After waiting for a month, Light Speed finally gave up.
I'm not quite sure about the significance of releasing this version of the model. Compared to version 2.3, I don't see any improvement in this model. Even with the addition of prompt word enhancement in the ComfyUI template workflow, there is still poor character consistency, chaotic movements, occasional screen crashes, and even a simple action like taking off glasses cannot be completed normally. The successful scenes can only be found in the official demonstration videos.
Are you even able to run the model fully to really see it's capabilities?
Are you even able to run the model fully to really see it's capabilities?
I have tried the three official template workflows in ComfyUI. Apart from being a bit faster, I didn't notice any significant difference in the output results compared to LTX 2.3.
像素漂移得到了改善，但是涂抹感非常重。失去了质感。
Are you even able to run the model fully to really see it's capabilities?
I have tried the three official template workflows in ComfyUI. Apart from being a bit faster, I didn't notice any significant difference in the output results compared to LTX 2.3.
Using a workflow doesn't really tell anything, there's a difference using full BF16 or quantized Q8.
If you optimize for speed then you don't see how good the quality can be.
Are you even able to run the model fully to really see it's capabilities?
I have tried the three official template workflows in ComfyUI. Apart from being a bit faster, I didn't notice any significant difference in the output results compared to LTX 2.3.
Using a workflow doesn't really tell anything, there's a difference using full BF16 or quantized Q8.
OK, maybe you're right, but my equipment doesn't support the use of the full-scale model. I'm not the intended user of this model. I hope the team can continue as they are now. At least for me, there's no reason not to switch to another model.