Find the first stage that loses detail
Do not judge only an editor preview or a filename containing “1080p.” Locate the same moment in the source, edited picture master and local export. Compare in one player with the same window size and playback conditions. Inspect small text, edges and texture, then watch motion at normal speed. A still does not establish motion quality.
| First loss | What to check next |
|---|---|
| Source is already unclear | Find a usable original or obtain necessary coverage. Larger export dimensions cannot recover unrecorded detail. |
| Source is clear; edited master worsens | A low-resolution download or preview copy, enlarged crop, or upscaled source in a multi-video assembly. |
| Master is clear; local export worsens | Actual dimensions, scaling cap, encoding settings and repeated encoding of an earlier export. |
| Local export is clear; uploaded playback worsens | Processing status, selected playback quality and viewing conditions before remaking the local edit. |
If one file looks different between players, record player and display scaling too rather than immediately blaming export settings.
Separate output dimensions from picture detail
Aspect ratio describes proportions; resolution describes actual pixel dimensions. For fitting landscape and portrait material, see mixed orientations. This page traces detail loss after the picture has been chosen.
A small video surrounded by padding can live in a large frame while its subject still occupies few pixels. Enlarging a small crop likewise enlarges existing pixels. Inspect the visible region as well as frame dimensions.
The cited multi-source video-cut path fits segments proportionally into a shared canvas with padding. Canvas selection considers adopted sources’ orientation, area and duration. A low-resolution source can be enlarged within that canvas. This enables assembly; it neither promises equal detail nor performs automatic super-resolution.
Geometry probing also accounts for rotation and pixel aspect ratio. Check the displayed picture rather than copying stored dimensions alone.
Check the settings the current skill actually consumes
These settings describe only the pinned video-assemble composition path. Check other editors and packaging paths separately.
- OUTPUT_MAX_HEIGHT: A positive value caps output height while retaining proportions. Zero adds no downscale at this stage. The implementation shrinks; it does not upgrade low-resolution pictures.
- OUTPUT_CRF: This path’s x264 quality setting, default 18. A larger value generally reduces size with more loss. A lower value cannot restore detail already absent from the input. Eighteen is not a losslessness guarantee.
- OUTPUT_PRESET: Default veryfast. It participates in encoding speed and compression tradeoffs, not resolution selection.
Encoding settings apply when this stage re-encodes video. Burned subtitles, masking, visual overlays, packaging, scaling or forced re-encoding enter that path. Without these operations the video stream is copied, so changing CRF alone does not re-encode it. Sound follows the existing audio plan separately.
video-cut already applies its own encoding when building the edited master. A lower composition CRF cannot undo that loss. Rebuild affected pictures from a still-clear source instead of repeatedly treating an encoded export as the original. Foreground packaging and JianYing export require their own implementation and output checks; these three variables are not universal switches.
Original diagnostic plan: a fine thread in a binding tutorial
This is a fictional paper case with no files, probes or viewing findings. A creator edits a bookbinding tutorial. Essential coverage shows a needle emerging through a paper hole, thread tightening and a completed knot. Assume both source and edited picture master are 1920×1080 and their hole and thread details are readable. An old export used OUTPUT_MAX_HEIGHT=480, and the creator finds the needlework unclear. Subtitles, cuts and sound are approved.
Verify the assumptions: Open the corresponding files, locate piercing and tightening, probe actual dimensions and compare playback under the same conditions. If the master already loses detail, return to that stage. If the old export was not actually reduced to 480 pixels high, the setting is not an established cause.
If comparison confirms the first loss at that downscale, this creator chooses a local candidate retaining 1080-pixel height. Render from the clear picture master with OUTPUT_MAX_HEIGHT=0, OUTPUT_CRF=18 and OUTPUT_PRESET=veryfast in this composition path. Keep accepted subtitles, picture order and sound decisions; use a separate candidate workspace and output, retaining the old version. Removing the downscale does not establish readability; inspect the result.
Complete review: Compare source, master, old export and candidate through piercing, tightening and knot completion. The needle must emerge, thread tighten and knot finish. Do not substitute a still or delete a step to appear sharper. Ensure subtitles do not obstruct the needlework after sizing changes, sound remains complete and synchronized, and probe candidate dimensions.
If detail is readable, record the file and observations for the creator’s adoption. Otherwise investigate source detail, enlarged crops and encoding separately rather than claiming the settings fixed it. A platform’s smaller-file requirement calls for a separate delivery and readability comparison; removing the cap here is not a universal delivery specification.
Preserve accepted content while revising picture quality
Tell the AI which content is accepted: speech, subtitle text, cuts, action order, music and complete mix. Request diagnosis of the first loss and proposed input/output changes before rendering a picture candidate. “Make it clearer” does not authorize rewriting narration, changing a voice or recutting the story.
For an existing complete mix, follow keeping approved audio with a treatment compatible with the new picture rather than silently producing new sound. Deliver actual files, probe values, inspected passages and unresolved questions. Render success establishes file creation, not satisfactory viewing.
FAQ
Does exporting 720p as 4K restore detail?
Larger dimensions do not restore missing detail. Check available originals and the pixels occupied by the subject. The cited path does not perform super-resolution.
What if only YouTube playback looks blurry?
Separate platform processing from the local export. YouTube says initial processing provides lower quality, with higher qualities available after processing. Check status and playback options before comparing. See YouTube’s quality guidance; do not assign one fixed waiting time to every platform.
Do it with the skill
Use /video-recap ($video-recap in Codex) to review the supplied source, edited master, export and output settings. First identify evidence of the earliest detail loss and propose a revision. Preserve accepted speech, subtitle text, shot order and mix. For discussion only, run no editing, voice production or assembly. Execute a candidate through a supported path after its production is requested.
Read the method: Actual output size and encoding configuration · Re-encoding and video stream copying · Downscale-only height cap · Display geometry and multi-source canvas choice · Edited-master scaling and encoding