Skip to main content
The Hedra v3 API dispatches every leading image, video, and audio model behind one endpoint, one key, and one bill. This page is a human-readable tour of the catalog and the newest additions. For the always-current, machine-readable list — including exact input schemas, resolution and duration menus, and prices — use:
Each model below lists the public id you submit to; find its Run a model entry in the API Reference for a copyable request body and every input field.

​
Recently added

​
Image models

Ideogram V4 — ideogram-v4 Text-to-image with three render tiers merged into one public id. Select a tier with input.quality (turbo, balanced, or quality). Priced per output megapixel. Strong for poster-ready text and layout. MAI-Image-2.5 — mai-image-2-5 Microsoft AI’s image model. Base and pro tiers are merged into one id; select a tier with input.quality. Supports both text-to-image and edit modes. HiDream-O1-Image — hidream-o1-image Standard and Dev tiers merged into one id (select with input.quality). Supports text-to-image, edit, and subject personalization from up to 10 reference images. Qwen Image 2.0 — qwen-image-2 Alibaba’s Qwen image model. Standard and pro tiers merged into one id (select with input.quality). Supports text-to-image and edit with 1–3 reference images per edit. Reve 2.1 — reve-21, reve-21-edit, reve-21-remix Three input modes, each exposed as a separate public id:
  • reve-21 — text-to-image
  • reve-21-edit — single-source edit
  • reve-21-remix — remix with 1–8 reference images

​
Video models

Vidu Q3 — vidu-q3, vidu-q3-reference Video with native dialogue and sound up to 16 seconds — the longest single clip in the catalog. Generate from a text prompt, a start frame, or between a start and end frame. standard and turbo tiers via input.quality. The separate vidu-q3-reference model takes 1–4 reference images (reference-to-video). Luma Ray 3.2 — luma-ray-32 Text-to-video at 540p, 720p, or 1080p, in 5- or 10-second clips. Wan 2.7 — wan-2-7 Alibaba’s video model with native audio, exposed under one public id. Three input modes:
  • Text-to-video, 2–15 seconds.
  • Image-to-video from a first frame with an optional last frame, 2–15 seconds.
  • Reference-to-video from up to 4 reference images, 2–10 seconds — keeps subjects consistent across the clip.
LTX-2.3 — ltx-2-3 Fast and Pro tiers merged into one id (select with input.quality). Supports resolutions up to 4K (2160p); Fast reaches 20 seconds. Audio is synthesised in the same forward pass as the video — there is no audio-off toggle. PixVerse V6 — pixverse-v6 Text-to-video and image-to-video at 360p, 540p, 720p, or 1080p, in 1–15 second clips. Native audio via the generate_audio toggle. Accepts seed for reproducibility.

​
How to pick a model

  1. List what’s available for your modality. GET /v3/models returns every model with its id, human name, and modality.
  2. Read the model’s typed schema. GET /v3/models/{id}/openapi.json returns the exact input fields — resolution menus, duration bounds, whether the model accepts reference frames, and so on.
  3. Estimate the cost. POST /v3/models/{id}/estimate returns the credit cost for a specific request body before you run it.
  4. Submit the job. POST /v3/models/{id} returns a job_id; poll it with GET /v3/jobs/{job_id}/status and read the output at GET /v3/jobs/{job_id}.
New model families are added continuously. Prefer GET /v3/models at runtime over hard-coding a specific model id when your integration should follow the latest release.