Finishing & Delivery Checklist

Platform specs change. The numbers below are common targets as of October 2026. Always check the current spec sheet for the platform or festival you're delivering to.

1. Picture lock #

• Story, pacing and runtime approved (by you, your team or a trusted viewer)
• Runtime is within the brief or festival limit, including credits
• Aspect ratio decided and consistent (don't crop after the fact)
• Frame rate consistent across all clips (conform mixed-fps generations before the final edit)
• All placeholder shots replaced
• Lock the cut. No more timing changes once sound and color start.

Mixed frame rates: AI clips often arrive at 16, 24, 25, 30 or 32 fps. Choose one timeline rate (24 fps for a filmic look) and conform each clip deliberately: retime, interpolate carefully, or accept a cadence change. Don't leave it to the NLE's default.

2. Sound #

Dialogue #

• Every line is intelligible on phone speakers
• Noise, clicks and reverb cleaned (dialogue repair tools)
• Consistent room tone under dialogue scenes (no dead-silent gaps)
• Lip sync checked at 50% speed on every dialogue shot

Sound design & music #

• Ambience bed in every scene (no "vacuum" silence unless intentional)
• Key actions have sound: footsteps, doors, impacts (Foley-level detail sells realism)
• Music edited on beats and bars, with no abrupt cuts mid-phrase
• Music ducks under dialogue
• J- and L-cuts smooth the scene transitions

Mix & loudness #

• Balanced mix: dialogue first, then effects, then music
• Checked on headphones, laptop speakers and phone
• Loudness measured with a LUFS meter and set to the target:

• Stereo checked in mono (no phase cancellation)
• Optional: stems exported (dialogue / music / effects) for future versions and localization

3. Color #

• Correct first: exposure and white balance balanced shot to shot
• Match: shots within each scene match. This matters most when mixing models and sessions.
• Grade: the creative look applied consistently
• Skin tones natural (check the vectorscope skin-tone line)
• No crushed blacks or clipped highlights unless intended
• No banding in skies or gradients (work in 10-bit, add fine grain or dither)
• Grain or texture applied after the grade and upscale, consistently across shots
• Viewed on a calibrated or reference display if possible, and also on a phone
• HDR: only if you can grade and monitor HDR properly. Otherwise deliver SDR (Rec.709).

4. Quality control #

Use the review passes in the AI Artifact Field Guide §8: faces, hands and bodies, continuity, edges, motion, text, sound, full-screen.

Technical QC

• No black or flash frames between cuts
• No duplicate frames or stutters at clip joins
• No unrendered or offline media
• No watermarks from free tiers or tools
• No accidental real-world logos, brands or text
• Titles and captions inside the title-safe area (and not covered by the platform UI on vertical video)
• Audio and video in sync from first frame to last
• Head and tail: about 0.5–1s of black or silence at start and end, unless the platform or festival says otherwise

Watch the final export file itself, start to finish, on the device your audience will use. Encoding can introduce problems that weren't in the timeline.

5. Upscaling, interpolation & finishing passes #

Order matters. A typical sequence:

1. Cleanup and fixes (inpaint, paint, deflicker)
2. Upscale (precise mode for faces and key details)
3. Interpolation (only where needed, slow shots only)
4. Final grade trim
5. Grain, texture and halation
6. Titles and graphics
7. Export

• Compared upscaled vs. original at 100% for hallucinated detail
• Interpolation hasn't caused warping on fast motion or across cuts

6. Captions & accessibility #

• Captions or subtitles created (auto-transcribe, then correct by hand)
• Timing checked: each caption readable, roughly 1–7 seconds on screen, no more than 2 lines
• SDH (subtitles for the deaf and hard of hearing) includes speaker IDs and key sounds, e.g. [door slams], [ominous music]
• Exported as a sidecar file (SRT or VTT) and/or burned in (burned-in captions help on social, where many people watch muted)
• Optional: audio description for important visual-only story points
• Optional: translated subtitles or AI dubbing (with human review)

7. Credits & disclosure #

• Title card
• End credits: roles (use names from the Roles Glossary), collaborators, voice actors, composers
• Asset credits where licenses require attribution (CC-BY music and SFX, LoRAs and workflows if the author asks)
• AI disclosure line or card, e.g. "Created with [tools]. AI-generated imagery; script, edit and sound design by [name]." (AiFC asks for it in your submission write-up; on screen is recommended too.)
• Copyright line for your human-authored contributions, e.g. "© 2026 [Name]"
• Content Credentials (C2PA) kept on the export if your tools support it

8. Export settings #

Masters (keep these) #

Delivery files (common starting points) #

Bitrate tip: for uploads, use a high bitrate (e.g. 20–50 Mbps for 1080p, more for 4K) so the platform's re-encode starts from a clean source. Grain and fine detail need more bitrate.

9. Submission package (AiFC challenges) #

• Final film (spec above)
• Process breakdown (use the Process Breakdown Template)
• Any craft artifact the brief asks for (lighting diagram, blocking sketch, character sheet)
• 3+ BTS stills or a screen recording
• Prompt and settings log for the key shots
• AI disclosure and rights confirmation
• Reflection (≤100 words)

10. Archive #

• Project files, masters, stems, captions and the generation log in one folder
• Workflow JSONs and model or LoRA versions recorded (models change; your record shouldn't)
• License and consent documents saved
• Two backups, one off-site or in the cloud