Krea2 dont work for me.
Both head swap and body swap throws errors.
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.attn.wv.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.attn.wv.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.down.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.down.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.gate.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.gate.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.up.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.up.lora_B.weight
[krea2edit] NOTE: connect 'target_latent' (the same latent that feeds KSampler.latent_image) to pre-encode the source here instead of on the first sampling step. Without it the VAE is loaded mid-sampling and can evict part of the diffusion model, slowing every remaining step.
[INFO] Requested to load Flux2
[INFO] loaded completely; 17604.02 MB loaded, full load: True
Both head swap and body swap throws errors.
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.attn.wv.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.attn.wv.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.down.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.down.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.gate.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.gate.lora_B.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.up.lora_A.weight
[WARNING] lora key not loaded: diffusion_model.txtfusion.refiner_blocks.1.mlp.up.lora_B.weight
[krea2edit] NOTE: connect 'target_latent' (the same latent that feeds KSampler.latent_image) to pre-encode the source here instead of on the first sampling step. Without it the VAE is loaded mid-sampling and can evict part of the diffusion model, slowing every remaining step.
[INFO] Requested to load Flux2
[INFO] loaded completely; 17604.02 MB loaded, full load: True
One question: are you using the exact workflow I provided with the same models, or are you using other LoRas in conjunction?
Your workflow and no other Loras.
BTW: On a Macbook pro.
Normal Krea2 text to image works fine.
EDIT: Using Krea2 turbo Q8 gguf.. Maybe the reason?