Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
Twitch vertical clips: crop an older 16:9 clip to 9:16
Twitch Dual Format makes vertical clips going forward. For a 16:9 clip without one, crop a 9:16 slice with video-filter: width 0.316, height 1.
Vercel 413 FUNCTION_PAYLOAD_TOO_LARGE with Sume video
Vercel Functions cap request and response bodies at 4.5 MB. Do not proxy a generated video through one; hand the client the media.sume.com URL instead.
Vercel Blob signed URL as a Sume image_url is rejected
Sume rejects signed or private URLs in input fields such as input.image_url. A Vercel Blob signed URL will fail; use a public-access store for Sume inputs.
Vercel function max duration 800s: Sume wait budget
Vercel Pro and Enterprise allow 800s, with 1800s in beta. Sume still caps a blocking wait at 30s, so video jobs need a poll loop with its own deadline.
Vercel Queues retry can resubmit a paid Sume job
A Vercel Queues consumer that times out is redelivered, and maxDeliveries is unlimited by default. Reuse one Idempotency-Key per message so Sume bills once.
Vercel Queues for Sume job webhooks: dedupe on job_id
Vercel Queues delivers at least once and bills keyed sends at 2x. To relay Sume webhooks through it, use job_id as the key and ack fast on the 10s limit.
Walmart main image size: 2200x2200 and Sume's multiples of 16
Walmart wants a 2200x2200 px main image under 5 MB. 2200 is not a multiple of 16, so generate 2208x2208 on Sume's gpt-image-2.5, then resize.
Wan 3.0 on Runway, or the wan-3.0 id as an API job on Sume
Runway's changelog lists Wan 3.0 in tool mode and workflows on paid plans. On Sume, wan-3.0 is a model id you call from code and poll as a job.
Whisper API cost per minute, and what an hour costs
OpenAI lists Whisper at $0.006 per minute and gpt-4o-mini-transcribe at $0.003. Here is the per-hour math next to Sume STT 1.0's per-minute rate.
WordPress 7.1 client-side media processing and Sume images
WordPress 7.1 resizes and converts uploads in the browser (default quality 0.82). Sume has no output_compression, so request the format you want.
xAI video status done/expired vs Sume completed/failed
xAI's Grok Imagine video status values are pending, done, expired and failed. Sume uses pending, in_progress, completed, failed and cancelled. Port a poll loop.
YouTube inauthentic content policy and bulk AI video runs
YouTube's inauthentic content policy targets mass-produced, generic AI templates. How a 100-item Format bulk queue can give every video its own input.
Zapier Catch Raw Hook: verify a Sume webhook signature
Zapier's Catch Raw Hook returns unparsed data plus headers, which a Sume HMAC check needs. Here is what to verify and the size limits to know.
Zed MCP OAuth scopes: what Sume asks for (mcp:read, mcp:write)
Zed 1.18.0 fixed OAuth for MCP servers with non-default scopes. Sume has two scopes, mcp:read (required) and mcp:write (opt-in on the consent page).
Vidu API: S2-Avatar live voice vs Sume avatar video jobs
Vidu S2-Avatar is a real-time voice model. Sume has no live session: you submit a script and a scene photo to an avatar job and fetch the finished video.
Vidu S2-Editing: live stream edits vs a Sume Omni clip edit
Vidu S2-Editing edits an incoming video stream in real time. Sume edits one finished clip with a prompt through Gemini Omni video_to_video, as an async job.
PixVerse API and R2 world model: Sume has no live session
PixVerse R2 is a persistent real-time world. Sume has no live session or action controls: video jobs return a clip you poll for or get by webhook.
Sora Videos API replacement: seconds and size on Sume
OpenAI removed the Videos API on 2026-09-24. Map its seconds and size fields to Sume's duration, resolution and aspect_ratio; size returns 400 on Sume.
Sora video.completed webhook to Sume job.completed
Sora emitted video.completed and video.failed. Sume sends job.completed, job.failed and job.canceled with an x-sume-webhook-signature header. Map the handler.
Sora input image must match size: the Sume frame_images rule
Sora's input image was the first frame and had to match the video size. Sume uses frame_images plus aspect_ratio, and rejects size on v1 models.
Kling 4.0 30 seconds: Sume lists kling-3 at 4 to 15
Kling 4.0 announces 3 to 30 second generation. Sume lists kling-3 at 4 to 15 seconds; seedance-2.5 and wan-3.0 are the ids that reach 30 seconds.
Kling 4.0 Omni Reference: 15 inputs; kling-3 on Sume takes none
Kling 4.0 Omni Reference allows up to 15 combined references. On Sume, kling-3 takes none; wan-3.0 and minimax-h3 list image, video and audio caps.
Kling 4.0 multiple keyframes: ten, versus Sume's two frames
Kling 4.0 announces up to 10 keyframes. Sume's video routes take a first and last frame; longer chains mean separate clips cut and joined.
Kling 4.0 video editing: five video inputs, Sume edits one
Kling 4.0 editing takes up to five video inputs. Sume's Omni edit route takes one video_url and cannot combine it with references.
