Guides
Sume guides: the API, Formats, agents, models, media tools and pricing, each answer dated and sourced.
Holiday send-by cues for Alaska, Hawaii, international and military
USPS 2026 send-by dates differ by region: Dec 16 for AK/HI/PR Ground Advantage, Dec 9 international, Nov 4 military. Burn each in as a $0.20 caption cue.
Meta catalog video: the ad spend per SKU that pays for the clip
Meta says catalog ad sets with product video saw 20% more conversions per dollar. Break-even spend per SKU is 5 x clip cost x CPA/margin; $0.93 and $2.19 clips.
One 1080x1920 master for Meta, TikTok, Pinterest and Shopify limits
One vertical MP4 can pass four platforms: Instagram Feed 1080x1920, TikTok 540x960 minimum, Pinterest 2 GB and 6-15 s, Shopify 1 GB. Limits read 2026-10-05.
Pinterest 6 vs 15 second product video: price for 100 SKUs at 720p
Pinterest recommends 6-15 s video. At 720p on Sume, 100 SKUs cost $75.00 vs $187.50 on Wan 3.0, or $226.80 vs $567.00 on Seedance 2. Full grid inside.
Put a promo code on screen in an AI product video with caption cues
Burn a Black Friday code into a silent clip with $0.20 caption cues, key the retry by SKU and code, then pull stills with video-frames to check the text.
10-SKU AI video ad pilot before Black Friday: priced line by line
A 10-product video ad test costs $22.74 on Sume: image edit, 5 s Seedance 2 clip, $0.20 captions and a $0.10 Timeline end card, plus $3.78 for two retakes.
Which SKUs get an AI video first? Rank by profit inside a fixed budget
Rank SKUs by 30-day gross margin and fill a fixed video budget at Sume prices: $0.625 Wan or $1.89 Seedance 2 clips plus $0.30 for captions and Timeline.
Bulk text to speech from a CSV: one job per row, safe retries
Voice 500 CSV rows with Sume TTS 1.0. One idempotency key per row id, retry-after on 429, and the queue-capacity numbers per plan decide how fast it goes.
Match STT results to file ids: the job echoes your Idempotency-Key
Sending 300 recordings to Sume STT? Set Idempotency-Key to your own file id. The job record returns it, so results map back after a crash or a retry.
Route low-confidence transcripts to review: STT language_probability
Sume STT returns language_code and language_probability. Flag results under a threshold you set and send them to a person. Python, about 10 cents per file.
Search a podcast archive for a phrase and jump to the time: STT words
Transcribe episodes with Sume STT, keep words[] with start times in a small index, and find any phrase with a mm:ss link. Python, about 60 cents per hour.
Predis Core: 325 images or 33 videos vs per-image pricing on Sume
Predis Core lists 1,300 credits, about 325 images or 33 videos. Divide the plan price and compare to Sume's per-image estimate of $0.2225.
Price per finished video: Predis, Submagic, Zebracat vs Sume
Dividing plan prices by listed video counts gives $0.69 to $2.60 per video. A Sume captioned 30-second render is $0.32; an avatar ad is $5.52.
Test a video poll loop with unittest and a local server, no spend
A 30-line stdlib file tests a Sume video poll loop against a fake server: pending, in_progress, completed in order, and a failed job that stops with no sleep.
A zsh function: submit a video prompt, keep your terminal free
A 19-line .zshrc function sends a prompt to Sume Wan 3.0 (2 s, 480p, $0.125), returns at once and saves the MP4 later. Do not name a variable status.
A 480p proof wall for 100 SKUs: when it beats going straight to 720p
Proofing 100 SKUs at 480p costs $43.94 on Seedance 2-mini or $87.89 on Seedance 2, against $189.00 for 720p finals. It pays only if enough SKUs get dropped.
Release a 50-SKU holiday batch in two bulk queues: 5 first, 45 after
Submit 5 SKUs as a pilot queue, review the clips, then queue the other 45. Sume bulk queues take 1-100 items at concurrency 1-16. Runnable payload builder.
Webhook receiver that queues finished SKU videos for human approval
Verify the format.run.terminal signature, dedupe on run_id, and park each finished SKU video as pending review before anything is published. Runnable Python.
Annual vs monthly video tool plans: months before the year wins
Zebracat, Predis, Biteable and Submagic quote lower annual prices. The break-even is 6 to 9 months of use, and a prepaid wallet has no term at all.
Do Sume wallet credits expire? 365 days per top-up lot
Each Sume top-up becomes its own credit lot with a 365-day expiry. How to size a top-up so nothing lapses, and how that differs from monthly credits.
Faceless channel, 16 videos a month: the Sume wallet budget
Four 5-minute faceless videos a week cost about $5.29 each on Sume, so $85 a month. How to size top-ups and a retry reserve against 365-day credit lots.
Tcl http package: submit a Gemini Omni video job and follow the 302
A 22-line Tcl script using the http, json and tls packages submits a 3 s 360p Gemini Omni job on Sume ($0.1125), polls it and follows the content redirect.
Does YouTube's AI label hurt reach or monetization? What it says
YouTube states a disclosure label alone does not change recommendations or monetization eligibility. What it covers, what it does not, and a workflow.
