Sooooo get rid of the hybrid loader and just use the load diffusion model node..... 😕

Doesn't work with the last frame ksampler 2 extend. It leaves a gap. It needs a fl2va model alongside ref2va model in hybrid loader.

Doesn't work with the last frame ksampler 2 extend. It leaves a gap. It needs a fl2va model alongside ref2va model in hybrid loader.

I'm extending my videos with - zero - issues.

Does not work with Minimax H3 Hybrid Loader node when paired with minimax_h3_fl2va_pruned_int8_convrot.safetensors or minimax_h3_ref2va_pruned_int8_convrot.safetensors. I think it needs a compatible version of Singularity that can do fl2va as I am unable to do basic ksampler 2 extend workflows using your model.

Error is

ComfyUI Error Report

Error Details

• Node ID: 306:309
• Node Type: MiniMaxH3HybridLoader
• Exception Type: RuntimeError
• Exception Message: RuntimeError: MiniMax H3 hybrid loader: the two checkpoints have different key sets, cannot hybridise. base-only=['model.diffusion_model.adaln_t_table', 'model.diffusion_model.audio_patch_proj.bias', 'model.diffusion_model.audio_patch_proj.weight', 'model.diffusion_model.blocks.0.adaln_proj.linear.bias', 'model.diffusion_model.blocks.0.adaln_proj.linear.weight']... overlay-only=['adaln_t_table', 'audio_patch_proj.bias', 'audio_patch_proj.weight', 'blocks.0.adaln_proj.linear.bias', 'blocks.0.adaln_proj.linear.weight']...

etc etc
Hey, really sorry about that. Yeah, that error is basically a checkpoint-format mismatch: the base and overlay checkpoints have different key sets—one has the model.diffusion_model. prefix, the other has raw keys like adaln_t_table—so MiniMaxH3HybridLoader can’t hybridise them. Right now my model is only optimized for the ref path, not fl2va, so it won’t work properly with an fl2va hybrid loader setup. I’m hoping a V2 release will include an fl version. I’m already putting together the training set and testing the training code, and if nothing blows up, I should be able to start training in about a week. Sorry for the hassle—I’ll keep you posted when V2 is ready.