Kling 4.0 prompt length: 8,000 tokens vs Omni's 20,000 characters
Kling 4.0 allows prompts up to 8,000 tokens. Sume documents a 20,000-character prompt cap for gemini-omni-flash-1.1; tokens and characters differ.
Wan 3.0 reference video requirements: 15 seconds and 16 fps
On Sume, wan-3.0 reference videos total 15 seconds or less and need 16 fps or more, up to 5 clips. Check each clip with video inspect and trim to fit.
Gemini Omni reference video length: 3 seconds per clip on Sume
On Sume, each Gemini Omni Flash reference video can be at most 3 seconds, with up to 3 clips. Trim longer footage first, then address clips as VIDEO_REF tags.
Higgsfield Genjutsu API on Sume: one source video, 1 to 8 images
Genjutsu Motion Transfer on Sume needs exactly one source video plus 1 to 8 reference images. Text-only or malformed bodies are not valid. Request shapes below.
Runway Enhance Frame Rate: 300 s cap and credits vs Sume fps
Runway's Enhance Frame Rate takes inputs up to 300 seconds at 1 credit per 2 seconds. Sume's Timeline sets output.fps to 24, 25, 30 or 60. Compared.
Seedance 2.5 reference limit: BytePlus says 50, Sume varies by model
BytePlus allows up to 50 multimodal references for one Seedance 2.5 clip. Sume documents limits per model, so read the catalog for the id you send. Examples.
gpt-image-2 transparent background: OpenAI preview vs Sume
OpenAI put transparent backgrounds for gpt-image-2 in preview (png or webp). On Sume, the background parameter is listed for ChatGPT Image 2.5, not gpt-image-2.
GPT Image output_compression: OpenAI has it, Sume returns 400
OpenAI lets jpeg and webp take output_compression 0-100. Sume's docs say it is not served in v1 and sending it returns 400 unsupported_parameter.
GPT Image edit with a mask and several images: which is masked?
OpenAI applies the mask to the first of several input images. Sume forwards input_references in order with mask_url, so put the image to edit first.
Grok Imagine image edit limit: xAI 5 sources, Sume catalog
xAI's Aug 28 release notes raise image editing to 5 source images. On Sume, Grok is edit-capable with n of 1; read input_references per model.
Topaz Labs and Adobe: what Sume's image upscale API takes
Adobe completed its Topaz Labs acquisition on Sep 23, 2026, and Topaz stays standalone. If you need an upscale API, here is what Sume's accepts.
Midjourney --tile seamless pattern API: what Sume accepts
Midjourney fixed --tile seams for V8.1 and V8.2 on Sep 24, 2026. Sume's image API has no tile flag; here is what a pattern request can set instead.
Nano Banana Google Search grounding: not a Sume parameter
Google documents Search grounding for its Gemini image models. Sume lists Nano Banana 2 but no grounding field, and unlisted fields return 400.
Nano Banana thinking mode and thought images on Sume
Google's Gemini 3 image models use thinking mode and uncharged thought images. Sume lists no thinking field, and you pay only for completed images.
Nano Banana 14 reference images vs Sume's 10 input_references
Google says Gemini 3 image models mix up to 14 reference images. Sume's Nano Banana models take up to 10 input_references; ChatGPT Image 2.5 takes 16.
AI product color changer: colorways through an image edit API
To recolor a product photo, send it as an input_references image with a prompt naming the new color. Sume's edit calls take up to 10 images per request.
AI change text in image: an edit call with a reference
To change wording inside a picture, send it as an input_references image with a prompt giving the new text. Sume lists edit-capable models.
GPT Image resolution above 2560x1440: experimental, and Sume's cap
OpenAI marks GPT Image sizes above 2560x1440 as experimental. Sume's 2.5 custom pixels accept up to a 3840 edge, so check the five size rules before you send.
Edit a high-resolution photo with AI: edit first, then upscale
Sume image edits return at catalog tiers (512, 1K, 2K, 4K), not your photo's full megapixels. Edit with aspect_ratio auto, then enlarge with Image Upscale 1.0.
TTS locale vs language field: getting an en-GB accent
Cartesia's locale field picks a regional accent such as en-GB on Sonic 3.6. Sume's TTS request has a language field only, so pick the accent via the voice.
Text to speech reading a confirmation code digit by digit
To make TTS read a code one digit at a time on Sume, write the transcript with the digits separated, pin a Sonic id, and listen before you ship it.
Hinglish text to speech API: romanized Hindi with language hi
For romanized Hindi or Hinglish on Sume TTS, send language `hi` even though the text is Latin script, and pin a Sonic model id on the Router.
Odia and Urdu text to speech API: the language codes
Cartesia's Sonic 3.6 adds Odia (`or`) and Urdu (`ur`). On Sume, send the code in `language`, pin `sonic-3.6`, and check the voice before you pay.
TTS voice-language mismatch warning: confirm, then retry
Sume's TTS warns when a voice's language differs from `language`. No job or charge exists yet; after the user agrees, retry with `confirm_language_mismatch`.