[Bug] LoRA getting loaded four times in a row with z-image-turbo. #1071
Description
Git commit
Operating System & Version
Windows 11
GGML backends
HIP
Command-line arguments used
-t 6 --diffusion-model "\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf" --llm "\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf" --vae "\Models\vae\flux_ae.safetensors" --embd-dir "\Models\embeddings" --lora-model-dir "\Models\LoRa" -o "\1765311742.png" -p "lora:z-image\characters\ShadowheartZIT_byKonan:1 Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a forest." -n ", text, english text, web adress, censored, patreon username, multiple views" --cfg-scale 1 --scheduler discrete --sampling-method euler --steps 20 -H 1536 -W 832 --seed 1765311742 -b 1 --clip-skip 0 --easycache 0.2,0.15,0.95 --diffusion-fa --color --verbose
Steps to reproduce
Load a LoRA, with any z-image-turbo models.
What you expected to happen
The LoRA gets loaded one time.
What actually happened
The LoRA gets loaded four times.
Logs / error messages / stack trace
Full log
mode: img_gen,
output_path: "<redacted>\1765311843_Z-Image-Turbo_none_832x1536_1765311843.png",
verbose: true,
color: true,
canny_preprocess: false,
preview_method: none,
preview_interval: 1,
preview_path: "preview.png",
preview_fps: 16,
taesd_preview: false,
preview_noisy: false
}
SDContextParams {
n_threads: 6,
model_path: "",
clip_l_path: "",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "",
llm_path: "<redacted>\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf",
llm_vision_path: "",
diffusion_model_path: "<redacted>\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf",
high_noise_diffusion_model_path: "",
vae_path: "<redacted>\Models\vae\flux_ae.safetensors",
taesd_path: "",
esrgan_path: "",
control_net_path: "",
embedding_dir: "<redacted>\Models\embeddings",
wtype: NONE,
tensor_type_rules: "",
lora_model_dir: "<redacted>\Models\LoRa",
photo_maker_path: "",
rng_type: cuda,
sampler_rng_type: NONE,
flow_shift: INF
offload_params_to_cpu: false,
control_net_cpu: false,
clip_on_cpu: false,
vae_on_cpu: false,
diffusion_flash_attn: true,
diffusion_conv_direct: false,
vae_conv_direct: false,
chroma_use_dit_mask: true,
chroma_use_t5_mask: false,
chroma_t5_mask_pad: 1,
prediction: NONE,
lora_apply_mode: auto,
vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
force_sdxl_vae_conv_scale: false
}
SDGenerationParams {
prompt: "<lora:z-image\characters\ShadowheartZIT_byKonan:1> Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a forest.",
negative_prompt: ", text, english text, web adress, censored, patreon username, multiple views",
clip_skip: 0,
width: 832,
height: 1536,
batch_count: 1,
init_image_path: "",
end_image_path: "",
mask_image_path: "",
control_image_path: "",
ref_image_paths: [],
control_video_path: "",
auto_resize_ref_image: true,
increase_ref_index: false,
pm_id_images_dir: "",
pm_id_embed_path: "",
pm_style_strength: 20,
skip_layers: [7, 8, 9],
sample_params: (txt_cfg: 1.00, img_cfg: 1.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: discrete, sample_method: euler, sample_steps: 20, eta: 0.00, shifted_timestep: 0),
high_noise_skip_layers: [7, 8, 9],
high_noise_sample_params: (txt_cfg: 7.00, img_cfg: 7.00, distilled_guidance: 3.50, slg.layer_count: 3, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE,
sample_method: NONE, sample_steps: 20, eta: 0.00, shifted_timestep: 0),
easycache_option: "0.2,0.15,0.95",
easycache: enabled (threshold=0.2, start=0.15, end=0.95),
moe_boundary: 0.875,
video_frames: 1,
fps: 16,
vace_strength: 1,
strength: 0.75,
control_strength: 0.9,
seed: 1765311843,
upscale_repeats: 1,
}
[DEBUG] stable-diffusion.cpp:160 - Using CUDA backend
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:69 - ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:69 - ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:69 - ggml_cuda_init: found 1 ROCm devices:
[INFO ] stable-diffusion.cpp\ggml_extend.hpp:69 - Device 0: AMD Radeon RX 7900 XT, gfx1100 (0x1100), VMM: no, Wave Size: 32
[INFO ] stable-diffusion.cpp:235 - loading diffusion model from '<redacted>\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf'
[INFO ] model.cpp:370 - load <redacted>\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf using gguf format
[DEBUG] model.cpp:412 - init from '<redacted>\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf'
[INFO ] stable-diffusion.cpp:282 - loading llm from '<redacted>\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf'
[INFO ] model.cpp:370 - load <redacted>\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf using gguf format
[DEBUG] model.cpp:412 - init from '<redacted>\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf'
[INFO ] stable-diffusion.cpp:296 - loading vae from '<redacted>\Models\vae\flux_ae.safetensors'
[INFO ] model.cpp:373 - load <redacted>\Models\vae\flux_ae.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from '<redacted>\Models\vae\flux_ae.safetensors', prefix = 'vae.'
