thumbnail-production

Use for a standalone thumbnail request or the packaging stage of a film. Deliver the requested images and editable composition with a private brief and asset record. A thumbnail-only request does not require loading film-production instructions or rendering a new film. Follow the project's privacy and design rules.

Research and brief

Read the approved content brief and destination. For a video, also read its script or finished cut, title and opening payoff. Record the truthful promise and which visible evidence supports it. A dramatic image must not imply a result, endorsement, screen state or personal experience the film does not establish.

If research is requested, inspect relevant thumbnails, not just their view counts. Record URLs, observation date, comparable publication ages and channel baselines where available. Missing analytics leave performance unverified. Borrow a composition idea in original form; public thumbnails are not cleared assets. Use available public sources or an already authorized research tool. Do not require a new subscription, MCP connection or scraping service.

Write a compact brief in the existing private production area:

Field	Required content
Promise	Title, opening payoff, source evidence and destination
Concepts	Distinct visual ideas, each with its focal subject, composition, exact short text or no text, and intended crop
References	Source links and what each informed; distinguish observed design from inferred performance
Assets	Cleared photos/stills, source IDs, permitted edits and any remote-generation destination
Delivery	Requested concepts/variants, formats, sizes, editable source, generation allowance and selection criterion

Make a few meaningfully different concepts when comparison is requested; color swaps alone may not test an idea. Do not turn three variants per concept into a mandatory paid batch. Reuse settled choices and authority. If concept selection, spend or photo transfer still needs a decision, finish the reviewable brief before asking. Otherwise continue production.

Photos and face references

Prefer a suitable cleared photograph or film still when it serves the idea. A face is optional. For an existing reference collection, select only relevant angles and expressions; dozens of photos do not guarantee identity accuracy. A crop or composite can preserve the actual face without asking a model to recreate it. For generative edits, verify likeness against the input.

Use the supplied or project-authorized images, following the consuming project's asset-source instructions. Preserve originals and edit copies. Keep private source collections outside shipped assets and version control. Permission to use a photo is separate from permission to send it to a named remote image service; reuse explicit transfer authority when supplied.

Record a private source ID and permitted use for each selected image. Check other people's consent, visible documents/screens and location clues. Inspect final derivatives for embedded location and personal metadata; remove it from delivery copies and recheck, preserving the original. Do not publish face-bank files, library paths or identifying source metadata.

Produce and inspect

Use the project's existing composition workflow for photographic layouts and exact type. For requested AI generation or image edits, use the available image-generation skill/tool under its instructions and authorized asset/provider scope. Keep text editable when generated lettering fails. Save the actual prompts, selected inputs and composition privately. Do not install the source author's studio or pick their model by default. If no suitable tool is available, deliver the brief and name the blocked image output explicitly.

Produce separate platform exports from the selected composition. Verify current requirements and the actual upload surface before export:

• YouTube long-form: use a 16:9 composition; at least 1280 by 720 keeps a comparison set above the documented experiment downscaling threshold. Check current file limits. Where eligible, Studio can compare up to three choices concurrently. For a thumbnail-only test, hold the title fixed. Shorts use a separate thumbnail path and do not support this test.
• LinkedIn: distinguish a native video thumbnail from a post image or linked-page preview. Desktop native video upload has a thumbnail control; check the actual account/composer. Match the intended video's shape and inspect feed cropping. A YouTube export is not proof of LinkedIn fit, and a cover is not the same as the opening frame during autoplay.

Inspect each export at full size and a small feed preview. Check readable type, focal subject, likeness, hands/edges, crop, duration-badge overlap, claim accuracy and private details. Probe actual format, dimensions and byte size; a filename is not proof. Preserve editable source and known good exports. Deliver a preview/contact sheet, exact artifact paths, the selected concept, asset provenance and unrun platform checks. No upload follows from image production alone.

Learn from the release

When authorized results exist, record the tested assets, title, dates, eligibility, sample limits and the platform's verdict. YouTube uses watch time rather than CTR alone; inconclusive and same-performance outcomes are valid. High views on another creator's video do not isolate the thumbnail's effect. Keep LinkedIn results separate and do not claim equivalent native thumbnail testing. Feed one supported observation into the next brief, or report insufficient evidence. A source review or attractive export does not prove growth.

Sources

Original procedure informed by Noah Vincent's article, October 26, 2026; public article/store descriptions only, reuse rights unknown, nothing copied. Checked October 26, 2026 against YouTube testing, LinkedIn native video. Recheck platform limits and controls before delivery. No comparative generation or growth trial.