chosen underlying model. [Product page](https://www.adobe.com/products/firefly/features/ai-video-generator.html)                         |
| Higgsfield                            | Its current site emphasizes a creative suite, model access, effects, reusable executions and a community film showcase.                                                                                                                        | Whether presets accelerate useful work, repeatability across a campaign, and portability of assets. [Product site](https://higgsfield.ai/)                                                                                                  |
| HeyGen                                | Avatars, script-oriented production and localization are central offerings.                                                                                                                                                                    | Delivery and terminology fidelity across target languages, revision cost and team review. [Product site](https://www.heygen.com/)                                                                                                           |
| Synthesia                             | Business video creation, training, templates, shared workspaces and localization.                                                                                                                                                              | Updating a real training library, reuse, approval flow and integration with its delivery system. [Product site](https://www.synthesia.io/)                                                                                                  |
| Wan — open model alternative          | The Wan2.2 repository publishes code and models under an Apache 2.0 license. This is an examined open baseline, not a claim it is the latest Wan release.                                                                                      | Hardware requirements, operations, achievable throughput and task-specific quality. [Repository](https://github.com/Wan-Video/Wan2.2)                                                                                                       |
| OpenAI — Sora                         | The discontinuation notice places the app shutdown in April and API shutdown on September 24, 2026.                                                                                                                                            | Migration/export needs for existing users; exclude from new-provider assumptions unless availability changes. [Notice](https://help.openai.com/en/articles/20001152-what-to-know-about-the-sora-discontinuation)                            |

## What this means for competition

A startup cannot assume that bundling several model endpoints is a durable advantage. The same model may be available through multiple workspaces, and a supplier may offer its own aggregation or workflow tools. The defensible question is what the customer accumulates in the product: reusable references, approved brand assets, structured project history, team decisions, or outcome data.

There is also no single “best video model” for every job. A tool that produces an impressive landscape can fail a product shot if it changes the package. A natural-looking avatar can fail a training workflow if updating a sentence requires rebuilding the entire project. Evaluate at the level of the paid job.

[Back to the report](/file/68cd0367-5524-48ea-af1e-3210f975a888)