YouTube AI labels you cannot appeal: Veo and C2PA metadata
YouTube says labels from its own AI tools or full-AI C2PA metadata cannot be appealed, though you can update disclosure status in Studio. What it means for you.
YouTube's AI label for unrealistic content: the expanded description
YouTube puts a label on realistic AI content, but unrealistic AI content is disclosed in the expanded description. The realistic test and the API field.
Find each script line in a recorded read: match STT words to script
Transcribe a studio read with Sume STT, then match its words to your script lines to get start and end seconds per line for cutting takes. Python.
Podcast quote clips end abruptly: STT boundary_lead_ms tail padding
Sume STT segmentation boundary_lead_ms (0 to 500 ms, default 70) sets how long a sentence's tail runs before the cut. Tune it, then cut with timeline audio.
Redact names from a transcript and the audio: STT words, then split
Sume STT returns word timings with a type field. Find the names you list, blank them in text, and cut the same spans from the audio for $0.01.
Fit a voiceover to a 30-second ad: measure pace, then set TTS speed
Send timestamps.words, read the last word end time, and rescale generation_config.speed (0.6 to 1.5) until a Sume TTS take fits its slot. Two takes, 2 cents.
Text to speech API with emotion: audition four values, keep the winner
Sume TTS 1.0 takes a free-text emotion up to 64 characters in generation_config. Run a four-take audition for 4 cents and store the winner.
What to log from a TTS job result so you can recreate a voiceover
A finished Sume TTS 1.0 job echoes model_id, voice, language, output_format and generation_config. Save them with the audio URL to rebuild the same take.
How to mark a Vimeo video as AI-generated: the steps for a Sume clip
Vimeo's AI label is a per-video checkbox in Settings, Content. The steps, the three cases that need it, and what to log for a Sume-generated clip.
Vimeo's AI label vanishes in the embed when the title is hidden
Vimeo shows the AI label under the title and in the embedded player, but not when the title is hidden. What to do on a landing page that embeds an AI clip.
Vimeo labels its own AI tools' output itself: what you control
Vimeo may apply an AI label you cannot remove when its own tools made the video. What that means for a clip you generate elsewhere, such as on Sume.
MCP timeline_create provider_fields_not_accepted: no ffmpeg fields
Sume's timeline_create refuses model, filtergraph, ffmpeg_args, codec, preset and crf with provider_fields_not_accepted. Describe the edit, not the encoder.
MCP unsupported_payload_keys: read the allowed array
A key a Sume MCP tool does not list returns unsupported_payload_keys with unexpected, allowed, tool and sometimes path. Fix it from the allowed list.
MCP unsupported arguments: move the field into payload
A Sume MCP call with quality beside payload fails with unsupported_tool_arguments. Read hint move_to_payload and the adjustments array, then retry once.
Python: price a Seedance 2.5 clip from seconds and resolution
A 17-line function turns resolution and seconds into video tokens and a Sume price: 4 s at 480p is $1.07, 30 s at 1080p is $42.65.
AppleScript: submit a Wan 3.0 video job and save it to the Desktop
A 22-line AppleScript that runs curl and jq through do shell script, polls a Sume video job and saves the MP4 to the Desktop. A 2 s 480p Wan 3.0 clip is $0.125.
macOS launchd: one AI video a day with a per-date Idempotency-Key
A launchd agent that wakes late can run twice. A per-date Idempotency-Key and an existing-file check keep a daily 3 s Gemini Omni clip ($0.1125) to one charge.
avatar-image-to-video_create: audio_url and duration_seconds
avatar-image-to-video_create needs a Sume-hosted audio_url and duration_seconds (1 to 300) matching the audio. Missing either gives missing_payload_field.
MCP conflicting_payload_fields: image_url or avatar_id, not both
Sume's avatar-image-to-video_create and kling-motion-control_create take image_url XOR avatar_id or avatar_handle. Both returns conflicting_payload_fields.
Check that a recorded ad says the required disclaimer: STT words
Transcribe an ad with Sume STT, match the disclaimer in words[], and fail the check if it is missing or starts in the last seconds. Python, 1 cent per minute.
Does the ad voice start in the first second? STT first-word check
Run recorded ads through Sume STT and read words[0].start. Flag any clip where speech starts after 1.0 s. A 20-ad batch is about 10 cents of usage.
Kling 3.0 720p or 1080p: the same price per second on Sume
Sume bills kling-3 by the second and audio setting only, so 1080p costs the same as 720p: $0.14 a second audio off, $0.21 on. A 10 s clip is $1.40 or $2.10.
Kling 3.0 audio is on by default and adds 50% to the price
On Sume, kling-3 treats a missing generate_audio as true: $0.21 a second, not $0.14. A 10 s clip is $2.10, or $1.40 with generate_audio false.
kling-motion-control_create: motion_video_url, 1 to 30 s, no model
kling-motion-control_create needs a public HTTPS motion_video_url and duration_seconds of 1 to 30, and refuses a model field with provider_fields_not_accepted.
MCP wrong_tool from generate_video: lip sync and motion control
Sume's generate_video refuses minimax/h3-max/lip-sync and the Kling 3.0 motion-control model with wrong_tool. next_action names the tool to call instead.