Skip to content

Add Missing You x Seedance 2.0 and Take Me Stars MV pipelines#26

Open
virtari wants to merge 2 commits intocalesthio:mainfrom
virtari:chef8080-missing-you-production
Open

Add Missing You x Seedance 2.0 and Take Me Stars MV pipelines#26
virtari wants to merge 2 commits intocalesthio:mainfrom
virtari:chef8080-missing-you-production

Conversation

@virtari
Copy link
Copy Markdown

@virtari virtari commented Apr 14, 2026

Summary

  • Missing You MV — Full Seedance 2.0 pipeline: character reference generation (FLUX Pro), 36 scene generation with numbered filenames, iterative scene fixes, and assembly to chef8080_missing_you_FULLMV_720p.mp4
  • Take Me Where The Stars Go MV — colorbalance fix applied, final 70.7 MB assembly
  • Scene improvements — Camera movement added to all scenes (push-in, dolly, tracking); two rounds of iterative fixes; ending changed to reunion-on-bridge (scene 036)
  • Memory / learnings — Production notes saved: Seedance moving camera, Character Bible, Anchor-and-Extend, tempostrategia, professional lighting terminology, bookend structure

New files

File Purpose
missing_you_seedance_mv/mv_generate_chars.py FLUX Pro character ref image generation
missing_you_seedance_mv/mv_generate_scenes.py 36-scene Seedance 2.0 generation pipeline
missing_you_seedance_mv/mv_assemble.py FFmpeg assembly — concat, mix audio, fade

Test plan

  • python mv_generate_chars.py generates 4 character reference images
  • python mv_generate_scenes.py --anchor-only generates anchor clips (001–003)
  • python mv_generate_scenes.py generates remaining 33 production scenes
  • python mv_assemble.py outputs chef8080_missing_you_FULLMV_720p.mp4

🤖 Generated with Claude Code

ari_v and others added 2 commits April 12, 2026 15:15
Adds a complete end-to-end music video production system for the track
"Missing You" by Chef-8080, built on top of OpenMontage tooling.

Production scripts:
- produce_missing_you.py: initial pipeline attempt (v1)
- produce_missing_you_v2.py: portrait 9:16 social teasers with Pexels
  footage + Whisper word-level karaoke subtitle sync
- produce_final_mixed.py: final mixed pipeline combining FAL AI-generated
  woman character shots (Kling v1.6) with Pexels city rain atmosphere,
  interleaved 2:1 city/woman edit pattern, ASS karaoke subtitles
- generate_fal_shots.py: FAL Kling v1.6 generation of 5 consistent
  woman character shots for portrait teasers
- mv_generate_scenes.py: full music video scene generator with 30 FAL
  clips across 10 scene pairs, consistent Sofia/Marcus characters across
  2000 and 2025 eras, 16:9 YouTube format
- mv_assemble.py: 60-shot timeline assembler for full 3-minute YouTube
  16:9 720p video with era-specific color grades, Whisper-timed ASS
  karaoke subtitles, and professional cut rhythm

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- missing_you_seedance_mv/: Full Seedance 2.0 pipeline for "Still Missing You"
  - mv_generate_chars.py: FLUX Pro character reference generation (Sofia/Marcus 2000+2025)
  - mv_generate_scenes.py: 36 scenes with director-level prompts, @Image/@Audio references,
    numbered filenames (001-036), film grade, beat-sync audio segments
  - mv_assemble.py: FFmpeg concat + audio mix → chef8080_missing_you_FULLMV_720p.mp4

- take_me_stars_mv/: Kling v3 i2v pipeline for "Take Me Where The Stars Go"
  - mv_generate_frames.py: FLUX Pro still frame generation with Tapa C film grade
  - mv_generate_scenes.py: 31 scenes, Kling v3 standard i2v with elements (character ref)
  - mv_assemble.py: 54-segment timeline, vintage color grades per location, subtitle burn
  - lyrics.ass: Full ASS subtitle file

- mv_template/: Reusable Seedance 2.0 scene generator template

- Root mv_assemble.py + mv_generate_scenes.py: Minor updates from production learnings

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
@virtari virtari requested a review from calesthio as a code owner April 14, 2026 08:02
@calesthio
Copy link
Copy Markdown
Owner

Integrating Seedance 2.0 is a great idea. It is currently one of the strongest video generation models, so having a real OpenMontage path for it would be valuable.

That said, this PR does not feel mergeable in its current form because it is adding project-specific production scripts rather than integrating Seedance into the platform architecture.

Main issues:

  • The scripts hardcode author-specific local paths, local downloads, and output locations (C:\Users\ari_v\..., specific MP3 inputs, local ffmpeg path), so they are not portable or runnable for other contributors.
  • The implementation bypasses the repo’s pipeline-first design. OpenMontage expects new production capability to flow through pipeline manifests, stage skills, tool registry discovery, and the existing project/artifact conventions rather than direct one-off fal_client scripts in the repo root.
  • Several files look like personal production workspace state or cached references rather than reusable product code (take_me_stars_mv/refs/*, uploaded URL caches, project-specific assets/scripts).
  • There is no tool-level integration, no pipeline integration, and no test coverage showing how this becomes a supported capability in OpenMontage.

I think the underlying direction is good, but I would recommend re-scoping this before merge. A stronger version of this contribution would be one of:

  • a proper Seedance provider/tool integration discoverable through the tool registry,
  • a pipeline/stage integration that uses the existing OpenMontage workflow, or
  • a clearly isolated example/template under an examples or experiments area, with all machine-specific paths removed.

So: strong idea, but this needs to be productized and aligned with the repo architecture before it makes sense to review deeply for merge.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants