Plastic look? Darker output?
Hi.
Your generations look brilliant, while my look like there would be too strong lora - glass/plastic look.
I used speed lora or not used it - the same result.
The output seems to be darker than the original video - changes light?
I tried with Res_2s/beta 4 steps sampling.
What sampler/scheduler to use, how many steps?
Did you test different resolutions? I wish 2, 2.5 or 3 mega pixels.
Could you please upload videos with workflow so I can see how did you achieve your quality?
Hi.
Your generations look brilliant, while my look like there would be too strong lora - glass/plastic look.
I used speed lora or not used it - the same result.
The output seems to be darker than the original video - changes light?
I tried with Res_2s/beta 4 steps sampling.What sampler/scheduler to use, how many steps?
Did you test different resolutions? I wish 2, 2.5 or 3 mega pixels.
Could you please upload videos with workflow so I can see how did you achieve your quality?
The configuration I use is exactly the one in the workflow, nothing more than that. Also, I haven't tested it with high resolutions because H3 takes a long time to generate at those settings. If the LoRA is too strong, you can reduce the strength, and yes, it can alter color and lighting. Try using the new DMD LoRA extracted from VDN (https://huggingface.co/drbaph/MiniMax-H3-Turbo-Lora-ComfyUI/blob/main/experimental/minimax_h3_dmd_ref2va_8step_turbo_pruned.safetensors), sometimes it can help, you need test what the config will be better for you. Also, the base model documentation states that the maximum generation resolution is 2K; plus, you need a very powerful machine to use guided latents at that resolution.