Models
The models behind Sume's video, image, speech, and music endpoints: ids, clip lengths, resolutions, aspect ratios, audio, references, and migration.
Start with What AI models does Sume support? Every model id, by API surface
GPT Image 2.5 quality levels: low, medium, high, xhigh, max and auto
GPT Image 2.5 has six quality values: auto, low, medium, high, xhigh and max. What each does to price on Sume and which one runs when you send nothing.
GPT Image 2.5 release date and model ids: Flare and Sunburst
OpenAI's GPT-Image-2.5 Flare and Sunburst carry the snapshot date 2026-09-08. The model ids, dated snapshots, and how Sume names the same two models.
GPT Image 2.5 resolution: the size rules and the 4K ceiling
GPT Image 2.5 takes custom sizes: both edges a multiple of 16, longest edge 3840, 655,360 to 8,294,400 pixels. What that means for 1080p, 1440p and 4K.
Grok Imagine image API: the x-ai/grok-image id on Sume
Sume lists Grok Imagine as x-ai/grok-image on its image API: thirteen aspect ratios (no 4:5), one reference image for edits, and a $0.025 price line per image.
Grok Imagine aspect ratio: 21:9 and 5:2 on xAI vs Sume
xAI added 21:9 and 5:2 to Grok Imagine image on Aug 28. Which ratios Sume's grok-image accepts, what the 400 looks like, and a call that works.
Grok Imagine Video 1.5 reference images API: xAI vs Sume
xAI's API takes up to seven reference images, text-to-video and 1080p for Grok Imagine Video 1.5. Sume's API reference lists it as first-frame image-to-video.
How much VRAM for AI video generation? Numbers from model cards
From 14 GB to 80 GB: the VRAM figures Tencent, Wan-AI and Genmo publish for their open video models, and what each number was measured with.
HunyuanImage 3.0 API: open weights, no listed Sume model
HunyuanImage 3.0 is an 80B-parameter open-weights image model from Tencent. Sume's image docs do not list it; here is what the card says and the hosted route.
HunyuanVideo 1.5 low VRAM: 14 GB with offloading
Tencent's HunyuanVideo-1.5 card lists 14 GB as the minimum GPU memory with offloading on. The OOM settings and step-distilled option it names.
HunyuanVideo API: is there a hosted one, or only weights?
HunyuanVideo is published as downloadable weights. Sume's docs list no HunyuanVideo API, so here is what running it takes and the listed video ids.
Ideogram 4.0 API: request shape and what Sume lists
Ideogram 4.0 has its own endpoint with text_prompt or json_prompt. Sume's image catalog lists Ideogram V3 only; here is how the two differ.
Imagen 4 API shut down in the Gemini API: what Sume's catalog lists
Google says Imagen is shut down in the Gemini API and points to Nano Banana. Sume's image catalog lists google/imagen-4-fast, imagen-4-ultra and Nano Banana ids
AI video generator for Instagram Reels: 9:16 models and durations
Which Sume video models take 9:16 and for how many seconds: seedance-2.5 and wan-3.0 reach 30, minimax-h3 5 to 15, gemini-omni-flash-1.1 3 to 10, and sume/auto.
Kling 3 4K API: Kling's own tier and what Sume lists
Kling 3.0 lists 4K on Kling's own capability map, but Sume's kling-3 lists 720p and 1080p and prices no resolution tier. Sume's 4K route: gemini-omni-flash-1.1.
Kling 3 duration and aspect ratio limits: Kling's own vs Sume's
Kling's own map lists 3 to 15 seconds for Kling 3.0; Sume's kling-3 takes 4 to 15 seconds at 16:9, 9:16 or 1:1. Values outside the list are refused.
Kling 3 first and last frame API: request, limits and a gotcha
On Sume, kling-3 takes a first and optional last frame via frame_images, for 4-15 seconds. Send frame one in the shape you want; aspect_ratio is text-only.
Kling 3 lip sync with your own audio: what each API does
Kling lists lip sync driven by text or audio. Sume's kling-3 takes no audio input, only a generate_audio switch; models with audio references are listed.
Kling 3 native audio: what the on and off price difference is
On Sume, kling-3 costs $0.14 a second silent and $0.21 with audio. Audio is on unless you send generate_audio false, so the default is the higher price.
Kling 4.0 API: what Sume lists and what to use today
Kling 4.0 API availability could not be confirmed from Kling's own pages on 2026-09-29. Here is what Sume's docs list today for video, and where Kling appears.
Kling motion control API in Python: submit, poll, fetch
Call Kling 3.0 motion control from Python with requests: submit with an Idempotency-Key, poll the status URL until terminal, then fetch the result when ready.
Kling motion control camera movement and orientation
In Kling motion control, character orientation decides whether the camera follows the video or the image. What Kling says, and Sume's character_orientation.
Kling motion control face consistency: element binding
Kling 3.0 Motion Control keeps a face steady by binding a facial element to the character image. How it works, and what Sume's endpoint takes instead.
Kling motion control with multiple people in the video
Kling motion control animates one character. With two or more people in the reference, it uses the one with the largest share of the frame.
