by Bedovyy
Open source · 2k downloads · 20 likes
Anima FP8 is an AI model optimized for image generation, offering an excellent balance between speed and quality. It comes in two versions: FP8, recommended for maximizing generation speed while maintaining high quality, and NVFP4Mixed, which is slightly lighter but with marginally lower quality. This model is particularly well-suited for users seeking optimized performance on modern hardware configurations, such as RTX 5090 graphics cards. Its use cases include creating artistic images, rapidly generating visuals for creative projects or prototypes. What sets it apart is its ability to deliver reduced generation times without excessively compromising visual quality, thanks to advanced quantization techniques.
There are two models - FP8 and NVFP4Mixed.
To use torch.compile, use the TorchCompileModelAdvanced node from KJNodes, set the mode to max-autotune-no-cudagraphs, and make sure dynamic is set to false.
Tested on
--fast option, torch2.10.0+cu130| quant | none | sage+torch.compile |
|---|---|---|
| bf16 | 7.13s/4.21it/s | 5.16s/5.81it/s (+38%) |
| fp8 | 6.66s/4.50it/s (+11%) | 4.52s/6.64it/s (+58%) |
| nvfp4mix | 6.37s/4.71it/s (+12%) | 4.99s/6.01it/s (+43%) |


| quant | sample |
|---|---|
| bf16 | ![]() |
| fp8 | ![]() |
| nvfp4mixed | ![]() |
{
"format": "comfy_quant",
"block_names": ["net.blocks."],
"rules": [
{ "policy": "keep", "match": ["blocks.0", "blocks.1."] },
{ "policy": "float8_e4m3fn", "match": ["q_proj", "k_proj", "v_proj", "o_proj", "output_proj", ".mlp"] },
{ "policy": "nvfp4", "match": [] }
]
}
{
"format": "comfy_quant",
"block_names": ["net.blocks."],
"rules": [
{ "policy": "keep", "match": ["blocks.0."] },
{ "policy": "float8_e4m3fn", "match": ["v_proj", "adaln_modulation", ".mlp"] },
{ "policy": "nvfp4", "match": ["k_proj", "q_proj", "output_proj"] }
]
}