{{ message }}
Repository navigation
Commit b3889ea
Add modular pipeline for HunyuanVideo 1.5 (#13389)
* Add modular pipeline support for HunyuanVideo 1.5
* Fix I2V latent/cond spatial dimension mismatch
* Fix guidance_scale default to 7.5 matching ClassifierFreeGuidance
* Fix tokenizer type: use Qwen2TokenizerFast to match model
* Fix system message string formatting to match standard pipeline
* Rewrite HunyuanVideo 1.5 modular: use standard pipeline methods directly
* Remove I2V exports (T2V only for now)
* Fix encoder: use static methods directly instead of encode_prompt
* Inline all standard pipeline methods, remove runtime dependency
* Add HunyuanVideo 1.5 image-to-video modular blocks
* Fix missing FrozenDict import in before_denoise.py
* auto-generated docstrings via #auto_docstring
* Fix ruff lint and format issues
* use InputParam/OutputParam templates and fix ruff
* Address LTX review feedback here like add AutoBlocks, refactor I2V latents, lift encoders
* Add workflow map, workflow tests, auto docstrings, export only AutoBlocks
* Address Claude CI review
* Address claude CI review 2
---------
Co-authored-by: YiYi Xu <yixu310@gmail.com>1 parent e0c1ec4 commit b3889ea
13 files changed
Lines changed: 2036 additions & 0 deletions
File tree
- src/diffusers
- modular_pipelines
- hunyuan_video1_5
- utils
- tests/modular_pipelines/hunyuan_video1_5
| Original file line number | Diff line number | Diff line change | |
|---|---|---|---|
| |||
88 | 88 | | |
89 | 89 | | |
90 | 90 | | |
| 91 | + | |
| 92 | + | |
| 93 | + | |
| 94 | + | |
91 | 95 | | |
92 | 96 | | |
93 | 97 | | |
| |||
123 | 127 | | |
124 | 128 | | |
125 | 129 | | |
| 130 | + | |
| 131 | + | |
| 132 | + | |
| 133 | + | |
126 | 134 | | |
127 | 135 | | |
128 | 136 | | |
| |||

0 commit comments