diving-z-image-turbo_v5.0_fp8

ComfyUI model

Origin Repo(model) on CivitAI diving-z-image-turbo(v5.0)

full fp8(e4m3fn) converted from the fp16 model Download

using comfy-dit-quantizer, edited and used all-fp8.json as:

{
  "format": "comfy_quant",
  "block_names": [""],
  "rules": [
    { "policy": "keep", "match": [] },
    { "policy": "float8_e4m3fn", "match": [""] },
    { "policy": "nvfp4",  "match": [] }
  ]
}

using "format": "comfy_quant" can prevent error message in ComfyUI like:

unet unexpected: ['model.diffusion_model.cap_embedder.1.comfy_quant', 'model.diffusion_model.context_refiner.0.attention.out.comfy_quant', 'model.diffusion_model.context_refiner.0.attention.qkv.comfy_quant', 'model.diffusion_model.context_refiner.0.feed_forward.w1.comfy_quant', 'model.diffusion_model.context_refiner.0.feed_forward.w2.comfy_quant', 'model.diffusion_model.context_refiner.0.feed_forward.w3.comfy_quant', 'model.diffusion_model.context_refiner.1.attention.out.comfy_quant', 'model.diffusion_model.context_refiner.1.attention.qkv.comfy_quant', 'model.diffusion_model.context_refiner.1.feed_forward.w1.comfy_quant', 'model.diffusion_model.context_refiner.1.feed_forward.w2.comfy_quant', 'model.diffusion_model.context_refiner.1.feed_forward.w3.comfy_quant', 'model.diffusion_model.final_layer.adaLN_modulation.1.comfy_quant', 'model.diffusion_model.final_layer.linear.comfy_quant', 'model.diffusion_model.layers.0.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.0.attention.out.comfy_quant', 'model.diffusion_model.layers.0.attention.qkv.comfy_quant', 'model.diffusion_model.layers.0.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.0.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.0.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.1.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.1.attention.out.comfy_quant', 'model.diffusion_model.layers.1.attention.qkv.comfy_quant', 'model.diffusion_model.layers.1.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.1.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.1.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.10.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.10.attention.out.comfy_quant', 'model.diffusion_model.layers.10.attention.qkv.comfy_quant', 'model.diffusion_model.layers.10.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.10.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.10.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.11.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.11.attention.out.comfy_quant', 'model.diffusion_model.layers.11.attention.qkv.comfy_quant', 'model.diffusion_model.layers.11.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.11.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.11.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.12.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.12.attention.out.comfy_quant', 'model.diffusion_model.layers.12.attention.qkv.comfy_quant', 'model.diffusion_model.layers.12.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.12.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.12.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.13.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.13.attention.out.comfy_quant', 'model.diffusion_model.layers.13.attention.qkv.comfy_quant', 'model.diffusion_model.layers.13.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.13.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.13.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.14.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.14.attention.out.comfy_quant', 'model.diffusion_model.layers.14.attention.qkv.comfy_quant', 'model.diffusion_model.layers.14.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.14.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.14.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.15.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.15.attention.out.comfy_quant', 'model.diffusion_model.layers.15.attention.qkv.comfy_quant', 'model.diffusion_model.layers.15.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.15.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.15.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.16.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.16.attention.out.comfy_quant', 'model.diffusion_model.layers.16.attention.qkv.comfy_quant', 'model.diffusion_model.layers.16.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.16.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.16.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.17.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.17.attention.out.comfy_quant', 'model.diffusion_model.layers.17.attention.qkv.comfy_quant', 'model.diffusion_model.layers.17.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.17.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.17.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.18.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.18.attention.out.comfy_quant', 'model.diffusion_model.layers.18.attention.qkv.comfy_quant', 'model.diffusion_model.layers.18.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.18.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.18.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.19.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.19.attention.out.comfy_quant', 'model.diffusion_model.layers.19.attention.qkv.comfy_quant', 'model.diffusion_model.layers.19.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.19.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.19.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.2.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.2.attention.out.comfy_quant', 'model.diffusion_model.layers.2.attention.qkv.comfy_quant', 'model.diffusion_model.layers.2.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.2.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.2.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.20.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.20.attention.out.comfy_quant', 'model.diffusion_model.layers.20.attention.qkv.comfy_quant', 'model.diffusion_model.layers.20.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.20.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.20.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.21.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.21.attention.out.comfy_quant', 'model.diffusion_model.layers.21.attention.qkv.comfy_quant', 'model.diffusion_model.layers.21.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.21.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.21.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.22.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.22.attention.out.comfy_quant', 'model.diffusion_model.layers.22.attention.qkv.comfy_quant', 'model.diffusion_model.layers.22.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.22.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.22.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.23.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.23.attention.out.comfy_quant', 'model.diffusion_model.layers.23.attention.qkv.comfy_quant', 'model.diffusion_model.layers.23.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.23.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.23.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.24.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.24.attention.out.comfy_quant', 'model.diffusion_model.layers.24.attention.qkv.comfy_quant', 'model.diffusion_model.layers.24.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.24.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.24.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.25.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.25.attention.out.comfy_quant', 'model.diffusion_model.layers.25.attention.qkv.comfy_quant', 'model.diffusion_model.layers.25.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.25.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.25.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.26.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.26.attention.out.comfy_quant', 'model.diffusion_model.layers.26.attention.qkv.comfy_quant', 'model.diffusion_model.layers.26.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.26.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.26.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.27.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.27.attention.out.comfy_quant', 'model.diffusion_model.layers.27.attention.qkv.comfy_quant', 'model.diffusion_model.layers.27.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.27.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.27.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.28.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.28.attention.out.comfy_quant', 'model.diffusion_model.layers.28.attention.qkv.comfy_quant', 'model.diffusion_model.layers.28.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.28.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.28.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.29.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.29.attention.out.comfy_quant', 'model.diffusion_model.layers.29.attention.qkv.comfy_quant', 'model.diffusion_model.layers.29.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.29.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.29.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.3.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.3.attention.out.comfy_quant', 'model.diffusion_model.layers.3.attention.qkv.comfy_quant', 'model.diffusion_model.layers.3.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.3.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.3.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.4.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.4.attention.out.comfy_quant', 'model.diffusion_model.layers.4.attention.qkv.comfy_quant', 'model.diffusion_model.layers.4.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.4.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.4.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.5.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.5.attention.out.comfy_quant', 'model.diffusion_model.layers.5.attention.qkv.comfy_quant', 'model.diffusion_model.layers.5.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.5.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.5.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.6.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.6.attention.out.comfy_quant', 'model.diffusion_model.layers.6.attention.qkv.comfy_quant', 'model.diffusion_model.layers.6.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.6.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.6.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.7.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.7.attention.out.comfy_quant', 'model.diffusion_model.layers.7.attention.qkv.comfy_quant', 'model.diffusion_model.layers.7.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.7.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.7.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.8.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.8.attention.out.comfy_quant', 'model.diffusion_model.layers.8.attention.qkv.comfy_quant', 'model.diffusion_model.layers.8.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.8.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.8.feed_forward.w3.comfy_quant', 'model.diffusion_model.layers.9.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.layers.9.attention.out.comfy_quant', 'model.diffusion_model.layers.9.attention.qkv.comfy_quant', 'model.diffusion_model.layers.9.feed_forward.w1.comfy_quant', 'model.diffusion_model.layers.9.feed_forward.w2.comfy_quant', 'model.diffusion_model.layers.9.feed_forward.w3.comfy_quant', 'model.diffusion_model.noise_refiner.0.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.noise_refiner.0.attention.out.comfy_quant', 'model.diffusion_model.noise_refiner.0.attention.qkv.comfy_quant', 'model.diffusion_model.noise_refiner.0.feed_forward.w1.comfy_quant', 'model.diffusion_model.noise_refiner.0.feed_forward.w2.comfy_quant', 'model.diffusion_model.noise_refiner.0.feed_forward.w3.comfy_quant', 'model.diffusion_model.noise_refiner.1.adaLN_modulation.0.comfy_quant', 'model.diffusion_model.noise_refiner.1.attention.out.comfy_quant', 'model.diffusion_model.noise_refiner.1.attention.qkv.comfy_quant', 'model.diffusion_model.noise_refiner.1.feed_forward.w1.comfy_quant', 'model.diffusion_model.noise_refiner.1.feed_forward.w2.comfy_quant', 'model.diffusion_model.noise_refiner.1.feed_forward.w3.comfy_quant', 'model.diffusion_model.t_embedder.mlp.0.comfy_quant', 'model.diffusion_model.t_embedder.mlp.2.comfy_quant', 'model.diffusion_model.x_embedder.comfy_quant']

personally think this model converted this way perform better than using ComfyUI_DiffusionModel_fp8_converter and ComfyUI-SNR-quant(into total fp8) (tested, but not fully, while these two perform the same)

I don't know why, maybe because it's e4m3fn?

Thanks to @Bedovyy