Repository navigation
Adapt gallery stories and preserve supplied-video choreography - #5
Merged
Merged
Conversation
cth9191
changed the title
cth9191
marked this pull request as ready for review
September 6, 2026 20:50
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Motion references need different treatment depending on the request. Gallery prompts adapt their visual and motion techniques around the viewer takeaway. A supplied video defaults to a new generation conditioned on that clip, preserving its sequence, composition, timing and camera moves while generating the new subject and lettering. Explicit loose inspiration or editing requests override that default.
The supplied-video guidance uses a concise source-to-replacement map and compares corresponding moments. It keeps retries on the original reference-generation route instead of switching to direct editing or manual typography. This corrects the edit-first routing introduced during development. There is no fixed scene count or scoring framework; archived gallery source prompts remain unchanged.
Validation: the skill validator, git diff --check and both GitHub validation jobs pass. Updated guidance is installed in Codex and Claude Code. Earlier gallery validation covered all 10 original source hashes and catalog timelines. Previous free adaptation, direct-edit and manual-composite trials did not satisfy the requested demonstration; they do not establish reliable fidelity or text replacement. Two fresh Seedance 2.5 omni_reference generations used only the original clip. Sampled frames show the source sequence and camera/UI composition are followed much more closely, but both retain source branding and health-related text despite explicit substitutions. Neither meets the new-subject brief. The demo exposes the untouched outputs and this limitation; no direct edit or manual finishing was used for these trials.