Hugging Face's logo

Lightricks
/
LTX-2.5-Diffusers

Text-to-Video
Diffusers
Safetensors
LTX2Pipeline
image-to-video
video-to-video
image-text-to-video
audio-to-video
text-to-audio
video-to-audio
audio-to-audio
text-to-audio-video
image-to-audio-video
image-text-to-audio-video
ltx-2
ltx-video
ltxv
lightricks
ltx-2.5
Model card Files
xet
 Community
3

Distilled snippets: disable STG + modality guidance explicitly

#9
by multimodalart HF Staff - opened
base: refs/heads/main
←
from: refs/pr/9
Discussion Files changed
+0
-0
initial commit6f0ce97a
 Update README.md7a651ad4
 Update README.md654941f0
 Update README.md97f7ea0a
 Upload folder using huggingface_hub882da37a
 Update README.md1e2faee0
 Update README.mddc195d7a
 Update README.md681610a0
 Update README.md5cf1cd2d
 Update README.md015937f6
 Add latent_upsampler for the two-stage distilled recipe (#1)bf2b6c27
 Update README.md73389066
 Update README.md03df0416
 Add modular_model_index.json to enable modular loading (#3)1a7b1df0
 Upload folder using huggingface_hub1757fe62
 Upload folder using huggingface_hub77c21cd2
 Upload folder using huggingface_hub19322ace
 Align the diffusers usage section with the shipping API (#2)b7f0659c
 Upload ltx-2.5-22b-distilled-lora-450-bf16.safetensors with huggingface_hubeec17bd4
 Nest the full-DiT load inside the pipeline init in the SFT diff (#4)5660a68e
 Swap the diffusion decoder processor with set_attn_processor (#5)7ede5ac1
 Enable tiling in the diffusion decoder section (#6)ec370845
 Add the install-from-main note (#7)9fafb143
 Move the install note next to the download step (#8)7b197ba2
 Distilled snippets: disable STG + modality guidance explicitly43c98d0c
multimodalart
 LTX.io org
•
edited 2 days ago

stg_scale and modality_scale default to the SFT values and are gated independently of guidance_scale (> 0 and > 1 respectively), so the distilled snippets were running STG + modality guidance. This adds the missing overrides; both audio-side values are passed explicitly because audio_stg_scale or stg_scale resolves a falsy 0.0 back to the video value.
simplify: just the missing kwargs195ef4a7
multimodalart changed pull request status to closed
multimodalart
 LTX.io org

Superseded by https://huggingface.co/Lightricks/LTX-2.5-Diffusers/discussions/10 — this branch's base was orphaned by the history rewrite, closing.
multimodalart changed pull request status to open
multimodalart changed pull request status to closed

Sign up or log in to comment