Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
WireMock scenarios: fake a Sume job going queued to completed
Use one WireMock scenario per job so GET /v1/jobs/{id}/status answers queued, processing, then completed, and test the poll loop without spending.
xAI video API polling: pending, done, failed vs Sume job states
xAI's video API returns a request_id and three states. Here is how they map onto Sume's five /v1/videos statuses, with a runnable Python polling loop.
AI battle map for tabletop RPGs: square image, grid drawn in code
Generate a top-down battle map on Sume at $0.15 for a 2K square, then draw the grid in Pillow so squares are exact. Revise one room with a reference edit.
Why did that stock avatar rank first? Catalog search score weights
Sume's avatar catalog search scores handle 8, keywords 7, search card 6, display name 5. Read search_reasons and fix a query that ranks the wrong presenter.
Do [pause] tags and dashes in an avatar script cost extra seconds?
Sume Avatar 1.0 counts every space-separated token as a word, so [pause] tags and lone dashes add planned seconds. See the math and the silence-scene fix.
Claude Free plan has one custom connector: is Sume worth the slot?
Claude's Free plan allows one custom remote MCP connector. What Sume's hosted MCP gives you in that slot, why read-only fits, and when to upgrade.
Claude Research with a Sume connector: turn off write tools first
Anthropic advises disabling write-action tools when a custom connector is used with Research. For Sume that means read-only OAuth, no paid tools.
Continue: add Sume hosted MCP, agent mode only, timeout above 55 s
Continue runs MCP tools in agent mode only. Add Sume's hosted MCP as a streamable-http YAML file, send a key header, and raise requestOptions timeout past 55 s.
Copilot CLI with a local Ollama model: do Sume MCP tools still work?
Copilot CLI can now discover Ollama models, but a local model is not offline mode. What that means for Sume's remote MCP tools, spend gates and tool calling.
Eleven v4 seed and voice settings: what each Sume router id accepts
Which of stability, similarity_boost, style, speed and seed work on eleven-v4, v4-turbo, v3, multilingual-v2 and turbo-v2.5 in the Sume TTS Router, with ranges.
ElevenLabs voice ids in the Sume TTS Router: list them, avoid 400s
Eleven v4 and other eleven-* ids run through the Sume TTS Router. How to list voices with family=eleven, which id shape each family takes, and what gets a 400.
Image-to-video with sound by API: Sume ids that take a first frame
Kandinsky 6.0 has an image-to-audio-video mode. On Sume these ids take a first_frame image and return sound, and these do not. Request code included.
Is AI video 'Full HD' native or upscaled? Kandinsky 6.0 vs Sume rows
Kandinsky 6.0 renders 864x480 and adds Full HD with a plug-in. Sume's H3 Max 1080p is a latent refinement from 768p. What that means for price and detail.
Is Kandinsky 6.0 Video on Sume? No: query the catalog by need
Sume lists no Kandinsky 6.0 id. Map each Kandinsky feature to a Sume request field, then filter GET /v1/videos/models for sound and a 5 s duration.
Jira Rovo MCP plus Sume hosted MCP: turn a ticket into a release clip
Use Atlassian Rovo MCP and Sume hosted MCP in one session: read a Jira ticket, dry-run a clip, cap spend, and post the media.sume.com link back.
Kandinsky 6.0 5-second audio clip: price it on Sume rows
Kandinsky 6.0 makes 5-second clips with sound but lists no price. Here is what a 5-second clip with audio costs on the Sume rows that bill sound.
Kandinsky 6.0 lip-syncs in one pass; on Sume a talking face is Fabric
Kandinsky 6.0 Video makes speech and lip-sync inside the clip. Sume's docs send on-camera speech to Fabric or H3 Max Lip Sync with your audio, not a video id.
Kandinsky 6.0 Pro: 292 s a clip on an H100, 100 clips take 8.1 hours
Kandinsky's published timing is 292 seconds per 5-second Pro HD clip on an H100. That is 8.1 hours for 100 clips, set against hosted per-clip prices on Sume.
Is Kandinsky 6.0 Video on Sume? No, and the closest audio ids
Kandinsky 6.0 Video is open weights with no hosted API, and Sume lists no Kandinsky id. These Sume ids give 5-second clips with sound and a first frame.
Kandinsky 6.0 video with sound: is it on Sume, and what to use
Kandinsky 6.0 makes video and audio together, MIT-licensed. Sume does not list it. Here is what Sume lists for synced sound, and when to self-host Kandinsky.
Kandinsky 6.0 image-to-audio-video vs a first frame on Sume
Kandinsky 6.0 animates a still with synced audio. On Sume, frame_images starts a clip from your image, and some ids also take an audio reference. What fits.
Kandinsky 6.0 Video or a hosted API: a 5-second clip with sound
Kandinsky 6.0 Video is MIT open weights for 5 s clips with synced audio. Sume does not list it; two Sume ids make sound natively. Prices and limits.
Kandinsky 6.0 stops at 5 s: a 30 s ad as one job or six clips
Kandinsky 6.0 Video makes 5 s clips. A 30 s ad on Sume is one Wan 3.0 job at $3.75 or six clips stitched for $0.17 in assembly fees. What each route costs you.
Kandinsky 6.0 Lite takes 406 s on an RTX 5090: local or hosted?
