IM CONFUSED
What does the 8gb lora even do??? it gave worse visuals and a very bad model fast but not good.
If you refer to a distilled LoRA,
It is used on top of full dev checkpoint and allow you to run a generation in just 8 steps instead of 30. dev checkpoint + distilled LoRA basically behaves the same as distilled checkpoint
Do you have a distilled-dev workflow like with LTX 2.0 / 2.3 where the first stage is made with dev and then high-resolution refined with the distilled LoRA?
@kabachuha
The flow that you propose is doable with LTX 2.5 and is quite similar to the LTX 2.3 (Just make sure to use the new text encoder)
@BotLifeGamer
To visualise my point: dev checkpoint + distilled LoRA (green group) is the same as distilled checkpoint (blue group). In fact this is the way distilled LoRA was created. It is basically a diff between the full dev checkpoint and the distilled checkpoint. So if you are loading a distilled model you should not apply distilled LoRA on top of it.