Skip to content

Shenrui-Ma/video-regen-recipes / templates/character-short/bit-by-bit-heartache

bit-by-bit-heartache

Verified

用角色参考图在本机制作 Heartache 连续动作配乐短片.

FileKindSize
.gitignoreother430 B
LICENSEother1.1 KB
LICENSES.mdmarkdown1.4 KB
README.mdmarkdown9.3 KB
SKILL.mdskill md7.3 KB
assets/README.mdmarkdown4.9 KB
assets/default/manifest.jsonconfig3.8 KB
assets/runtime-assets.jsonconfig5.5 KB
profile.jsonconfig3.8 KB
prompts/ref2va.template.txtmarkdown1.6 KB
prompts/yaoguang.historical.txtmarkdown5.0 KB
references/background-removal.mdmarkdown5.5 KB
references/changelog.mdmarkdown916 B
references/character-input.mdmarkdown4.7 KB
references/deployment.mdmarkdown4.9 KB
references/editing.mdmarkdown3.7 KB
references/environment.lock.jsonconfig1.5 KB
references/first-segment-validation.jsonconfig1.8 KB
references/generation.mdmarkdown3.8 KB
references/media-preflight.mdmarkdown4.2 KB
references/native-continuation.mdmarkdown17.8 KB
references/node-sources.jsonconfig6.7 KB
references/node-sources.mdmarkdown3.8 KB
references/package-verification.jsonconfig1.7 KB
references/reference-check.jsonconfig968 B
references/reproduction.mdmarkdown10.1 KB
references/validation.mdmarkdown2.6 KB
references/workflow-recovery.mdmarkdown2.2 KB
references/workflow.lock.jsonconfig1.2 KB
references/yaoguang-run.jsonconfig7.4 KB
scripts/character/prepare_character.pyscript6.5 KB
scripts/distribution/fetch_assets.pyscript5.8 KB
scripts/distribution/package_skill.pyscript5.6 KB
scripts/plan_segments.pyscript3.4 KB
scripts/prepare_driver.pyscript5.2 KB
scripts/render_default.pyscript5.5 KB
scripts/runtime/graphs.pyscript11.3 KB
scripts/runtime/heartache.pyscript14.6 KB
scripts/runtime/local_io.pyscript3.3 KB
scripts/runtime/media.pyscript13.2 KB
scripts/runtime/run_segment.pyscript13.3 KB
sources.mdmarkdown2.4 KB
tests/test_background_removal.pyscript2.6 KB
tests/test_character_input.pyscript12.2 KB
tests/test_default.pyscript879 B
tests/test_distribution.pyscript12.3 KB
tests/test_media_preflight.pyscript2.1 KB
tests/test_prepare_driver.pyscript1.9 KB
tests/test_release_review.pyscript1.7 KB
tests/test_reproduction.pyscript1.9 KB
tests/test_runtime_cli.pyscript11.3 KB
tests/test_runtime_graph.pyscript4.8 KB
tests/test_runtime_io.pyscript2.1 KB
tests/test_runtime_media.pyscript10.2 KB
tests/test_runtime_runner.pyscript1.6 KB
tests/test_segment_plan.pyscript3.4 KB
tests/test_skill_contract.pyscript1.6 KB
workflows/continue.editor.jsonconfig23.8 KB
workflows/first.editor.jsonconfig19.8 KB
workflows/node-schema.jsonconfig22.2 KB
workflows/provenance.jsonconfig911 B

references/generation.md

View on GitHub
# 新角色生成:当前可执行路径

从独立模板根目录按 [README](../README.md)执行。全部H3首段、续接、输入和后处理脚本都随包提供;无需其他个人Skill或私有Save/Load节点。

## 输入和条件

角色参考图→LoadImage→Ref2VA图像条件;577帧驱动→LoadVideo→GetVideoComponents.images→Ref2VA视频条件。驱动音轨不接入模型;正式配乐在发布画面拼接后重铺。参考条件提供身份和动作引导,不等于硬首帧或逐帧骨骼约束。

固定基础为1344×768、24fps、20步、res_multistep/simple、BasicGuider、denoise=1。模型与依赖版本参考见[环境参考记录](https://github.com/Shenrui-Ma/shenrui-comfyui-toolkit/tree/667eafddb9f42bb6c72ad27b665def9ada36df45/environments/h3/README.md)(默认复用用户已有环境),角色文字使用[prompt模板](../prompts/ref2va.template.txt)。不得把历史爻光描述沿用于新角色。

## 采样与发布计划

`plan_segments.py`以实测容量求计划;采样帧按17k+5网格对齐。首段最多发布全部采样帧,后段保留22帧context成本,仅发布其后新画面;末段补齐网格的尾帧不进入发布结果。驱动切片与该段context+新增画面对齐,不能把历史四段的切点套到新计划。

例如实测最大107采样帧时,首段发布107,后续满段各发布85;总长由实际驱动577帧决定。这仅是容量示例。新机器先做完整首段和续接校准,记录模型常驻、RAM、显存与解码峰值。

## 真续接

1. Sampler产生完整联合AV latent。
2. Core `LTXVSeparateAVLatent`拆成video/audio,分别用Core `SaveLatent`保存,外部sidecar保存发布区间、前驱及SHA。
3. 下一段Core `LoadLatent`两路读取,`LTXVConcatAVLatent`合并,接到公开`MiniMaxH3MotionContext.context_latent`。
4. MotionContext的第二输出是INT裁切量,连`ImageFromBatch.batch_index`;length为计划发布帧数。
5. 保留完整latent与已经裁过context的发布画面,两者用途不同;后期只拼发布画面,不再次裁22帧。

Core LoadLatent会读成F32。这里只声明已核验F32往返,不泛化到其他dtype。私有canonical旧格式的字段及历史四段区别留在[历史续接记录](native-continuation.md),不作为新安装的依赖。

## 运行和恢复

用`scripts/runtime/heartache.py`准备、检查、显式执行。每段runner先写提交意图,取得prompt_id后只围绕原ID查history;提交结果不明时禁止自动重投。用户取消后停止后续提交,完整保留已完成前驱。

采样成功但解码失败时,不应重新采样。已有两路latent可接无Sampler解码图恢复;确认header/shape、前驱、history和SHA后,完成原发布区间。修改前驱采样则必须新建版本并重验全部依赖下游。

## 参考背景与后处理预检

只借用人物身份而不保留参考图背景时,可先使用[去背景流程](background-removal.md)。保留alpha图,再明确合成纯色RGB送入H3,防止忽略alpha时读入隐藏背景颜色。该步骤不改变动作驱动、seed或采样规格。

按[媒体预检](media-preflight.md)在实际runner的PATH测试AAC和完整后处理。发布视频已成功而配乐失败时,仅恢复后处理,不重新采样。

## 后处理

拼接连续发布画面,原曲从时间线零点连续铺设、最后半秒淡出。每个阶段可生成第一至N段累计成片。技术校验包括区间连续、帧数、尺寸、24fps、视频画面未被换源、音轨覆盖时长、完整解码和SHA;身份、动作和接缝视觉效果由使用者确认。

[原版逐帧/逐字节对照材料](reproduction.md)与新角色再生成是两种验收目标:随机性、量化、硬件和不同分段都可能改变像素结果。
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 194kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 194kUpdated Claude CodeCodexVerifiedInstall