Hedra Inference

We’ve launched a prototype of our next-generation VAE, accelerating our in-house and open-source diffusion models by up to 50%.

Agent multiplayer

Realtime multiplayer is life, share your Hedra space with your team for realtime collaboration.

Hedra Develop

We’ve launched the world’s most agent- and developer-friendly API for managed image, video, and audio inference. With straightforward USD billing, it’s now broadly available.

We’re rapidly expanding the catalog with the goal of offering the largest collection of visual inference models on the market. If there’s a model you’d like to see, let us know.

• Introduced a redesigned, mobile-friendly Developer Console with a new home page, quickstart, model catalog, API key management, billing, and playground.

• The playground now builds its controls from each model’s capabilities and accepts pasted or dropped media.

• Runs stream into a dedicated Result tab with request previews, validation errors, progress, ETA, outputs, cost, and processing time.

• Improved API wallet funding, balance errors, auto-recharge eligibility, and dollar-based usage reporting.

• Expanded the API documentation with model artwork, field descriptions, input limits, webhook schemas, signed headers, and structured errors.

• Added more reliable webhooks with deduplication keys, redelivery indicators, signed delivery metadata, and a longer retry schedule.

• Improved generation status reporting with lifecycle logs, terminal progress, enhanced prompts, output asset IDs, and clearer expired-file errors.

• Generation requests now use POST /v3/models/{model}; the older /queue route has been retired.

• Public models now use one stable model ID, with mode, quality, and sizing selected through inputs.

• API terminology now consistently uses jobs, typed public IDs, UTC timestamps, USD pricing, and structured client errors.

• Standardized fields including result_url, modality, start_image, source_video, and image-reference arrays.

• Video duration must now be provided when required by the selected model.

• Standardized webhook delivery statuses, signatures, retry information, and error envelopes.

New models

• Gemini Omni Flash Video

• Seedream 5.0 Pro

• Ideogram V4

• Luma Ray 3.2

• Wan 2.7

• LTX-2.3 Fast and Pro

• PixVerse V6

• Vidu Q3

• MAI Image 2.5

• HiDream O1

• Reve 2.1

• Qwen Image 2.0

Controls

• Added Video Edit mode to Agent.

• Expanded reference-media support across Seedance, Kling, Veo, Gemini Omni Flash, FLUX.2, GPT Image, Seedream, Grok, Nano Banana, and Hailuo.

• Added reproducible seed controls to more image and video models.

• Added negative prompts to 26 supported models.

• Added advanced guidance, CFG, and step controls where supported.

• Added per-model output selection for PNG, JPEG, and WebP.

• Upgraded multimodal prompt enhancement to Gemini 3.5 Flash.

• Added per-clip trimming and clearer warnings when reference audio or video exceeds a model’s limits.

• Improved multi-speaker Hedra Avatar inputs.

• Clarified avatar duration and audio requirements.

• Improved aspect-ratio and resolution validation across image and video models.

• Retired redundant Nano Banana integrations and VEED Fabric 1.0 Fast where equivalent or better options already existed.

Agent

• Studio galleries now show generated media without mixing in uploads.

• Fixed Library search so it includes recent and uploaded assets that were previously missing.

• Improved reference-media drag-and-drop between Agent, the asset library, Composer, and Canvas.

• Agent approval cards now calculate lip-sync costs from the driving audio.

• Improved handling for Agent responses that finish after a tool call without additional text.

• Moderation messages now distinguish celebrity or likeness restrictions from prompt-language problems.

• Added clearer handling for rejected reference media and video moderation failures.

• Removed prompt shortcuts that conflicted with text input.

• Japanese IME composition no longer conflicts with keyboard shortcuts.

• Fixed misleading credit-recovery messaging.

Canvas

• Added labeled collaborator cursors.

• Improved the mobile toolbar, view switching, navigation, and asset library.

• Fixed Canvas dragging the wrong selected asset.

• Fixed video flicker when opening and closing chat.

• Fixed abrupt Canvas zooming when using Deep Research.

• Fixed stale generation states on Canvas nodes.

• Failed Canvas chunks are now contained instead of crashing the whole app.

• Canvas retry actions now work correctly.

• Improved media synchronization as generations cross into the ready state.

• Improved reliability when reconciling collaborative Canvas updates.

Assets, playback, and downloads

• Image and video downloads now use the browser’s native download flow.

• Video downloads use the asset title as the filename.

• Restored selectable video-quality controls in Chromium browsers.

• Removed the 1.2-megapixel restriction on start-frame uploads.

• Reduced memory usage when attaching large images.

• Asset details now show creator names.

• Asset details now include richer reference-audio and reference-video information.

• Improved accessibility labels across asset previews and panels.

• Fixed repeated polling of failed videos.

• Fixed retries continuing for deleted assets.

• Fixed broken video drag-and-drop.

• Improved video upload and finalization reliability.

• Improved playback and delivery reliability for Mux assets.

• Fixed completed Topaz 1080p upscales being lost during finalization.

• Improved BytePlus reference-media validation and transcoding.

• Improved corrupt-media detection and error reporting.

• Added support for additional FFmpeg-compatible video MIME types.

Skills and personalization

• Added a dedicated Personalize area for Skills and Connectors on desktop and mobile.

• Public skills can now be cloned.

• Private skills can be renamed and edited more reliably.

• Added workspace skill sharing, publishing, activation, and team catalog support.

• Added the Skill Builder public skill.

• Added an Agent tool for creating skills.

• Added a public multi-shot video workflow.

• Added workflow as a recognized skill category.

• Improved skill thumbnails and catalog presentation.

• Improved connector discovery and prompt-menu selection.

• Improved mobile connector management.

• Hidden utility skills no longer appear in the public catalog while remaining visible when used.

Billing and account experience

• API credit-pack customers can enable auto-recharge using the card saved at checkout.

• Subscription cancellation screens now show the scheduled end date.

• Corrected displayed and charged prices for audio generation.

• Corrected avatar estimates to use the actual audio duration.

• Corrected pricing for Grok video, Sora 2 Pro, Flux.2, and several image models.

• Improved insufficient-balance errors so they identify which wallet needs funding and where to add funds.

• Improved Developer Console usage and billing charts.

• Improved user name synchronization.

• Workspace invitations now retain information about who sent them.

• Fixed API-key permission and rotation behavior.

Reliability, accessibility, and performance

• Improved retry behavior for transient failures from Gemini Omni Flash, Together, Topaz, BytePlus, and other providers.

• Fixed multiple image and video models that were sending incompatible provider parameters.

• Invalid inputs are now rejected earlier with clearer, field-specific messages.

• Terminal generations no longer display stale ETAs.

• Improved application recovery when a lazily loaded page fails.

• Reduced unnecessary errors caused by deleted assets.

• Improved app-home performance, particularly on mobile.

• Improved marketing-site and blog loading performance.

• Improved dark-mode model artwork and catalog contrast.

• Kept video controls accessible on short browser windows.

• Improved accessibility across asset panels and preview surfaces.

• Added a useful route back home from the 404 page.