Add LTX-2.4 support to the LTX2 transformer, pipelines, and conversio…
78fcaf83
Fix resolution-dependent timestep shift being a no-op in all LTX2 pip…
420ad559
Add LTX-2.4 duration head for automatic num_frames prediction
2bc3c933
Add LTX-2 modular video and audio decoder blocks
963f1e4d
Add LTX-2 modular pipeline (t2v + i2v) and parity harnesses
6ca2c697
Add script to create tiny LTX-2.4 test pipeline ckpt (under assumptio…
d39c7189
make fix-copies
d6bbed14
Run modular connector once on CFG-concatenated batch for bitwise parity
f8bf0fa8
Use tiny LTX2VocoderWithBWE in the LTX-2.4 test checkpoint
181861e3
Match LTX-2.4 transformer flags in the test checkpoint
d0335ba4
Add CPU offload flag to parity scripts to be able to load full checkp…
94cf57be
Support dynamic timestep shift and duration head in LTX-2 modular pip…
ee19558d
Add duration predictor support for tiny checkpoint and parity scripts
4111b09f
Merge branch 'main' into ltx-2-4-modular
d4085b3d
Refactor LTX-2 modular guidance into video/audio guiders
3582e4a6
Batch cond+uncond CFG forward to mirror standard LTX-2 pipeline
932d8241
Drive LTX-2 modular guidance through the guider API (per-pass inputs)
c372297b
Drive LTX-2 modular guidance via prepare_inputs_from_block_state
fefc2b7b
Route LTX-2 batch-invariant denoiser kwargs via denoiser_input_fields
7482c46b
Fix LTX-2 parity harnesses: guidance overrides reach both pipelines
e2cb6f52
Document denoiser_input_fields standalone-run caveat for LTX-2 denoiser
7e11e364
Address LTX-2 modular self-review: fix dead enhancer trigger + docstr…
337d0bf1
Fix stale I2V parity-harness --help text for the loosened fp32 gate
2aa7ecac
make style and make quality
3a73088d
Add the LTX-2.4 diffusion VAE decoder with a native-parity harness
ec0b841d
Add a converter for the LTX-2.4 diffusion VAE decoder
acf5c7de
Initialise use_slicing and document the diffusion VAE decisions
255fc8dd
Record real-weight parity against the reference decoder
5fac7aa9
Add model-level tests for the LTX-2.4 diffusion VAE decoder
926818ac
Fold the diffusion VAE converter into convert_ltx2_to_diffusers.py
887c1861
Make the diffusion decoder work through LTX2Pipeline
3928fc26
Make the diffusion decoder work through the modular decode step
c6b8c4a9
Note that the pipeline decode block predates the LTX-2.4 integration PR
fbd87338
Address diffusers self-review on the diffusion VAE decoder
dbd6acfb
Add the API doc page for AutoencoderKLLTX2VideoDiffusionDecoder
2303e01e
Mark the parity harness transient and drop the notes doc from the diff
7ee18bb7
Accept the diffusion decoder in the image-to-video pipeline too
558d8e44
Reorganize LTX-2 modular blocks into coarser component-aligned phases
2b6106eb
Enable LTX-2 prompt enhancement by default when the enhancer is loaded
661a6e4c
Rename LTX-2.4 product identifiers to LTX-2.5.
f21853fe
Add image-conditioning H.264 CRF matching Lightricks training.
dd6c5598
Align LTX-2 pipelines with the Lightricks reference behavior.
524ec976
Update LTX-2 docs for 2.5 naming, defaults, and opt-in enhancement.
52fc83c0
Replace LTX2AutoDuration with min_seconds/max_seconds call args.
0ec61d60
Merge branch 'pr-2-ltx-2-4-integration' into ltx-2-4-modular
b24e69f0
Align LTX-2 modular blocks with the merged standard-pipeline changes
5d18b5d2
Evaluate the decoder's SwiGLU in token tiles
8e5df58d
Declare _no_split_modules on the diffusion decoder
031b5621
Match the encoder defaults to the published LTX-2.4 config
a68d6d23
Cover the token-tiled SwiGLU path in the tests
68755f0c
Let the parity harness run against the LTX-2.5 reference too
c984a005
Add LTX-2.X modular condition blocks
e117b223
Add LTX-2.X modular in-context (IC-LoRA) blocks
58052bf5
Make the diffusion decoder a decode pipeline instead of a VAE
d71af8c5
Add a modular decode block for the LTX-2 diffusion decoder
42f45102
Test the diffusion decode pipeline, and give it a video-only output
474452d4
Fix the decode example: output_type="latent" is already denormalized
8e978fdd
Fix IC-LoRA video self-attention mask grouping
dd92e468
Add LTX25AutoBlocks and LTX25ModularPipeline
0a147159
Drop the encoder half in the diffusion decoder converter
1d6fdce3
Share the neighborhood attention mask and parameterise the AdaLN chun…
d97dcac0
Export the LTX-2.5 modular classes from the top-level namespace
d354288b
Name the diffusion decoder LTX-2.5 and save it to diffusion_decoder/
31414b84
Move the LTX-2 text-input expansion into the core denoise step
24404d18
Describe the connector inputs on LTX2TextInputStep
1f0e2e2a
Merge branch 'ltx-2-4-modular' into ltx-2-4-modular-condition-ic
c6bd5a64
Build the condition and in-context workflows from LTX2AutoBlocks
5bb24c67
Run the text-input step in the condition and in-context denoise steps
47cf7585
Broadcast keyframe and reference conditioning across the generation b…
4f7a060b
Merge pull request #12 from huggingface/ltx-2-4-modular-condition-ic
63fdd15e
Merge remote-tracking branch 'origin/ltx-2-4-modular' into ltx-2.5
33b3c06a
Merge branch 'ltx-2-4-modular' into ltx-2-4-diffusion-vae-modular
7a73ce6c
feat: use natten for processing. (#15)
6593e3b0
Merge PR #8 (LTX-2.4 diffusion VAE decoder) into ltx-2.5
64839af9
Merge PR #8 update: fetch NATTEN as a Hub kernel
5b6f9def
make vae properties accessible without loading it. (#17)
4d329633
Address review: replace LTX2 mixins with Copied-from pipeline methods.
50bc9f8b
Point the modular enhance-prompt comment at LTX2Pipeline
6e92e152
Merge PR #8 update: read latent statistics without loading the vae
f512cbdb
Remove the integrations/ dev scripts
bded3698
moular tests (#19)
7ee18313
Keep LTX-2.5.1 keyframes_abs_pos_embedding through convert. (#20)
f20c0004
remove test_ltx2_keyframes_abs_pos_embedding_load_and_forward
a6803d6d
Update the LTX-2.5 docs: two stage generation, distilled default, rep…
a1c06c9f
feat: implement tiling for the diffusion decoder. (#22)
c8e16b90
add a note on tiling (#23)
68c743be
Merge branch 'main' into ltx-2.5
fb8f932f
fix documentation.
83cca549
yiyixuxu
commented
on 2026-08-11
yiyixuxu
commented
on 2026-08-11
yiyixuxu
commented
on 2026-08-11
Login to write a write a comment.
Login via GitHub