The Kandinsky repo times one 5 s Full HD clip per GPU: 406 s for Lite on an RTX 5090. Clips per hour by card, and when a hosted Sume job is simpler.
Need a 7-second ad? Kandinsky's fixed 5 s vs Sume exact seconds
Kandinsky 6.0 makes 5-second clips, so a 7-second ad needs a join. On Sume, Omni 360p bills exactly 7 s for $0.2625. Rows, floors and the join cost.
Layered paper-cut art from a photo: FLUX.2 flex or pro on Sume
Photo to layered paper-cut art with a reference edit: FLUX.2 flex at $0.0625 or pro at $0.0375 on Sume, four takes for $0.25 or $0.15.
MCPJam auth debugger on Sume's MCP host: what each step returns
Point MCPJam's auth debugger at mcp.sume.com and read the results: S256-only PKCE, public clients, authorization_code only, a one-hour token, a resource check.
Neovim mcphub.nvim: Sume hosted MCP with autoApprove for read tools
mcphub.nvim reads remote MCP servers from servers.json. Add Sume with a Bearer header, autoApprove only read tools, and leave paid generation to confirm.
Photo to linocut print: Ideogram 4.5 edit and a 2-color threshold
A linocut look from a photo: one Ideogram 4.5 edit from $0.0375, then a Pillow threshold that makes it truly two-colour. Low, medium, high compared on Sume.
Photo to stained glass window: one Seedream 4.5 edit
Turn a photo into a stained glass window with one reference edit: Seedream 4.5 at $0.05 an image, an explicit ratio, and four takes for $0.20 on Sume.
Prism 2K audio-video needs 4+ 80 GB GPUs; Sume ids at 2K and 4K
Tencent Hunyuan's Prism does native 2K video-audio but wants four or more 80 GB GPUs. Which hosted Sume ids reach 2K or 4K, and what a 10-second clip costs.
Prism 2K video-audio needs 4+ GPUs: the hosted route up to 4K
Tencent Hunyuan's Prism makes 2K video with audio but needs a single 80 GB GPU at 720p and 4+ GPUs above. What Sume lists for 1080p and 4K sound clips instead.
Is my Sume avatar ready? Check resource_status and preview_video_url
Read a Sume Avatar 1.0 resource before your first talking video: resource_status, preview_image_url, nullable preview_video_url, voice.status, handle filter.
Remix cut refused with only a start time: the 120 s budget
Why a remix media cut with start but no end fails remix_cut_too_long, how the 120 second and 20 span limits add up, and when to use video trim instead.
Remix notes lint: FORMAT.md needs a Captions heading
What the remix media notes verb checks: a Captions heading in FORMAT.md, media.sume.com links only, the 200,000 byte cap, and why a lint failure still saves.
Remix tile: 96 cells per call, and how to count yours
A remix media tile call is refused above 96 cells with remix_tile_too_many_cells. How each selector is counted, worked examples, and ways to split a request.
Remix tile accepts a reference ingest id; other verbs do not
Use a finished reference ingest id as understanding_id for remix tile, including around without a transcript job. Boundaries and cut still need a probe.
What did one script_run cost? Sume script_runs and its children
Sume /v1/usage lists script_runs with rows and money per call, and counts script_run_children in includes. Read the cost of one script_run to the cent.
Prism audio tags vs Sume generate_audio and audio references
Prism's README describes music, sfx and speech tags. Sume gives a generate_audio flag and audio references on some ids. Which ids, and the limits.
Slack MCP draft message plus Sume: post a generated clip after review
Slack's MCP server can draft, schedule and send messages. Pair it with Sume's hosted MCP: generate the clip, post the durable media.sume.com link as a draft.
pending_usd_micros vs held: what Sume's settle sweeper still owns
Sume /v1/usage splits open holds into held and pending_usd_micros. Read the two fields and settle_state to tell parked rows from spend, and when final flips.
unpriced_usd_micros: legacy rows still inside Sume's debited total
unpriced_usd_micros in Sume /v1/usage is captured spend from legacy rows with no price-book stamp. It is inside debited, so never subtract or add it twice.
TanStack Start webhook route for Sume jobs: verify the raw body
Receive a signed Sume job webhook in a TanStack Start server route: read request.text(), verify with the SDK, answer 204, and dedupe on job_id.
Tilt-shift miniature effect: Nano Banana 2.1 edit or Pillow blur
Make a photo look like a tiny model town: a Nano Banana 2.1 edit with auto ratio from $0.10 on Sume, or a free Pillow gradient blur. Both compared.
UGC avatar scene prompt: why dim, glary lighting words are kept
Sume's first-frame step copies lighting words from your scene prompt as written, so ordinary light like TV glow or a side window stays. How to write it.
Which Sume media routes need a flag, and which are on prod?
Remix media, reference ingest and the trending-videos rebuild are flag-gated; video inspect, frames, trim, detach, filter and timeline are on dest and prod.
Word timestamps for an Eleven v4 voice: the router says 400, use STT
The Sume TTS Router rejects timestamps on eleven-* ids. Get word timings by running the finished narration through Sume STT for one cent a minute.
Zapier Next Gen Zaps: pause for a human reply before a Sume agent run
Next Gen Zaps can pause for a human reply. Use that pause to approve a Sume agent run, then send the approved generation_spend_cap_usd and an idempotency key.