Modelpublic

W-Zil/Krea2_Turbo_LowVram_AllFiles_WorkFlows

sourceHugging Faceotherupdated 1mo agoView on Hugging Face

0likes110downloads

Model Card

Krea2 Turbo LowVram — All Files & Workflows

Personal collection of models and supporting files to run Krea2 Turbo (Int8) on ComfyUI on GPUs with limited VRAM (tested on 15GB).

Files

File	Type	ComfyUI folder
`krea2_turbo_int8_convrot.safetensors`	Main diffusion model (Int8)	`models/diffusion_models`
`qwen3vl-4b-q4_k_m.gguf`	Text encoder (Qwen3-VL, GGUF format)	`models/text_encoders`
`qwen3vl_4b_fp8_scaled.safetensors`	Text encoder (Qwen3-VL, fp8 format)	`models/text_encoders`
`qwen_image_vae.safetensors`	VAE	`models/vae`
`krea2_darkbrush.safetensors`	LoRA	`models/loras`

Note: the two text encoders (gguf and fp8) are alternatives — you typically only need to download one of them, depending on the loader node used (GGUF loader vs standard loader).

Workflows

Workflow	Description
image_krea2_turbo_t2i WORK_Tested.json	Standard text-to-image generation with Krea2 Turbo.
image_krea2_turbo_t2i_int8.json	Text-to-image generation using the Int8-quantized model — lower VRAM usage
image_krea2_turbo_int8_image_style_reference.json	Int8 pipeline with style reference conditioning — upload one or more reference images to guide the aesthetic/style of the output.

To use: download the JSON file and drag it directly onto the ComfyUI canvas, or load it via Workflow → Open.

Requirements

python

from huggingface_hub import hf_hub_download

hf_hub_download(
    repo_id="W-Zil/Krea2_Turbo_LowVram_AllFiles_WorkFlows",
    filename="FILE_NAME",
    local_dir="/path/to/destination",
    token="YOUR_TOKEN",
)

Status

✅ Tested and working — ComfyUI + custom nodes + models