Connection order

[my2interid]

I have these nodes
model-lora-lora-Patch Sage Attention KJ-ModelSamplingAuraFlow-Sampler.
Where do I connect the CacheDiT Accelerator?

[eccentricworx]

I have these nodes model-lora-lora-Patch Sage Attention KJ-ModelSamplingAuraFlow-Sampler. Where do I connect the CacheDiT Accelerator?

My guess is, right before nothing is changing in your flow anymore. Which means if you change LORAs a lot, it should be between the model and the LORAs. I have this Node right after the model as i tend to change a lot in the chain for testing, the model always stays...

[my2interid]

Some AI systems say that it should be connected at the very end, right before the sampler. Because of this conflicting information, I am confused about the correct connection order.

Load Diffusion Model (Load the LTX-Video checkpoint)

Patch Sage Attention (KJ): First modify the attention mechanism to improve memory efficiency.

Cache-DiT: Apply caching on top of the modified model structure to eliminate redundant computations.

KSampler: Perform sampling using the fully optimized model.

[my2interid]

Through multiple controlled tests, I consistently observed that this node order delivers noticeably faster video generation compared to other configurations.