Use cases
Worked examples on Sume's catalog Formats and endpoints: product video and photography, try-on, app and product ads, logo animation, and more.
Start with Ready-made Formats for product video: the Sume Format catalog
Amazon "up to 50% off" needs 10% of ASINs at 50%: a caption check
Amazon says an up-to claim needs 10% of products at the maximum discount. A short Python check counts the ASINs and burns the matching caption cue.
Walmart holiday creative survey: price first, then burn the price line
Walmart Connect reports shoppers ranked price the top holiday ad message in food, electronics and toys (2023 survey). Re-burn a price line at $0.20 a clip.
Reel speed 0.5x to 3x: burn captions after the final speed change
Instagram lets creators change Reel speed from 0.5x to 3x. Burned-in captions speed up with the clip, so a 1.5 s line is 0.5 s at 3x. Burn them last.
Shorts 2x playback speed: captions flash by, so set phrasing max_words
YouTube's June 2026 Shorts update lists 2x playback speed. At 2x a caption phrase is on screen half as long; limit max_words and burn captions with Sume.
Festive caption colors for holiday ads: slam style design overrides
TikTok's holiday guide asks for bold captions and festive overlays at peak. One caption job with design colors burns a red-and-green slam line for $0.20.
TikTok holiday budget split 20-25/60-65/10-15 and what 30 clips cost
TikTok suggests 20-25% early, 60-65% peak, 10-15% January. On a $10,000 plan, 30 captioned Omni clips at $1.20 each are $36, or 1.8% of the early phase at most.
160x600 skyscraper ad image: 3.75:1 is past GPT Image 2.5's 3:1 cap
A 160x600 ad is 1:3.75, taller than the 3:1 limit on GPT Image 2.5 custom sizes. Render 576x1728 (1:3) and crop 20% of the width in Pillow. Math included.
60-second AI video ad from 10-second shots: hero and filler cost
Six 10-second shots, two on Seedance 2.5 at 1080p and four on Wan 3.0 at 480p, make a 60-second spot for $30.96 on Sume. All six on 2.5 would be $85.32.
728x90 leaderboard ad image: GPT Image 2.5 can't, Nano Banana 2 can
A 728x90 ad is 8.09:1. GPT Image 2.5 caps custom sizes at 3:1, but the Sume catalog lists 8:1 for google/nano-banana-2. Check the row, request 8:1, crop 1.1%.
970x250 billboard ad image: render 1920x640, crop 23% of height
A 970x250 ad is 3.88:1, wider than the 3:1 limit on GPT Image 2.5. Render 1920x640 (3:1), scale to 970 wide, and crop the height. Table of checks and prices.
A4 poster at 300 dpi: 2480x3508 is over the GPT Image 2.5 pixel cap
A4 at 300 dpi is 2480x3508, 8,699,840 pixels, above the 8,294,400 cap. Render 2400x3392 on GPT Image 2.5, then scale up 3.4% in Pillow and tag it 300 dpi.
Brand name misheard in YouTube auto captions: burn your own script
YouTube warns that auto captions can misread accents and noisy audio. For a brand name that must be right, send script_text to Sume and burn your own wording.
Change only one thing in an AI image edit: prompt template and mask
How to edit one detail of an image without the rest drifting: a keep-and-change prompt template for GPT Image 2.5, then a mask_url fallback. Code and prices.
Check an AI thumbnail stays readable at 320x180 with a Pillow sheet
Generate a 1280x720 thumbnail with GPT Image 2.5, then shrink it to 640x360, 320x180 and 160x90 side by side to catch tiny text and weak contrast.
Child's drawing to a picture-book page with GPT Image 2.5
Turn a child's drawing into a polished picture-book page: host the scan at a public HTTPS URL, pass it as a reference, and ask for 4:3. Code and Sume prices.
Custom Shorts thumbnail needs YPP and a computer: export a frame
YouTube's help page says Shorts custom thumbnails upload in Studio on a computer. Export a frame with Sume's video frames route, and check size caps first.
Cut a quote clip from a long video: transcript, then video_trim
Find the sentence in a Sume transcript, then cut it with video_trim. A Python script, with timing padding, $0.02 per trim, and the 900-second output cap.
300x250 display ad image: generate 1440x1200, then resize by 4.8
A 300x250 ad is 6:5, so GPT Image 2.5 can render it at 1440x1200 (legal custom size) and Pillow scales it down by exactly 4.8. Size checks and price included.
Email header image 600x200: render 1920x640 at exactly 3:1
A 600x200 email header is 3:1, the widest shape GPT Image 2.5 accepts. Render 1920x640, then export 1200x400 for retina and 600x200 for 1x with Pillow.
EU deepfake audio label: repeat the disclaimer in a long voiceover
The EU Code lists an audible disclaimer for audio-only deepfakes, repeated in longer content. Join disclaimer clips into a voiceover with Sume timeline audio.
EU deepfake label on a recorded video: disclaimer and end credits
The EU Code of Practice lists an opening disclaimer, persistent icon and end credits for recorded video. Burn the text onto a Sume avatar clip with cues.
Floor-plan sketch to a furnished room render on GPT Image 2.5
Turn a floor-plan sketch into a furnished room render with GPT Image 2.5 at 16:9. The model does not measure, so name sizes in the prompt. Code and prices.
Google video enhancements add English-only voiceover: opt out or ship
Google video enhancements can flip, shorten and add English-only AI voiceover to your clip. Opt out per campaign, or ship a finished cut.
