Regarding ltx2.3-VBVRlora
Dear production team, the current VBVR-Wan2.2 loras have achieved excellent results in the Wan2.2 community ecosystem. The compliance with dynamic and prompt words has significantly improved, and the spatial relationship of objects has also been significantly enhanced. May I ask if you will consider implementing this technology on the ltx2.3 audio-video synchronization model? This model has just been open-sourced for a few days, and the improvement compared to ltx2.0 is obvious. The compliance with prompt words, dynamics, audio generation, and consistency have all been enhanced. However, the current open-source community ecosystem is not very complete, and there is a shortage of such high-quality loras. Here, I pay tribute to the production staff of the team. Thank you very much!!
Dear @wange1002 ,
Thank you for your interest and for the kind words about our work. We’re glad to hear that the VBVR-Wan2.2 LoRAs have been helpful to the community.
We have also noticed the recently open-sourced ltx2.3 model and are currently conducting an internal evaluation of its capabilities and potential integration. Once the evaluation is complete, we will consider possible support and further development within our ecosystem.
In the meantime, you’re welcome to join our Slack channel for further discussion and updates with the team and community. (https://video-reason.slack.com/join/shared_invite/zt-3qqf23icm-UC29fatWWYsIuzRNBR1lgg)
We truly appreciate your support and feedback. Thank you again for following our work.
Best regards,
Wang Ruisi
On behalf of the VBVR team
It's a great honor. Thank you very much. Thank you!
Hi, just a quick update, we will release a model trained on LTX-2.3 in one or two weeks. Stay tuned and thank you for your interest!
Hi, just a quick update, we will release a model trained on LTX-2.3 in one or two weeks. Stay tuned and thank you for your interest!
Thank you. Cant wait to try out that ;-) already works great with Wan
@RuneXX @RedHn @EclipseMist @Manderlore @DrWh0
Thank you for your interest. In this release, we present VBVR-Wan2.1 (Diffusers format), VBVR-Wan2.1-diffsynth (DiffSynth LoRA format), and VBVR-LTX2.3-diffsynth (DiffSynth LoRA format; Diffusers does not yet support LTX-Video-2.3, so only the DiffSynth LoRA format is released for this model).
@RuneXX @RedHn @EclipseMist @Manderlore @DrWh0
Thank you for your interest. In this release, we present VBVR-Wan2.1 (Diffusers format), VBVR-Wan2.1-diffsynth (DiffSynth LoRA format), and VBVR-LTX2.3-diffsynth (DiffSynth LoRA format; Diffusers does not yet support LTX-Video-2.3, so only the DiffSynth LoRA format is released for this model).
Appreciate your work and research! I created the first adaptation of VBVR for LTX 2.3 and uploaded the first version of it to civitAI 20 days ago. Up to version 3 which I stripped the ffn layers and noticed it improved motion a lot. My version is trained on 4800 videos and I noticed lots of improvements in temporal consistency and prompt adherence so I cant wait to see how well yours which is trained on a ton more than that performs. Did you also end up excluding the ffn from the final lora?
Testing your VBVR lora it causes a lot of stuttering like makes the video look robotic almost im not sure why
Testing your VBVR lora it causes a lot of stuttering like makes the video look robotic almost im not sure why
Used it a bit inside Comfy with WanVideoWrapper. Worked great for me at least