Fine-tune frames in Advanced Editor

Open the current editor

From a ready generated or imported video in Quick Mode, choose Fine-tune animation. The editor is designed for desktop widths of at least 1024 px; on smaller screens, use Quick Mode or a wider window. Editor controls follow your selected interface language; server-provided error details may still appear in English.

Your source video remains unchanged. The saved composition is an editing recipe that references source frames, not a newly generated video. Reopening a source can reuse its existing default composition; check the current sequence before replacing it.

Understand the five panels

The source and sequence playheads are independent. Selecting a frame is not always the same as moving a playhead. Watch the relevant preview when checking an edit.

Build a sequence

1. Click a source thumbnail; Shift-click extends a range and Ctrl/⌘-click toggles individual selections.
2. Use Insert after playhead, Append at the end, or drag source frames into the sequence.
3. Select output frames to move, duplicate, or remove them. Group selections preserve their order when moved.
4. Duplicate repeats each selected item beside itself; it can make a pose hold longer.
5. Reverse reverses the entire sequence, not only the current selection.
6. Use the Undo/Redo buttons to correct recent edits in this editing session.

The current limit is 240 expanded output frames, including repetitions. The UI does not expose a separate per-frame duration editor. Duplicate frames to make holds, and inspect the resulting sequence length.

Control timing and check a concrete example

Output FPS is adjustable from 1 to 60. At a fixed FPS, duration is expanded frame count divided by FPS. For example, 24 output frames at 12 FPS play for two seconds. Duplicating three of those frames yields 27 frames and 2.25 seconds; changing to 24 FPS then yields 1.125 seconds.

Changing FPS does not synthesize intermediate motion. Loop playback repeats the sequence but does not repair the last-to-first seam. Check feet, silhouette, and apparent speed at that boundary.

Source-frame timing comes from the video's decoded presentation timestamps. Do not interpret a source-frame number as an output-frame number after rearranging or repeating frames.

Adjust a frame or replace its image

Whole animation changes the global canvas, placement, scale, and pivot. Switch to Selected frames, then select output items in Animation Sequence to adjust their X/Y offsets and scale, or drag the picture in Animation Preview. Other occurrences of the same source frame stay unchanged. Multi-selection applies relative changes while preserving existing differences.

Reset transform resets selected transforms without removing replacement images. Enable Onion skin while paused to overlay neighboring output frames for alignment checks; these overlays are not exported.

With one item selected, Frame image → Export PNG downloads its current accepted image (the source frame if no replacement has been accepted). It saves pending edits first and uses the original source dimensions with transparency preserved. Position, scale, output-canvas settings, and onion skin are not baked into this file, so reimporting does not apply transforms twice. The filename includes the selected item's saved sequence number, for example animgen-frame-0012.png.

Edit that PNG in your own tool, keeping the canvas size, then select the same item and choose Import image. Import accepts a single PNG/WebP/JPEG up to 20 MB; other sizes retain their aspect ratio and fit the source canvas without automatic cropping. Compare using Show before, then choose Accept image to save. Discard preview only dismisses the preview. Revisit saved candidates through Saved image versions, or use Restore source to return to the source video frame. Both single-frame and animation export are disabled while a candidate awaits review. Single-frame export does not generate an AI image or charge generation credits.

The ? buttons beside Selected frame and Frame image contain the detailed adjustment and import/export tips. Click or use Enter/Space to open them; Escape or clicking outside closes them. Errors, processing progress, and billing notices remain visible in the inspector.

When enabled for the deployment, AI repair lets you draw a rectangle and describe a local correction. Review the estimated credits and maximum reservation before generating. The final charge uses actual token usage at the quoted rates and cannot exceed that maximum; unused credits are released. If usage is incomplete, the candidate stays usable while billing is checked. Unverified charges are waived after the reconciliation deadline (normally 24 hours). A failed repair releases its reservation.

The current frame and available neighboring/character references are sent to OpenAI. A successful candidate still needs manual acceptance; discarding it does not refund generation credits. The server preserves pixels outside the repair region of its prepared input, but animation continuity is not guaranteed. Check adjacent frames and the loop seam. Missing server configuration disables AI repair without disabling manual editing or image import.

Save before leaving or exporting

Changes are automatically saved after a short idle period. Check the status: Unsaved → Saving… → Saved. Save failed is not a successful save. Keep the page open and preserve a record of important edits while resolving the error.

Use one editing tab per composition to avoid revision conflicts. The editor's Back and Export actions attempt to flush pending changes; abruptly closing a tab or losing connectivity is not a guaranteed save path. Undo history is local to the editing session, not permanent server version history.

Export uses a saved composition revision and a snapshot of its recipe. Review the displayed frame count, FPS, canvas, formats, and quote before confirming. Later changes do not rewrite an export already submitted.

Current boundaries

The editor provides global and per-frame transforms, image replacement, and rectangular local AI repair when enabled for the deployment. It does not offer transform-keyframe interpolation, freehand masks, batch AI redraw, automatic pose alignment, multi-track compositing, or audio editing. Public API/MCP generation tools do not accept an Advanced Editor recipe.

Continue with canvas and pivot, output formats, and export troubleshooting.