[INFO ] stable-diffusion.cpp:312 - Version: Z-Image
[INFO ] stable-diffusion.cpp:340 - Weight type stat: f32: 640 | q8_0: 455
[INFO ] stable-diffusion.cpp:341 - Conditioner weight type stat: f32: 145 | q8_0: 253
[INFO ] stable-diffusion.cpp:342 - Diffusion model weight type stat: f32: 251 | q8_0: 202
[INFO ] stable-diffusion.cpp:343 - VAE weight type stat: f32: 244
[DEBUG] stable-diffusion.cpp:345 - ggml tensor size = 400 bytes
[DEBUG] stable-diffusion.cpp\llm.hpp:285 - merges size 151387
[DEBUG] stable-diffusion.cpp\llm.hpp:317 - vocab size: 151669
[DEBUG] stable-diffusion.cpp\llm.hpp:1134 - llm: num_layers = 36, vocab_size = 151936, hidden_size = 2560, intermediate_size = 9728
[INFO ] stable-diffusion.cpp:536 - Using flash attention in the diffusion model
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1883 - qwen3 params backend buffer size = 4076.43 MB(VRAM) (398 tensors)
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1883 - z_image params backend buffer size = 6272.77 MB(VRAM) (453 tensors)
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1883 - vae params backend buffer size = 94.57 MB(VRAM) (138 tensors)
[DEBUG] stable-diffusion.cpp:684 - loading weights
[DEBUG] model.cpp:1351 - using 6 threads for model loading
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\checkpoints\z-image\z_image_turbo.q8_0.gguf
|====================> | 453/1095 - 105.94it/s
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\text_encoders\Qwen3-4B-Instruct-2507.q8_0.gguf
|======================================> | 851/1095 - 116.23it/s
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\vae\flux_ae.safetensors
|==================================================| 1095/1095 - 145.23it/s
[INFO ] model.cpp:1580 - loading tensors completed, taking 7.55s (process: 0.00s, read: 5.24s, memcpy: 0.00s, convert: 0.02s, copy_to_backend: 1.49s)
[DEBUG] stable-diffusion.cpp:716 - finished loaded file
[INFO ] stable-diffusion.cpp:788 - total params memory size = 10443.77MB (VRAM 10443.77MB, RAM 0.00MB): text_encoders 4076.43MB(VRAM), diffusion_model 6272.77MB(VRAM), vae 94.57MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)
[INFO ] stable-diffusion.cpp:856 - running in FLOW mode
[DEBUG] stable-diffusion.cpp:3152 - generate_image 832x1536
[INFO ] stable-diffusion.cpp:3183 - sampling using Euler method
[INFO ] stable-diffusion.cpp\denoiser.hpp:364 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:3296 - TXT2IMG
[DEBUG] stable-diffusion.cpp:1147 - lora z-image\characters\ShadowheartZIT_byKonan:1.00
[INFO ] stable-diffusion.cpp:1016 - apply lora at runtime
[INFO ] model.cpp:373 - load <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors', prefix = 'lora.'
[INFO ] stable-diffusion.cpp\lora.hpp:40 - loading LoRA from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors'
[DEBUG] model.cpp:1351 - using 6 threads for model loading
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors
|==================================================| 480/480 - 2242.99it/s
[INFO ] model.cpp:1580 - loading tensors completed, taking 0.22s (process: 0.00s, read: 0.00s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.00s)
[INFO ] model.cpp:373 - load <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors', prefix = 'lora.'
[INFO ] stable-diffusion.cpp\lora.hpp:40 - loading LoRA from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors'
[DEBUG] model.cpp:1351 - using 6 threads for model loading
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors
|==================================================| 480/480 - 2388.06it/s
[INFO ] model.cpp:1580 - loading tensors completed, taking 0.20s (process: 0.00s, read: 0.00s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.00s)
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1883 - lora params backend buffer size = 162.19 MB(VRAM) (480 tensors)
[DEBUG] model.cpp:1351 - using 6 threads for model loading
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors
|==================================================| 480/480 - 2296.65it/s
[INFO ] model.cpp:1580 - loading tensors completed, taking 0.21s (process: 0.00s, read: 0.07s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.09s)
[DEBUG] stable-diffusion.cpp\lora.hpp:93 - finished loaded lora
[INFO ] model.cpp:373 - load <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors using safetensors format
[DEBUG] model.cpp:503 - init from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors', prefix = 'lora.'
