Quoted By:
Local instance of FLUX (schnell, not dev) up and running! we are so back!
benchmarks:
loading model with full weights (unpruned): 70-75 seconds
1024x1024 gen with same prompt: 45-50 seconds
1024x1024 gen after changing prompt: 115-130 seconds
*pruned models are already coming, out they will be faster (with slightly lower quality) but this is not bad at all for trying it out
for comparison, the normal-tier 1024s I typically spam take 25-30 sec each and changing prompt doesn't matter for gen speed. but those are all pruned models
I'll accept requests (at my discretion) for the rest of the thread