Automation.
A delivery is hundreds of small jobs that nobody should do by hand. Final Frame runs them automatically as files arrive, and records the result of every step.
What we automate.
The deal
ProcessWhat it does
Deal memo readingReads the memo and proposes the full deliverables list
Delivery spec libraryKeeps named destination specs that deliverables inherit from
Spec from the annexeBuilds a new delivery spec from a memo's annexe
Partner identifiersStores the IDs each destination issues for a title
Bill of goodsGenerates the delivery manifest a destination ingests
Media operations
ProcessWhat it does
File transferMoves masters in and out over industry transfer pipes
TranscodesConverts masters to the demanded codec and container
Frame rate conversionConforms material between the standard frame rates
UpscalingTakes HD masters to UHD where a deal demands it
PackagingAssembles IMF and supplemental packages to the spec
DCP creationBuilds cinema packages from the graded master
Streaming proxiesMakes a browser-playable copy of every upload
Metadata and catalogue
ProcessWhat it does
Title matchingResolves titles against EIDR, IMDb, TMDB and Wikidata
Auto-completeFills facts, credits and certificates across the catalogue
Synopsis writingWrites synopses at every length, in every deal language
Deep searchResearches the records the databases could not answer
Genre mappingMaps sourced genres onto your taxonomy, per market
Metadata filesWrites the metadata deliverable in the demanded shape
Video
ProcessWhat it does
Technical probeReads codec, resolution, frame rate and colour off uploads
Spec comparisonJudges each measured attribute against the delivery spec
Audio and dubs
ProcessWhat it does
Technical probeReads codec, sample rate and channels off every upload
Audio descriptionCarries AD as its own deliverable beside dubs and subtitles
Streaming copyMakes a playable preview with the master left untouched
Loudness meteringMeasures loudness, true peak and range to EBU R128
Images and artwork
ProcessWhat it does
Template placeholdersCreates the destination's image set, sized from the spec
Artwork creationGenerates key art from the title's reference imagery
Artwork versioningSizes and formats each piece per destination and market
Spec comparisonJudges resolution, ratio and format against the template
Textless
ProcessWhat it does
Text removalTakes burnt-in text out of a shot and rebuilds the frames under it
Textless mastersBuilds a clean master from the texted one, shot by shot
Textless elementsCuts each affected shot as its own clip, in, out and running time
Shot finderFinds every shot carrying text and lays them out on a filmstrip
Title cardsRebuilds a title card as its plain ground rather than guessing at it
Rights and availability
ProcessWhat it does
Rights inRecords what you hold, per work, territory, language and window
Rights out and holdbacksTracks what is licensed away and what is held back from sale
AvailabilityWorks out what is still sellable, by territory and by window
Release windowsPlans and projects each window kind across a release
Storage and ingest
ProcessWhat it does
Your own storageConnects your S3, Google Cloud Storage or Azure Blob in four steps
Cloud explorerBrowses your buckets and pulls an object straight into a placeholder
Watch foldersBrings arrivals into the right project without anybody touching them
API, webhooks and MCPDrives all of it from your own systems, agents and CI
Subtitles and documents
ProcessWhat it does
Subtitle generationTranscribes the picture's sound and writes timed cues from it
Subtitle translationTranslates a cue list into any delivery language, timing kept
SDH and forced narrativeProduces SDH, standard or forced-narrative versions of a file
Cue previewShows timed text as a reviewable cue list in the browser
Cue approvalClears a checked cue so only what still needs a look remains
Filename conventionsRenames files to the project's convention in one click
Document filingFiles certificates and schedules against the right row
The QC checks.
Every check records a pass or a fail, with a note explaining what it found.
Video
CheckWhat it looks for
Decode integrityThe whole file plays cleanly through the decoder
Black framesBlack segments longer than two seconds
Frozen framesA held picture longer than four seconds
Scan typeInterlaced or progressive, checked against the spec
Audio silenceSilent stretches longer than five seconds
Integrated loudnessLoudness against the spec's target and tolerance
True peakPeak level against the spec's stated ceiling
Loudness rangeDynamic range against the spec's stated ceiling
PSE / flash riskCertified flash analysis via the Harding engine
HDR metadataDolby Vision and HDR10 metadata validation
Audio and dubs
CheckWhat it looks for
Decode integrityThe whole file plays cleanly through the decoder
SilenceSilent stretches longer than five seconds
Integrated loudnessLoudness against the spec's target and tolerance
True peakPeak level against the spec's stated ceiling
Loudness rangeDynamic range against the spec's stated ceiling
Channel mappingThe channel layout the spec demands for the mix
Subtitles
CheckWhat it looks for
File parsesThe file reads correctly in the demanded format
Timecode orderEvery cue starts after the one before it
Cue overlapNo cue starts before the last one has ended
Cue durationsEvery cue ends after the time it starts
Characters per lineLine lengths against the spec's stated limit
Lines on screenLines per cue against the spec's stated limit
Minimum durationCue length against the spec's minimum, in frames
Sync to pictureCue timing checked against the picture itself
Images and documents
CheckWhat it looks for
Decode integrityThe image file decodes cleanly and completely
Resolution and formatSize, ratio and format against the template
Document formatThe file matches the type its name claims
Where a destination requires a certified report, such as Harding analysis for flash risk, the licensed engine runs behind the same per-file report.
A note on models.
The catalogue work, from enrichment and translation to synopsis writing and artwork generation, runs on current frontier models. We choose the model per job, because a grounded web search and a Japanese synopsis are different tasks with different best tools. When a better model becomes available, the job moves to it without any change on your side.
Every model call is recorded, with the process that asked, the model that answered and the tokens it used, so you can always see exactly what ran and what it cost. And the output always lands as values a person can review, edit and approve before anything ships.
Every process on this page ends in a record of what ran, what it found and who approved it. That record travels with the delivery, so both sides of a deal can see the same facts.