infer
Get started
CHANGELOG · UPDATED WITH EVERY RELEASE

What's new.Straight from the team.

Feature launches, model availability, and fixes worth knowing about — everything that changed on Infer, newest first.

Reference-to-video and full video controls in the Studio

The Studio composer now exposes each model's real capabilities — reference images, keyframes, aspect ratios, resolution, duration, FPS, and audio.

  • NewReference-to-video mode: guide a video generation with up to 5 reference images on models that support it (e.g. HappyHorse 1.1).
  • NewKeyframe slots: video uploads are now labeled as start/end frames, so you can pin the first and last frame of a generation.
  • NewPer-model video controls: aspect ratio, output resolution (720p/1080p), duration window, and an audio on/off toggle — each picker only shows what the selected model actually supports.
  • NewFPS control (24/50) on LTX Fast, and durations up to 20 seconds where the model allows it.
  • ImprovedThe API playground derives the same video knobs from each model's schema, so Studio and API stay in sync.

Unlimited plan and clearer tier limits

A new Unlimited tier joins Plus and Pro, and each tier's limits are now consistent across the Studio.

  • NewUnlimited plan: generate without per-run credit anxiety at a flat monthly price.
  • ImprovedPer-tier controls: Pro unlocks each model's full duration window and higher run counts; Plus stays on the fastest, most economical settings.
  • FixedYour Library stays viewable after a subscription lapses — past generations are never locked away.

HappyHorse 1.1 and updated Seedance durations

New top-ranked video model generation, and duration options tuned per model.

  • ModelsHappyHorse 1.1 is live for text-to-video and image-to-video, with all nine aspect ratios and reference-image support. HappyHorse 1.0 has been retired from the Studio picker.
  • ModelsSeedance 2.0 offers 5s and 10s durations (a 15s option was briefly available and has been withdrawn upstream).
  • FixedLong prompts with attached media no longer fail with URI-too-long errors.

Wan 2.2 text-to-video and live model availability

More models in the Studio, and the model list now updates live — new launches and availability changes appear without waiting for an app release.

  • ModelsWan 2.2 text-to-video is available in the Create Studio.
  • NewSite-wide announcement banner: maintenance windows and incidents are now surfaced at the top of the app.
  • ImprovedModel availability, image-input support, and first/last-keyframe support are now driven live from the model catalog.
  • FixedWhen a provider declines a generation for content-moderation reasons, you now see the actual reason instead of a generic error.

Console navigation refresh and reliability fixes

A cleaner API console, plus a batch of quality-of-life fixes across the app.

  • ImprovedThe API console navigation has been restructured, with a clearer switch between the Studio and the API surface.
  • FixedDownloads save the media file directly instead of opening it in a new tab.
  • FixedBlocked trials show a clear explanation instead of an endless spinner, and raw provider errors are no longer leaked into the UI.