Image-to-Video
MiniMax H3
comfyui
latent-upscaler
first-last-frame
flf
bridge
reference-to-video
r2v
controlnet
controlnet-union
openpose
canny
depth
v2v
text-to-speech
tts
video-generation
int8
block-sparse-attention
veda-sparse
spectrum
turbo-lora
lightx2v
extend
switchboard
group-bypass
workflow
optimization
quantization
2nd stage latent upscale fries image
#1
by MJS255 - opened
It's completely frying the image for me, haven't changed any nodes. Tried 4-10 steps still the same. Pruned int8 models . Looks completely fine with 2 stage disabled. Cu130 torch 2.13 5070. I2V workflow
I'm very sorry, but I removed the latent upscale feature from the workflow because it wasn't producing the intended quality consistently. I apologize for the inconvenience, and thank you for reporting this and for the detailed feedback.
@MJS255 Frying an image is not issue with the upscale itself but a combination of sampler/steps/turbo Lora strength, which are frying image even without upscaling. So you must adjust accordingly.
MJS255 changed discussion status to closed