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
Nano Banana 2 banner ratios 4:1 and 8:1: Pro does not list them
On Sume, Nano Banana 2 lists 4:1, 1:4, 8:1 and 1:8 on top of Pro's ratios. Pro returns 400 for them. Compare both lists and the Google page I read.
Pika Music: lyrics, voice and track references vs Sume Music Router
Pika Music takes prompts, tagged lyrics, a voice reference and a music reference. Sume's Music Router is prompt-first on Lyria 3.5. See what that changes.
Pika Speech: 5-minute requests, 5-second clones vs Sume TTS
Pika Speech lists 5 minutes per request, 48 kHz output and a 5-second clone at $0.01 a minute. What that means next to Sume's TTS Router and its limits.
Reasoning effort levels: GPT-6.1 Sol, Claude, Grok 4.7, DeepSeek
Effort names differ by vendor: five levels on GPT-6.1 Sol and Claude, four on Grok 4.7, three on DeepSeek. What Sume's picker shows and which rows have no knob.
Seedance bitrate_mode standard or high: which models accept it
bitrate_mode takes standard or high on Sume's Seedance 2.0 family. Seedance 2.5, Wan 3.0, MiniMax H3 and Gemini Omni Flash do not take it.
Seedream 4.5 on Sume: edits return one image, 4:5 is 1920x2400
What Sume sends to Seedream 4.5: edit calls ask for one image whatever n says, 4:5 and 5:4 become 1920x2400 and 2400x1920 boxes, and refs are cut to 10.
Swap a product in an existing video: Gemini Omni edit vs Seedance
To replace an object in a finished clip on Sume, use gemini-omni-flash-1.1 with video_url. Seedance reference-to-video makes a new clip from references.
Which agent LLM can read video frames? Image input by model
GPT-6.1 Sol, Claude 5.5, Grok 4.7 and DeepSeek Flash accept images; one DeepSeek row is text only. How a Sume agent checks a render with stills.
Which Sume image models accept aspect_ratio auto on an edit?
Only GPT Image, Nano Banana and Seedream 4.0 list auto on Sume. Grok, Seedream 4.5 and 5.0 Lite, and FLUX.2 return 400. Table plus a nearest-ratio helper.
Grok Imagine on Sume: 4:5 returns 400, 1080x1350 does not
Grok Imagine has no 4:5 on Sume. aspect_ratio 4:5 returns 400, but 1080x1350 snaps to the nearest native ratio, 2:3. What to send for Instagram portrait.
Which LLM for each step of a video agent: script to assembly
A hypothesis for matching LLMs to video agent steps (script, shot list, frame check, assembly), using documented tool prices, to test with a Format A/B run.
AI video model bake-off: one prompt, three models, your own scores
Leaderboards rank models on someone else's prompts. Run your own bake-off: one prompt, three Sume model ids, fixed settings, and a short scoring sheet.
Gemini Omni Flash 9:16: two ratios only, and a vertical ad check
On Sume, gemini-omni-flash-1.1 accepts 16:9 and 9:16 only, 3-10 s with audio always on. Ratio table across Sume video models.
Gemini TTS voices: 150+ or 2,000+? Two sources, one test plan
Google's changelog says 150+ Gemini TTS voices; a tracker says 2,000+. Do not pick by count. A voice test plan, and how voices are selected on Sume.
Gemini TTS limit: 16,384 output tokens at 25 per second is 10:55
Gemini 3.8 TTS is reported to cap output at 16,384 tokens with 25 audio tokens per second, about 10 minutes 55 seconds. Sume's TTS cap is 1,200 seconds.
Kling 4.0 makes 30 s clips; Sume's Kling 3 stops at 15 s
Kling's page says Kling 4.0 makes native 3-30 s clips. Sume lists kling-3 at 4-15 s, so here is how to get 30 s on Sume today with other models or two clips.
Kling 4.0 4K output vs Sume: 1080p for Kling 3, 4K on one model
Kling says Kling 4.0 outputs 4K. On Sume, kling-3 stops at 1080p and gemini-omni-flash-1.1 is the model that lists 4K. Rows and the 4K price per 10 s.
Kling 4.0 multilingual text in video: a test plan before you trust it
Kling says Kling 4.0 renders text, logos and emoji in 10+ languages. A 12-sample test plan to run on any video model before a localized ad depends on it.
New image model out? Check Sume's catalog before you integrate
FLUX 3 and Ideogram 4.5 launched on Higgsfield this week. Here is how to tell whether Sume lists a new image model, and what to use while you wait.
Reference audio for AI video: which Sume models take audio_url
Seedance 2.x, Wan 3.0 and MiniMax H3 honor audio references on Sume; Gemini Omni, Genjutsu and Recast take video but not audio. Rules and a check.
Seedance 2.5 aspect ratios on Sume: eight values incl. adaptive
Sume lists eight aspect ratio values for seedance-2.5: auto, adaptive, 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Uses in ad placements and what is missing.
Seedance 2.5, Wan 3.0, H3, HappyHorse, Ray3.2: which run on Sume?
Seedance 2.5, Wan 3.0 and MiniMax H3 have Sume catalog ids; HappyHorse 1.1 and Luma Ray3.2 do not. Ids, clip lengths and resolutions as read 2026-10-02.
Shortest AI video clip each Sume model accepts: 2 s Wan to 5 s H3
Wan 3.0 accepts 2 seconds, Gemini Omni 3, Seedance and Kling 4, MiniMax H3 5. How to get a shorter cut when the model's minimum is too long.
