I'm sure there's a specific term to describe this. I don't know it.
[DEBUG] main.cpp:549 - version: stable-diffusion.cpp version master-586-c97702e, commit c97702e
[DEBUG] main.cpp:550 - System Info:
SSE3 = 1 | AVX = 1 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | VSX = 0 |
[DEBUG] main.cpp:551 - SDCliParams {
mode: img_gen,
output_path: "output.png",
image_path: "",
metadata_format: "text",
verbose: true,
color: false,
canny_preprocess: false,
convert_name: false,
preview_method: none,
preview_interval: 1,
preview_path: "preview.png",
preview_fps: 16,
taesd_preview: false,
preview_noisy: false,
metadata_raw: false,
metadata_brief: false,
metadata_all: false
}
[DEBUG] main.cpp:552 - SDContextParams {
n_threads: 12,
model_path: "",
clip_l_path: "../clip_l.safetensors",
clip_g_path: "",
clip_vision_path: "",
t5xxl_path: "../t5xxl_fp16.safetensors",
llm_path: "",
llm_vision_path: "",
diffusion_model_path: "../flux1-dev-q3_k.gguf",
high_noise_diffusion_model_path: "",
vae_path: "../ae.safetensors",
taesd_path: "",
esrgan_path: "",
control_net_path: "",
embedding_dir: "",
embeddings: {
}
wtype: NONE,
tensor_type_rules: "",
lora_model_dir: ".",
hires_upscalers_dir: "",
photo_maker_path: "",
rng_type: cuda,
sampler_rng_type: NONE,
offload_params_to_cpu: false,
enable_mmap: false,
control_net_cpu: false,
clip_on_cpu: true,
vae_on_cpu: false,
flash_attn: false,
diffusion_flash_attn: false,
diffusion_conv_direct: false,
vae_conv_direct: false,
circular: false,
circular_x: false,
circular_y: false,
chroma_use_dit_mask: true,
qwen_image_zero_cond_t: false,
chroma_use_t5_mask: false,
chroma_t5_mask_pad: 1,
prediction: NONE,
lora_apply_mode: auto,
force_sdxl_vae_conv_scale: false
}
[DEBUG] main.cpp:553 - SDGenerationParams {
loras: "{
}",
high_noise_loras: "{
}",
prompt: "a lovely cat holding a sign says 'lol'",
negative_prompt: "",
clip_skip: -1,
width: -1,
height: -1,
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: 0, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: euler, sample_steps: 20, eta: inf, shifted_timestep: 0, flow_shift: inf),
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: 0, slg.layer_start: 0.01, slg.layer_end: 0.20, slg.scale: 0.00, scheduler: NONE, sample_method: NONE, sample_steps: 20, eta: inf, shifted_timestep: 0, flow_shift: inf),
custom_sigmas: [],
cache_mode: "",
cache_option: "",
cache: disabled (threshold=inf, 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: 42,
upscale_repeats: 1,
upscale_tile_size: 128,
hires: { enabled: false, upscaler: "Latent (nearest)", model_path: "", scale: 2, target_width: 0, target_height: 0, steps: 0, denoising_strength: 0.7, upscale_tile_size: 128 },
vae_tiling_params: { 0, 0, 0, 0.5, 0, 0 },
}
[DEBUG] stable-diffusion.cpp:184 - Using Vulkan backend
[DEBUG] ggml_extend.hpp:78 - ggml_vulkan: Found 1 Vulkan devices:
[DEBUG] ggml_extend.hpp:78 - ggml_vulkan: 0 = AMD Radeon RX 7900 XT (RADV NAVI31) (radv) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
[INFO ] stable-diffusion.cpp:205 - Vulkan: Using device 0
[INFO ] stable-diffusion.cpp:270 - loading diffusion model from '../flux1-dev-q3_k.gguf'
[INFO ] model.cpp:229 - load ../flux1-dev-q3_k.gguf using gguf format
[DEBUG] model.cpp:278 - init from '../flux1-dev-q3_k.gguf'
[INFO ] stable-diffusion.cpp:286 - loading clip_l from '../clip_l.safetensors'
[INFO ] model.cpp:232 - load ../clip_l.safetensors using safetensors format
[DEBUG] model.cpp:307 - init from '../clip_l.safetensors', prefix = 'text_encoders.clip_l.transformer.'
[INFO ] stable-diffusion.cpp:310 - loading t5xxl from '../t5xxl_fp16.safetensors'
[INFO ] model.cpp:232 - load ../t5xxl_fp16.safetensors using safetensors format
[DEBUG] model.cpp:307 - init from '../t5xxl_fp16.safetensors', prefix = 'text_encoders.t5xxl.transformer.'
[INFO ] stable-diffusion.cpp:331 - loading vae from '../ae.safetensors'
[INFO ] model.cpp:232 - load ../ae.safetensors using safetensors format
[DEBUG] model.cpp:307 - init from '../ae.safetensors', prefix = 'vae.'