300x600 half-page ad image: render 720x1440, keep a 600x1200 copy
A 300x600 half-page ad is 1:2. GPT Image 2.5 can render 720x1440 in one call; scale by 2.4 for 1x or by 1.2 for a 600x1200 retina file. Checks and prices.
Instagram voice translation covers nine languages; add a tenth by cues
Instagram's voice translation lists nine languages: English, Hindi, Spanish, Portuguese and five Indian ones. For other markets, burn your own caption cues.
Is a personal AI avatar video exempt from EU deepfake labels?
The EU Code exempts private personal use and asks only minimal disclosure for artistic work, per Tech Policy Press. Business avatar videos still need a label.
Logo sketch to a transparent PNG with GPT Image 2.5
Turn a hand-drawn logo sketch into a transparent PNG: send the sketch as a reference, set background transparent and output_format png, then verify alpha.
Voice drift in long narration: Nova 2 Sonic's 52% vs Sume chunks
Speaker drift is what splits a long voiceover into jobs. Amazon claims -52% on an internal set; here is how to measure drift across Sume TTS chunks yourself.
Napkin sketch to product render: one reference on GPT Image 2.5
Turn a photographed napkin sketch into a product render with one input reference on GPT Image 2.5. Set aspect_ratio auto to keep the shape. Code and prices.
Pinterest video ad aspect check: which Sume ratios fit 1:2 to 1.91:1
Pinterest says avoid video narrower than 1:2 or wider than 1.91:1. Of six Seedance ratios, five are allowed and two (1:1, 9:16) are recommended.
Shopify product video limits: 10 minutes, 1 GB, 4K, and a 15 s clip
Shopify's product media page allows video up to 10 minutes, 1 GB, and 4096x2160 in mp4, mov, or webm. A 15-second 1080x1920 clip uses a sliver of each limit.
Shorts series episode one: check four opening frames in two seconds
A Shorts series has a watch series button that leads into a series feed, so episode one is the entry. Check four frames, 0 to 2 seconds, with video inspect.
TikTok ad text allows no links, @ or hashtags: a Python lint
TikTok's in-feed spec says ad text shows in white, uncustomizable, with no clickable links, @ symbols, or hashtags. A short check catches them before upload.
TikTok AI Outline splits a video in six: plan six slots in Sume first
TikTok's AI Outline breaks a video into six parts. Mirror that outline in Sume with six timeline slots and the unbilled plan call, then render once.
TikTok non-Spark ad file check: 500 MB, 10 minutes, 516 kbps floor
TikTok's non-Spark spec: 9:16 at 540x960 or more, up to 10 minutes, 500 MB or less, 516 kbps minimum. A 15-second 1080p cut sits far inside the limits.
US Letter flyer at 300 dpi: 2550x3300 exceeds the GPT Image 2.5 cap
US Letter at 300 dpi is 2550x3300, 8,415,000 pixels, over the 8,294,400 cap. Render 2400x3104 with GPT Image 2.5, then fit it to 2550x3300 in Pillow.
YouTube Shorts count replays as views: render a looping 15-second clip
YouTube counts a Shorts view when a Short starts or replays. Build a 15-second loop from repeated Timeline 1.0 slots for $0.10; match last frame to first.
Amazon's AI-people tag: which images are exempt
Amazon wants contains-synthetic-performer on photorealistic AI people. Real people (even AI-altered), characters, no-people and non-photoreal images are exempt.
TikTok AI disclosure: toggle or on-screen note for a Sume clip
TikTok Shop accepts two disclosure methods: the AI-generated content toggle or an on-screen note. Sume can burn the note in; you set the toggle in TikTok.
TikTok Shop allows AI dubbing: localize a product video
TikTok Shop's AI policy lists AI-assisted translation and dubbing as permitted. Disclose synthetic voices, and note LIVEs ban AI voices. Batch per language.
Prove an AI clip still shows the right color and size
TikTok Shop bans AI that alters a product's size, color, material or features. Pull 1-24 stills with Sume video frames and compare them to the listing.
TikTok Shop campaigns lock SKUs: render variant clips first
TikTok's campaign FAQ says variations and SKUs cannot be added or removed while a product is registered. Freeze the SKU list, then render one clip per variant.
TikTok Shop campaign review takes 5-14 business days: date math
TikTok's campaign FAQ says review takes 5-14 business days. To hear back by Fri Nov 27, 2026, apply by Nov 9 at the latest; apply earlier to be safe.
TikTok Shop campaign score: 3.5 to enroll, 3.0 to stay, QA gate
TikTok Shop campaigns need a 3.5 Shop Performance Score to enroll and 3.0 to keep benefits. Add a pre-publish check that extracts stills from each AI clip.
TikTok Shop: "digital renderings are prohibited" and AI images
TikTok Shop's listing-quality page says digital renderings are prohibited and the main image must exceed 600x600. Keep generated art off the hero slot.
YouTube can add the AI label itself, even if you skip the toggle
YouTube may auto-apply AI labels for its own GenAI tools, C2PA metadata, or content its systems detect. Sume docs promise no C2PA, so disclose yourself.
YouTube AI disclosure won't limit reach or monetization
YouTube says disclosing AI content won't limit a video's audience or eligibility to earn money. Penalties are for repeated non-disclosure. Tell clients.
AI B-roll on YouTube: when real places and events need the label
YouTube asks for disclosure when AI alters footage of a real event or place, or makes a realistic scene that did not occur. Four triggers, applied to B-roll.