A 12-headline text rendering test for any image model
FLUX 3 and Ideogram 4.5 both claim better text. Run your own 12 headlines through the models Sume lists and review the outputs by eye, not by launch post.
Video model leaderboard rank: how to turn Elo into an API choice
Hedra's board puts Seedance 2.0 at Elo 1,225 (#2) and HappyHorse 1.1 at 1,149 (#4). What a rank tells an API buyer and what to test instead.
Dictation API: AssemblyAI cleaned text vs Sume STT word timings
AssemblyAI's Dictation API returns send-ready text. Sume STT returns a transcript with words[] timings and no cleanup flag, so you do the filler removal.
AssemblyAI text to speech: coming soon, and what Sume has today
AssemblyAI's product menu lists a Text-to-Speech API as coming soon. Sume TTS is available now: what a call takes, the 20000-character cap, and word timings.
Creatify Boreal talking clips vs Sume's still-plus-audio route
Creatify says Boreal's gains are smallest on single-person talking clips. Sume makes every speaking shot from an accepted still plus TTS audio via Fabric.
Deepgram nova-3-pharma vs Sume STT: drug-name transcripts
Deepgram added nova-3-pharma for English drug names. Sume STT has one public model, sume/stt-1.0, so check each drug name against word timings.
ElevenLabs character limits by model vs Sume TTS 20,000
ElevenLabs lists 5,000 characters for v3, 10,000 for v4 and 40,000 for Flash v2.5. Sume TTS 1.0 takes up to 20,000 characters in one request.
ElevenLabs speech to speech API: what Sume offers instead
ElevenLabs lists speech-to-speech voice changer models. Sume has no voice-to-voice endpoint: transcribe with STT, edit the text, then run TTS.
FLUX.2 flex steps and guidance: BFL has them, Sume does not
BFL lists adjustable steps and guidance only for FLUX.2 flex ($0.06/MP). Sume lists flux.2-flex but rejects unlisted parameters with 400.
FLUX.2 max grounding search: what it is, what Sume lists
Only FLUX.2 [max] does web-grounded generation at BFL. Sume lists flux.2-pro and flux.2-flex, so grounding is not available there.
GPT Image aspect ratios for a Pinterest pin (2:3) and Instagram (4:5)
Sume lists 17 aspect ratios, including 2:3 for a Pinterest pin and 4:5 for an Instagram portrait. A model only accepts the ratios its catalog row lists.
HappyHorse 1.0 API: what Runway lists and how to check Sume's models
Runway lists HappyHorse 1.0 at 3-15 seconds, 720p and 1080p. Sume's video docs name no such id, so read /v1/videos/models before you plan around it.
HeyGen cinematic avatar API: 3 looks per shot vs Sume's one
HeyGen cinematic_avatar takes a prompt and 1-3 avatar looks with no script. Sume Avatar Video resolves one avatar per final video from a script.
Kling 4.0 21:9 ultrawide: check Sume's aspect ratios
Kling 4.0 supports 21:9 output. Sume documents 21:9 as a ratio, but each model lists its own supported_aspect_ratios, so read them before sending.
Kling 4.0 Flash early access: how to tell what Sume lists
Kling 4.0 Flash opened to a limited group on Sept 28 with full launch in October. On Sume, a model is callable when the live catalog lists it.
Kling 4.0 stereo audio vs Sume's generate_audio flag
Kling 4.0 moves to two-channel stereo audio. Sume exposes generate_audio and per-model audio capability; check the catalog for what each model reports.
Ray 3.2 API aspect ratios and resolutions, and Sume's lists
Luma lists six Ray 3.2 aspect ratios (9:16 to 21:9) at 540p, 720p and 1080p. Sume reports ratios per model in supported_aspect_ratios and rejects size.
Seedance 2.0 4K price: Luma 261 credits/sec, Runway 150, Sume 1080p
Luma lists Seedance 2.0 at 4K for 261 credits per second; Runway's docs list 150. Sume's docs list seedance-2 at 480p, 720p and 1080p, with no 4K entry.
Nano Banana multiple images at once: n range vs Gemini's count
Google says Gemini won't always return the exact image count you ask for in a prompt. On Sume, set n, and read each model's n range from the catalog.
Nano Banana video to image: poster from a video via Sume stills
Google's Gemini API takes a video as context for a thumbnail or poster. Sume's images API takes image references only, so grab a still frame first.
Runway video_to_hdr alpha and ProRes 4444: what Sume returns
Runway's /v1/video_to_hdr now keeps source alpha and delivers ProRes 4444. Sume docs describe no alpha or ProRes output; a completed video job is a download.
Speechmatics Linden for voice agents vs Sume batch STT jobs
Speechmatics announced Linden for voice agents. Sume STT is a bounded batch job with a webhook, not a live stream, so live agents need streaming.
Synthesia Interactive Avatar API vs Sume rendered avatar clips
Synthesia headlines a live Interactive Avatar API. Sume avatar video is script-driven and rendered as a job: submit, poll, then fetch the clip.
Adobe Firefly Image 5 API: native 4 MP vs Sume resolution tiers
Firefly's Image5 model is described as native 4 MP with Instruct Edit. Sume does not list a Firefly model; it uses 512, 1K, 2K and 4K tiers.
Fish Audio Drama 3 single-word fix vs Sume sentence segments
Fish Audio says Drama 3 preview can fix a single word. Sume has no word-level repair: regenerate a sentence and join takes with Timeline audio concat.