[INFO ] stable-diffusion.cpp:356 - Version: Flux
[INFO ] stable-diffusion.cpp:384 - Weight type stat: f32: 720 | f16: 415 | q3_K: 304
[INFO ] stable-diffusion.cpp:385 - Conditioner weight type stat: f16: 415
[INFO ] stable-diffusion.cpp:386 - Diffusion model weight type stat: f32: 476 | q3_K: 304
[INFO ] stable-diffusion.cpp:387 - VAE weight type stat: f32: 244
[DEBUG] stable-diffusion.cpp:389 - ggml tensor size = 400 bytes
[INFO ] stable-diffusion.cpp:434 - CLIP: Using CPU backend
[DEBUG] clip_tokenizer.cpp:65 - vocab size: 49408
[INFO ] flux.hpp:1283 - flux: depth = 19, depth_single_blocks = 38, guidance_embed = true, context_in_dim = 4096, hidden_size = 3072, num_heads = 24
[DEBUG] ggml_extend.hpp:2046 - clip params backend buffer size = 235.06 MB(RAM) (196 tensors)
[DEBUG] ggml_extend.hpp:2046 - t5 params backend buffer size = 9083.77 MB(RAM) (219 tensors)
[DEBUG] ggml_extend.hpp:2046 - flux params backend buffer size = 5105.72 MB(VRAM) (780 tensors)
[INFO ] stable-diffusion.cpp:682 - using VAE for encoding / decoding
[INFO ] auto_encoder_kl.hpp:517 - vae decoder: ch = 128
[DEBUG] ggml_extend.hpp:2046 - vae params backend buffer size = 94.57 MB(VRAM) (138 tensors)
[DEBUG] stable-diffusion.cpp:806 - loading weights
[DEBUG] model.cpp:755 - using 12 threads for model loading
[DEBUG] model.cpp:777 - loading tensors from ../flux1-dev-q3_k.gguf
|===========================> | 780/1439 - 8.15GB/s
[DEBUG] model.cpp:777 - loading tensors from ../clip_l.safetensors
|=================================> | 976/1439 - 6.41GB/s
[DEBUG] model.cpp:777 - loading tensors from ../t5xxl_fp16.safetensors
|=========================================> | 1195/1439 - 11.13GB/s
[DEBUG] model.cpp:777 - loading tensors from ../ae.safetensors
|==================================================| 1439/1439 - 9.74GB/s
[INFO ] model.cpp:1006 - loading tensors completed, taking 1.47s (process: 0.00s, read: 0.55s, memcpy: 0.00s, convert: 0.01s, copy_to_backend: 0.40s)
[DEBUG] stable-diffusion.cpp:846 - finished loaded file
[INFO ] stable-diffusion.cpp:898 - total params memory size = 14519.13MB (VRAM 5200.30MB, RAM 9318.83MB): text_encoders 9318.83MB(RAM), diffusion_model 5105.72MB(VRAM), vae 94.57MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(RAM)
[INFO ] stable-diffusion.cpp:987 - running in Flux FLOW mode
[INFO ] stable-diffusion.cpp:3320 - generate_image 512x512
[INFO ] denoiser.hpp:499 - get_sigmas with discrete scheduler
[INFO ] stable-diffusion.cpp:2835 - sampling using Euler method
[DEBUG] conditioner.hpp:1157 - parse 'a lovely cat holding a sign says 'lol'' to [['a lovely cat holding a sign says 'lol'', 1], ]
[DEBUG] bpe_tokenizer.cpp:183 - split prompt "a lovely cat holding a sign says 'lol'" to tokens ["a</w>", "lovely</w>", "cat</w>", "holding</w>", "a</w>", "sign</w>", "says</w>", "'</w>", "lol</w>", "'</w>", ]
[DEBUG] t5_unigram_tokenizer.cpp:336 - split prompt "a lovely cat holding a sign says 'lol'" to tokens ["▁", "a", "▁lovely", "▁cat", "▁holding", "▁", "a", "▁sign", "▁says", "▁", "'", "l", "o", "l", "'", ]
[DEBUG] clip.hpp:318 - identity projection
[DEBUG] ggml_extend.hpp:1859 - clip compute buffer size: 1.42 MB(RAM)
[DEBUG] clip.hpp:318 - identity projection
[DEBUG] ggml_extend.hpp:1859 - t5 compute buffer size: 68.25 MB(RAM)
[DEBUG] conditioner.hpp:1272 - computing condition graph completed, taking 6017 ms
[INFO ] stable-diffusion.cpp:3189 - get_learned_condition completed, taking 6.02s
[INFO ] stable-diffusion.cpp:3354 - generating image: 1/1 - seed 42
[DEBUG] ggml_extend.hpp:1859 - flux compute buffer size: 341.50 MB(VRAM)
|==================================================| 20/20 - 1.30it/s
[INFO ] stable-diffusion.cpp:3385 - sampling completed, taking 15.37s
[INFO ] stable-diffusion.cpp:3403 - generating 1 latent images completed, taking 15.38s
[INFO ] stable-diffusion.cpp:3213 - decoding 1 latents
[DEBUG] ggml_extend.hpp:1859 - vae compute buffer size: 1984.25 MB(VRAM)
[DEBUG] vae.hpp:206 - computing vae decode graph completed, taking 2.58s
[INFO ] stable-diffusion.cpp:3229 - latent 1 decoded, taking 2.58s
[INFO ] stable-diffusion.cpp:3233 - decode_first_stage completed, taking 2.58s
[INFO ] stable-diffusion.cpp:3540 - generate_image completed in 24.17s
[INFO ] main.cpp:440 - save result image 0 to 'output.png' (success)
[INFO ] main.cpp:489 - 1/1 images saved
I tried other models. Exact same result. I don't think the issue are the models, because the image is always like the example above.
Git commit
Operating System & Version
Alpine Linux Edge
GGML backends
Vulkan
Command-line arguments used
Steps to reproduce
What you expected to happen
Image properly generated
What actually happened
The image has like, all channels separate?
I'm sure there's a specific term to describe this. I don't know it.
Logs / error messages / stack trace
Additional context / environment details
I tried other models. Exact same result. I don't think the issue are the models, because the image is always like the example above.