[INFO ] stable-diffusion.cpp\lora.hpp:40 - loading LoRA from '<redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors'
[DEBUG] model.cpp:1351 - using 6 threads for model loading
[DEBUG] model.cpp:1373 - loading tensors from <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors
|==================================================| 480/480 - 2388.06it/s
[INFO ] model.cpp:1580 - loading tensors completed, taking 0.20s (process: 0.00s, read: 0.00s, memcpy: 0.00s, convert: 0.00s, copy_to_backend: 0.00s)
[INFO ] stable-diffusion.cpp:1157 - apply_loras completed, taking 0.89s
[DEBUG] stable-diffusion.cpp:1158 - prompt after extract and remove lora: " Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a forest."
[DEBUG] stable-diffusion.cpp\conditioner.hpp:1701 - parse '<|im_start|>user
Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a forest.<|im_end|>
<|im_start|>assistant
' to [['<|im_start|>user
', 1], [' Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a
forest.', 1], ['<|im_end|>
<|im_start|>assistant
', 1], ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt "<|im_start|>user
" to tokens ["<|im_start|>", "user", "─è", ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt " Shadowheart, ultra realistic photo of an elf with black hair, a ponytail and green eyes. She is wearing a circlet and a dark dress with white patterns. She is standing in a forest." to tokens ["─áShadow", "heart", ",", "─áultra", "─árealistic", "─áphoto", "─áof", "─áan", "─áelf", "─áwith", "─áblack", "─áhair", ",", "─áa", "─ápony", "tail", "─áand", "─ágreen", "─áeyes", ".", "─áShe", "─áis", "─áwearing", "─áa", "─ácirc", "let", "─áand", "─áa", "─ádark", "─ádress", "─áwith", "─áwhite", "─ápatterns", ".", "─áShe", "─áis", "─ástanding", "─áin", "─áa", "─áforest", ".", ]
[DEBUG] stable-diffusion.cpp\llm.hpp:259 - split prompt "<|im_end|>
<|im_start|>assistant
" to tokens ["<|im_end|>", "─è", "<|im_start|>", "assistant", "─è", ]
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1697 - qwen3 compute buffer size: 5.36 MB(VRAM)
[DEBUG] stable-diffusion.cpp\conditioner.hpp:1914 - computing condition graph completed, taking 371 ms
[INFO ] stable-diffusion.cpp:2927 - get_learned_condition completed, taking 1263 ms
[INFO ] stable-diffusion.cpp:3038 - generating image: 1/1 - seed 1765311843
[INFO ] stable-diffusion.cpp:1519 - EasyCache enabled - threshold: 0.200, start_percent: 0.15, end_percent: 0.95
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1697 - z_image compute buffer size: 896.10 MB(VRAM)
|==================================================| 20/20 - 1.81s/it
[INFO ] stable-diffusion.cpp:1862 - EasyCache skipped 8/20 steps (1.67x estimated speedup)
[INFO ] stable-diffusion.cpp:3080 - sampling completed, taking 36.46s
[INFO ] stable-diffusion.cpp:3091 - generating 1 latent images completed, taking 36.61s
[INFO ] stable-diffusion.cpp:3094 - decoding 1 latents
[DEBUG] stable-diffusion.cpp\ggml_extend.hpp:1697 - vae compute buffer size: 8113.22 MB(VRAM)
[DEBUG] stable-diffusion.cpp:2297 - computing vae decode graph completed, taking 1.37s
[INFO ] stable-diffusion.cpp:3104 - latent 1 decoded, taking 1.37s
[INFO ] stable-diffusion.cpp:3108 - decode_first_stage completed, taking 1.37s
[INFO ] stable-diffusion.cpp:1128 - diffusion_lora_models:
[INFO ] stable-diffusion.cpp\lora.hpp:724 - (480 / 480) LoRA tensors have been applied, lora_file_path = <redacted>\Models\LoRa/z-image\characters\ShadowheartZIT_byKonan.safetensors
[INFO ] stable-diffusion.cpp:3404 - generate_image completed in 39.26s
Additional context / environment details
Also tested with Chroma and SDXL, can only reproduce with z-image-turbo.
Tested with multiples LoRAs.