Kling motion control not working: causes and fixes
A short, cut-off or wrong-person Kling motion control result usually traces to the reference video. Kling's causes, and the errors Sume's endpoint returns.
Kling motion control prompt: what it can change
In Kling motion control the video supplies the movement and the prompt covers background and looks. Kling's examples, and Sume's 2,000-character prompt.
Kling motion control reference video requirements
Kling asks for one character in one continuous shot, 3 to 30 seconds, short edge 340 px or more. Which limits Sume's endpoint also enforces.
LTX 2 open source: what Lightricks released, and the license
LTX-2 is an audio-video model with open weights under the LTX-2 community license. What the Hugging Face cards list, and why Sume's catalog has no LTX.
Luma Ray 3.2 20 seconds 1080p: how Sume's catalog limits read
Luma says Ray3.2 makes up to 20 seconds at 1080p. In Sume's video docs, seedance-2.5 is the one model stated to accept 20 seconds at 1080p.
Luma Ray 3.2 API: 16 keyframes, and Sume's two-frame route
Luma says Ray3.2 takes up to 16 keyframes per clip. Sume lists no Luma model; chain its first and last frames to cover more key images.
Luma Ray3.2 modify: what is confirmed, and Sume's edit route
Luma's Ray3.2 notes list keyframes, 1080p, HDR and reframe but no quotable modify details. Sume's video edit route is gemini-omni-flash-1.1 with video_url.
Lyria 3 Pro or Lyria 3.5: which model id do I send?
Sume's Music Router accepts sume/music-auto, lyria-3.5 and lyria-3-pro. What each id does, how to pin one, and what Google lists for the models.
Lyria 3.5 API: model id, request and how to call it
Lyria 3.5 is in the Gemini API as lyria-3.5. Here is what Google's page says about it and how to call it from Sume with one music request.
AI music generator with vocals: Lyria 3.5 lyrics by API
Lyria 3.5 can sing. How to ask for vocals or an instrumental, steer lyrics with section tags, and read the lyrics back from a Sume music job.
Lyria 3.5 release date: where it is available
Google announced Lyria 3.5 in the Gemini app on 2026-09-04. Where Google says you can use it, and where Sume lists it.
How long is a Lyria 3.5 song, and how do I set the length?
Lyria 3.5 makes songs of a couple of minutes, and length is set in the prompt. What Google says, what Sume rejects, and prompts that steer duration.
Midjourney V8.2 API: does Sume list a Midjourney model?
Sume's image catalog has no Midjourney model, so a Midjourney id returns 404 model_not_found. Read the live list with GET /v1/images/models and try sume/auto.
2K AI video generator API: MiniMax H3 at 2K on Sume
MiniMax H3 makes video up to 2K. On Sume you ask for resolution 2K on minimax-h3, an upscale of native 768p. Request, price for 10 seconds, and limits.
MiniMax H3 languages: which spoken languages does it support?
MiniMax says H3 has stable support for 11 dialogue languages and partial support for others. Sume's docs list no language set, so test yours first.
MiniMax H3 Max 1080p: a latent refinement from native 768p
MiniMax H3 Max on Sume offers 480p, 768p and 1080p. The 1080p tier is a latent refinement from native 768p. What that means for price and output checks.
What is MiniMax H3 Max? The post-trained variant, explained
MiniMax H3 Max is a variant post-trained by fal.ai on MiniMax H3 for faster generation. Its resolutions, lengths and modes, and how Sume lists it.
MiniMax H3 Max lip sync API: a still plus audio, 5 to 14.8 seconds
Sume runs MiniMax H3 Max lip sync at POST /v1/minimax/h3-max/lip-sync: send a still and Sume-hosted audio of 5 to 14.8 seconds. Body, resolutions and price.
AI video with native stereo sound: MiniMax H3 audio through the API
MiniMax H3 generates native stereo sound in the same pass as the video. On Sume audio is always on, generate_audio false is refused, and audio steers.
MiniMax H3 open weights on Hugging Face: what is in the release
MiniMax H3 has a model card on Hugging Face: two checkpoints, a 33B-parameter Transformer, a community license and a 4-GPU serving example. What to check first.
MiniMax H3 cache and sparse attention: what the open release has
MiniMax's H3 open release ships full-attention inference only; sparse attention comes later. About 13B parameters can be cached, not loaded.
AI video editing with a text prompt: MiniMax H3 and Sume's edit path
MiniMax H3 is described as editing existing video from instructions. What the vendor says, what Sume's H3 ids accept, and the id with an edit field.
Mochi video model: what Mochi 1 is, needs and can't do
Mochi 1 preview is Genmo's Apache 2.0 open video model: 480p output, about 60 GB VRAM on one GPU, and it is not in Sume's video catalog.
Nano Banana 2 Lite API: what Google shipped and what Sume lists
Nano Banana 2 Lite is Google's gemini-3.1-flash-lite-image model. What Google says it does, and which Nano Banana models Sume's image API lists today.
Nano Banana Pro custom aspect ratio: what Sume does
Nano Banana Pro on Sume takes 11 native aspect ratios. A custom ratio or WxH size is snapped to the nearest one. Which, and how to check.