What AI models does Sume support? Every model id, by API surface

Sume supports 10 video generation models on POST /v1/videos, 17 image models on POST /v1/images, 3 Music Router ids, and 5 TTS Router ids. On top of those come Avatar 1.0 and three models that animate a still image: VEED Fabric 1.0, MiniMax H3 Max Lip Sync, and Kling 3.0 Motion Control.

The id lists below were read from Sume's API contract code on 2026-09-26. Endpoints and rules come from the Models overview, Video generation, and Image API docs and the Sume API reference. The public GET /v1/catalog confirms them, as shown at the end.

Which models can I call on each API surface?

The table shows the model values each surface accepts, or its fixed public model id where there is no model picker.

How are video and image model ids written?

Video ids are bare catalog ids such as seedance-2, with no provider-org prefix, and the legacy POST /v1/video-router/generate takes the same ids. An OpenRouter-compatible video API covers sume/auto and when to pin a model instead.

The Image API names models in org/slug form, such as openai/gpt-image-2.5 or bytedance-seed/seedream-4.5, and accepts the bare Image Router ids in the table as aliases for their org/slug equivalents. The legacy /v1/image-router/* routes still work but are deprecated.

What do the avatar and still-image models do?

The avatar and still-image surfaces do different jobs:

• Avatar 1.0 is two steps: create a reusable avatar, then generate talking videos from a script or multi-scene video_inputs. quality is standard, plus (the default), or max. See How to create a reusable AI avatar.
• Avatar Face Swap (Beta) swaps a ready avatar face onto a public source video.
• VEED Fabric 1.0 turns a still image, or a ready avatar's still, plus Sume-hosted audio into a talking clip. MiniMax H3 Max Lip Sync takes the same body for 5–14.8 seconds of audio. The lip sync API guide compares them.
• Kling 3.0 Motion Control animates a still with the motion of a driving video you supply, and the output length follows that video. See the motion control API guide.

Which media tools does Sume offer?

The API also has tools for speech, stills, and clip assembly. STT, RMBG, and the two upscalers are documented in the Sume API reference; the other tools each have a guide linked from the Models overview.

• Speech and pictures: STT 1.0 (POST /v1/stt-1.0/transcribe) returns text with word-level timings, RMBG 1.0 (POST /v1/rmbg-1.0/remove) removes an image background, and Image Upscale 1.0 and Video Upscale 1.0 upscale stills and clips.
• Clip prep on one hosted clip: Video inspect, Video frames, Video trim, Audio detach, and Video filter.
• Assembly: Timeline 1.0 joins an audio spine and ordered video into one MP4, Timeline compose puts a still and a video in one frame, and Timeline audio joins or splits Sume-hosted audio.
• Captions and research: Video captions burns captions onto an existing public video URL, and Trending videos searches TikTok trending video metadata.

Which model surfaces are retiring?

Video 1.0 and Image 1.0 are retiring soon. Video 1.0 maps to sume/auto and ignores routing_preset, and Image 1.0's public URLs remain compatibility aliases for Image Router Auto, so send new work to POST /v1/videos or POST /v1/images. Video 1.0 and Image 1.0 are retiring soon walks through the switch. Music 1.0 is retiring too; its requests resolve through the Music Router.

How do I confirm the list myself?

GET /v1/catalog is one of the few /v1 routes that needs no API key. Each entry carries its invoke_url, model_ids, and availability, and paid capabilities include USD pricing metadata where available. Video ids sit under the video-router entry and image ids under image-router.

The per-surface catalogs need a key and add per-model detail: GET /v1/videos/models, GET /v1/images/models, GET /v1/music-router/models, and GET /v1/tts-router/models. Check a model there before you pin it, because limits are not uniform. On the Image API, a parameter the model's supported_parameters do not list is rejected with 400 unsupported_parameter.

curl -s https://api.sume.com/v1/catalog \
  | jq '.data[] | { id, invoke_url, model_ids }'