diffusers
Add Support for LTX-2.3 Models
#13217
Merged

Add Support for LTX-2.3 Models #13217

dg845 merged 43 commits into main from ltx2-3-pipeline
dg845
dg845 Initial implementation of perturbed attn processor for LTX 2.3
6c7e720d
yiyixuxu
yiyixuxu commented on 2026-03-06
dg845 Update DiT block for LTX 2.3 + add self_attention_mask
e90b90a3
dg845 Add flag to control using perturbed attn processor for now
f768f8da
dg845 Add support for new video upsampling blocks used by LTX-2.3
cde67486
dg845 Support LTX-2.3 Big-VGAN V2-style vocoder
236eb8db
dg845 Initial implementation of LTX-2.3 vocoder with bandwidth extender
1e89cb36
dg845 Initial support for LTX-2.3 per-modality feature extractor
5a44adb0
dg845 Refactor so that text connectors own all text encoder hidden_states n…
4ff31688
dg845 Fix some bugs for inference
835bed61
dg845 Fix LTX-2.X DiT block forward pass
19004efc
dg845 Support prompt timestep embeds and prompt cross attn modulation
4dfcfeb3
dg845 Add LTX-2.3 configs to conversion script
13292dde
HuggingFaceDocBuilderDev
dg845 Support converting LTX-2.3 DiT checkpoints
0528fde4
dg845 Support converting LTX-2.3 Video VAE checkpoints
c5e1fcc4
dg845 Support converting LTX-2.3 Vocoder with bandwidth extender
50da4df0
dg845 Support converting LTX-2.3 text connectors
42062803
dg845 Don't convert any upsamplers for now
e719d32c
dg845
dg845 Support self attention mask for LTX2Pipeline
fbb50d96
dg845 Fix some inference bugs
de3f869b
dg845 Support self attn mask and sigmas for LTX-2.3 I2V, Cond pipelines
5056aa82
dg845 Support STG and modality isolation guidance for LTX-2.3
f875031d
dg845 make style and make quality
652d363d
dg845
dg845 dg845 marked this pull request as ready for review 11 days ago
dg845 Make audio guidance values default to video values by default
d018534d
dg845 Update to LTX-2.3 style guidance rescaling
c0bb2ef2
tin2tin
dg845 Support cross timesteps for LTX-2.3 cross attention modulation
ab0e5b5c
dg845 Fix RMS norm bug for LTX-2.3 text connectors
f78c3dae
dg845 Perform guidance rescale in sample (x0) space following original code
63b3c9f2
dg845 Support LTX-2.3 Latent Spatial Upsampler model
6188af22
dg845
dg845 Support LTX-2.3 distilled LoRA
89f8cc43
dg845 Support LTX-2.3 Distilled checkpoint
f1a812aa
dg845
dg845 Support LTX-2.3 prompt enhancement
145e8e48
dg845 Make LTX-2.X processor non-required so that tests pass
8a580734
dg845 Fix test_components_function tests for LTX2 T2V and I2V
93247a0a
dg845 Fix LTX-2.3 Video VAE configuration bug causing pixel jitter
17b53f08
dg845 Merge branch 'main' into ltx2-3-pipeline
6ee66c9d
dg845 dg845 changed the title [WIP] Add Support for LTX-2.3 Models Add Support for LTX-2.3 Models 6 days ago
dg845 dg845 requested a review from yiyixuxu yiyixuxu 6 days ago
dg845 dg845 requested a review from sayakpaul sayakpaul 6 days ago
sayakpaul
sayakpaul commented on 2026-03-16
dg845 Apply suggestions from code review
c016ce56
dg845 Refactor LTX-2.X Video VAE upsampler block init logic
2feb4609
yiyixuxu
yiyixuxu commented on 2026-03-18
dg845 Refactor LTX-2.X guidance rescaling to use rescale_noise_cfg
27404098
dg845 Use generator initial seed to control prompt enhancement if available
5d8b6342
dg845 Remove self attention mask logic as it is not used in any current pip…
b0723de8
dg845 Commit fixes suggested by claude code (guidance in sample (x0) space,…
67a9ce33
dg845 Use constant shift following original code
4cbedd7c
yiyixuxu
yiyixuxu approved these changes on 2026-03-19
dg845 Merge branch 'main' into ltx2-3-pipeline
8a9a1480
tin2tin
dg845
tin2tin
dg845 dg845 merged 072d15ee into main 2 days ago
dg845 dg845 deleted the ltx2-3-pipeline branch 2 days ago
tin2tin

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone