tungsten-logo
Sign In
Heavy Make up with Deliberate Cyber
deliberate-cyber:v1.0
evevalentine2017
evevalentine2017
·
Mar 24
365
Input
prompt
Specify things to see in the output
Raw intricate portrait photograph of a bold and confident tattooed brunette with a seductive facial expression, gritty, intricate details, (intricate blue-- eyes), posing in a dark and moody industrial setting, leather & lace clothing, spiked dog collar, hard lighting, (detailed skin and hair texture), highly contrasted, high-quality photo shot on iPhone
negative_prompt
Specify things to not see in the output
epiCNegative, nsfw, (nipples:1), nudity, anime, extra hand, extra leg, extra arm, extra head, extra fingers, extra body parts, bad anatomy, distorted face, extra limbs, gross proportions, malformed limbs, missing arms, missing legs, mutated hands, fused fingers, too many fingers, long neck, extra head, cloned head, extra body, cloned body, cropped, low resolution , username, blurry, trademark, watermark, title, multiple view, Reference sheet, curvy, plump, fat, strabismus, clothing cutout, side slit, out of focus, overexposed, blurry, logo, watermark, signature, jpeg artifacts, illustration, 3d, sepia, painting, cartoons, sketch, multiple pictures, fused bodies, ugly, duplicate, mutilated, out of frame, (mutation:1.3), (deformed:1.3), (blurry), (bad anatomy:1.1), (bad proportions:1.2), (long neck:1.2), (worst quality:1.4), (low quality:1.4), (monochrome:1.1)
num_outputs
Number of output images
4
width
Output image width
768
height
Output image height
768
enhance_face_with_adetailer
Enhance face with adetailer
true
enhance_hands_with_adetailer
Enhance hands with adetailer
true
adetailer_denoising_strength
1: completely redraw face or hands / 0: no effect on output images
0.45
detail
Enhance/diminish detail while keeping the overall style/character
0
brightness
Adjust brightness
0
contrast
Adjust contrast
0
saturation
Adjust saturation
0
seed
Same seed with the same prompt generates the same image. Set as -1 to randomize output.
138353287
input_image
Base image that the output should be generated from. This is useful when you want to add some detail to input_image. For example, if prompt is "sunglasses" and input_image has a man, there is the man wearing sunglasses in the output.
input_image_redrawing_strength
How differ the output is from input_image. Used only when input_image is given.
0.55
reference_image
Image with which the output should share identity (e.g. face of a person or type of a dog)
reference_image_strength
Strength of applying reference_image. Used only when reference_image is given.
1
reference_pose_image
Image with a reference pose
reference_pose_strength
Strength of applying reference_pose_image. Used only when reference_pose_image is given.
1
reference_depth_image
Image with a reference depth
reference_depth_strength
Strength of applying reference_depth_image. Used only when reference_depth_image is given.
1
sampler
Sampler type
DPM++ 3M SDE Exponential
samping_steps
Number of denoising steps
80
cfg_scale
Scale for classifier-free guidance
7
clip_skip
The number of last layers of CLIP network to skip
2
vae
Select VAE
vae-ft-mse-840000-ema-pruned_fp16.safetensors
lora_1
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
FaceBeauty_qinglong_V3.safetensors
lora_2
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
bimbostyleFourUDel.safetensors
lora_3
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
embedding_1
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
epiCNegative.pt
embedding_2
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
embedding_3
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
disable_prompt_modification
Disable automatically adding suggested prompt modification. Built-in LoRAs and trigger words will remain.
false
Output
https://files.tungsten.run/uploads/526054d56c09466eab55f147569218af/00000-138353287.webp
https://files.tungsten.run/uploads/b0eb970e0e5c4204883246c3cf5926af/00001-138353288.webp
https://files.tungsten.run/uploads/bf9525de5a554325b12d593fd0fdd100/00002-138353289.webp
https://files.tungsten.run/uploads/45ec4d4ecf0c48988c4ccabf610f6c01/00003-138353290.webp
Finished in 203.3 seconds
Setting up the model... Preparing inputs... Processing... Loading VAE weight: models/VAE/vae-ft-mse-840000-ema-pruned_fp16.safetensors Full prompt: Raw intricate portrait photograph of a bold and confident tattooed brunette with a seductive facial expression, gritty, intricate details, (intricate blue-- eyes), posing in a dark and moody industrial setting, leather & lace clothing, spiked dog collar, hard lighting, (detailed skin and hair texture), highly contrasted, high-quality photo shot on iPhone Full negative prompt: epiCNegative, nsfw, (nipples:1), nudity, anime, extra hand, extra leg, extra arm, extra head, extra fingers, extra body parts, bad anatomy, distorted face, extra limbs, gross proportions, malformed limbs, missing arms, missing legs, mutated hands, fused fingers, too many fingers, long neck, extra head, cloned head, extra body, cloned body, cropped, low resolution , username, blurry, trademark, watermark, title, multiple view, Reference sheet, curvy, plump, fat, strabismus, clothing cutout, side slit, out of focus, overexposed, blurry, logo, watermark, signature, jpeg artifacts, illustration, 3d, sepia, painting, cartoons, sketch, multiple pictures, fused bodies, ugly, duplicate, mutilated, out of frame, (mutation:1.3), (deformed:1.3), (blurry), (bad anatomy:1.1), (bad proportions:1.2), (long neck:1.2), (worst quality:1.4), (low quality:1.4), (monochrome:1.1) 0%| | 0/80 [00:00<?, ?it/s] 1%|▏ | 1/80 [00:01<01:48, 1.37s/it] 2%|▎ | 2/80 [00:03<02:05, 1.60s/it] 4%|▍ | 3/80 [00:05<02:13, 1.74s/it] 5%|▌ | 4/80 [00:06<02:16, 1.80s/it] 6%|▋ | 5/80 [00:08<02:12, 1.76s/it] 8%|▊ | 6/80 [00:10<02:10, 1.77s/it] 9%|▉ | 7/80 [00:12<02:11, 1.80s/it] 10%|█ | 8/80 [00:14<02:09, 1.80s/it] 11%|█▏ | 9/80 [00:15<02:09, 1.82s/it] 12%|█▎ | 10/80 [00:17<02:06, 1.81s/it] 14%|█▍ | 11/80 [00:20<02:25, 2.11s/it] 15%|█▌ | 12/80 [00:23<02:31, 2.23s/it] 16%|█▋ | 13/80 [00:24<02:20, 2.10s/it] 18%|█▊ | 14/80 [00:26<02:10, 1.98s/it] 19%|█▉ | 15/80 [00:28<02:03, 1.90s/it] 20%|██ | 16/80 [00:30<02:01, 1.90s/it] 21%|██▏ | 17/80 [00:31<01:55, 1.84s/it] 22%|██▎ | 18/80 [00:33<01:53, 1.83s/it] 24%|██▍ | 19/80 [00:35<01:52, 1.84s/it] 25%|██▌ | 20/80 [00:37<01:48, 1.82s/it] 26%|██▋ | 21/80 [00:38<01:43, 1.76s/it] 28%|██▊ | 22/80 [00:40<01:40, 1.74s/it] 29%|██▉ | 23/80 [00:42<01:37, 1.72s/it] 30%|███ | 24/80 [00:44<01:38, 1.75s/it] 31%|███▏ | 25/80 [00:45<01:35, 1.73s/it] 32%|███▎ | 26/80 [00:47<01:35, 1.76s/it] 34%|███▍ | 27/80 [00:49<01:35, 1.79s/it] 35%|███▌ | 28/80 [00:51<01:31, 1.77s/it] 36%|███▋ | 29/80 [00:52<01:28, 1.74s/it] 38%|███▊ | 30/80 [00:54<01:26, 1.73s/it] 39%|███▉ | 31/80 [00:56<01:24, 1.72s/it] 40%|████ | 32/80 [00:57<01:20, 1.68s/it] 41%|████▏ | 33/80 [00:59<01:18, 1.68s/it] 42%|████▎ | 34/80 [01:01<01:16, 1.67s/it] 44%|████▍ | 35/80 [01:02<01:16, 1.70s/it] 45%|████▌ | 36/80 [01:04<01:16, 1.75s/it] 46%|████▋ | 37/80 [01:06<01:10, 1.65s/it] 48%|████▊ | 38/80 [01:07<01:06, 1.58s/it] 49%|████▉ | 39/80 [01:09<01:04, 1.58s/it] 50%|█████ | 40/80 [01:11<01:06, 1.65s/it] 51%|█████▏ | 41/80 [01:12<01:03, 1.62s/it] 52%|█████▎ | 42/80 [01:14<01:01, 1.62s/it] 54%|█████▍ | 43/80 [01:15<00:59, 1.61s/it] 55%|█████▌ | 44/80 [01:17<00:57, 1.59s/it] 56%|█████▋ | 45/80 [01:18<00:56, 1.61s/it] 57%|█████▊ | 46/80 [01:20<00:55, 1.62s/it] 59%|█████▉ | 47/80 [01:22<00:52, 1.61s/it] 60%|██████ | 48/80 [01:23<00:52, 1.65s/it] 61%|██████▏ | 49/80 [01:25<00:49, 1.61s/it] 62%|██████▎ | 50/80 [01:27<00:49, 1.64s/it] 64%|██████▍ | 51/80 [01:28<00:45, 1.59s/it] 65%|██████▌ | 52/80 [01:30<00:43, 1.55s/it] 66%|██████▋ | 53/80 [01:31<00:43, 1.61s/it] 68%|██████▊ | 54/80 [01:33<00:42, 1.62s/it] 69%|██████▉ | 55/80 [01:34<00:39, 1.57s/it] 70%|███████ | 56/80 [01:36<00:39, 1.65s/it] 71%|███████▏ | 57/80 [01:38<00:35, 1.56s/it] 72%|███████▎ | 58/80 [01:39<00:33, 1.53s/it] 74%|███████▍ | 59/80 [01:40<00:30, 1.46s/it] 75%|███████▌ | 60/80 [01:42<00:29, 1.49s/it] 76%|███████▋ | 61/80 [01:43<00:27, 1.44s/it] 78%|███████▊ | 62/80 [01:45<00:26, 1.49s/it] 79%|███████▉ | 63/80 [01:46<00:25, 1.49s/it] 80%|████████ | 64/80 [01:48<00:23, 1.48s/it] 81%|████████▏ | 65/80 [01:49<00:22, 1.53s/it] 82%|████████▎ | 66/80 [01:51<00:20, 1.47s/it] 84%|████████▍ | 67/80 [01:52<00:18, 1.45s/it] 85%|████████▌ | 68/80 [01:53<00:16, 1.40s/it] 86%|████████▋ | 69/80 [01:55<00:16, 1.47s/it] 88%|████████▊ | 70/80 [01:57<00:14, 1.45s/it] 89%|████████▉ | 71/80 [01:58<00:13, 1.45s/it] 90%|█████████ | 72/80 [01:59<00:11, 1.45s/it] 91%|█████████▏| 73/80 [02:01<00:09, 1.36s/it] 92%|█████████▎| 74/80 [02:02<00:07, 1.32s/it] 94%|█████████▍| 75/80 [02:03<00:07, 1.40s/it] 95%|█████████▌| 76/80 [02:05<00:05, 1.36s/it] 96%|█████████▋| 77/80 [02:06<00:04, 1.40s/it] 98%|█████████▊| 78/80 [02:07<00:02, 1.36s/it] 99%|█████████▉| 79/80 [02:08<00:01, 1.17s/it] 100%|██████████| 80/80 [02:09<00:00, 1.00s/it] 100%|██████████| 80/80 [02:09<00:00, 1.62s/it] Decoding latents in cuda:0... done in 0.99s Move latents to cpu... done in 0.02s 0: 640x640 1 face, 8.3ms Speed: 3.2ms preprocess, 8.3ms inference, 23.7ms postprocess per image at shape (1, 3, 640, 640)
prompt
Specify things to see in the output
Raw intricate portrait photograph of a bold and confident tattooed brunette with a seductive facial expression, gritty, intricate details, (intricate blue-- eyes), posing in a dark and moody industrial setting, leather & lace clothing, spiked dog collar, hard lighting, (detailed skin and hair texture), highly contrasted, high-quality photo shot on iPhone
negative_prompt
Specify things to not see in the output
epiCNegative, nsfw, (nipples:1), nudity, anime, extra hand, extra leg, extra arm, extra head, extra fingers, extra body parts, bad anatomy, distorted face, extra limbs, gross proportions, malformed limbs, missing arms, missing legs, mutated hands, fused fingers, too many fingers, long neck, extra head, cloned head, extra body, cloned body, cropped, low resolution , username, blurry, trademark, watermark, title, multiple view, Reference sheet, curvy, plump, fat, strabismus, clothing cutout, side slit, out of focus, overexposed, blurry, logo, watermark, signature, jpeg artifacts, illustration, 3d, sepia, painting, cartoons, sketch, multiple pictures, fused bodies, ugly, duplicate, mutilated, out of frame, (mutation:1.3), (deformed:1.3), (blurry), (bad anatomy:1.1), (bad proportions:1.2), (long neck:1.2), (worst quality:1.4), (low quality:1.4), (monochrome:1.1)
num_outputs
Number of output images
4
width
Output image width
768
height
Output image height
768
enhance_face_with_adetailer
Enhance face with adetailer
true
enhance_hands_with_adetailer
Enhance hands with adetailer
true
adetailer_denoising_strength
1: completely redraw face or hands / 0: no effect on output images
0.45
detail
Enhance/diminish detail while keeping the overall style/character
0
brightness
Adjust brightness
0
contrast
Adjust contrast
0
saturation
Adjust saturation
0
seed
Same seed with the same prompt generates the same image. Set as -1 to randomize output.
138353287
input_image
Base image that the output should be generated from. This is useful when you want to add some detail to input_image. For example, if prompt is "sunglasses" and input_image has a man, there is the man wearing sunglasses in the output.
input_image_redrawing_strength
How differ the output is from input_image. Used only when input_image is given.
0.55
reference_image
Image with which the output should share identity (e.g. face of a person or type of a dog)
reference_image_strength
Strength of applying reference_image. Used only when reference_image is given.
1
reference_pose_image
Image with a reference pose
reference_pose_strength
Strength of applying reference_pose_image. Used only when reference_pose_image is given.
1
reference_depth_image
Image with a reference depth
reference_depth_strength
Strength of applying reference_depth_image. Used only when reference_depth_image is given.
1
sampler
Sampler type
DPM++ 3M SDE Exponential
samping_steps
Number of denoising steps
80
cfg_scale
Scale for classifier-free guidance
7
clip_skip
The number of last layers of CLIP network to skip
2
vae
Select VAE
vae-ft-mse-840000-ema-pruned_fp16.safetensors
lora_1
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
FaceBeauty_qinglong_V3.safetensors
lora_2
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
bimbostyleFourUDel.safetensors
lora_3
LoRA file. Apply by writing the following in prompt: <lora:FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE>
embedding_1
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
epiCNegative.pt
embedding_2
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
embedding_3
Embedding file (textural inversion). Apply by writing the following in prompt or negative prompt: (FILE_NAME_WITHOUT_EXTENSION:MAGNITUDE)
disable_prompt_modification
Disable automatically adding suggested prompt modification. Built-in LoRAs and trigger words will remain.
false
https://files.tungsten.run/uploads/526054d56c09466eab55f147569218af/00000-138353287.webp
https://files.tungsten.run/uploads/b0eb970e0e5c4204883246c3cf5926af/00001-138353288.webp
https://files.tungsten.run/uploads/bf9525de5a554325b12d593fd0fdd100/00002-138353289.webp
https://files.tungsten.run/uploads/45ec4d4ecf0c48988c4ccabf610f6c01/00003-138353290.webp
Finished in 203.3 seconds
Setting up the model... Preparing inputs... Processing... Loading VAE weight: models/VAE/vae-ft-mse-840000-ema-pruned_fp16.safetensors Full prompt: Raw intricate portrait photograph of a bold and confident tattooed brunette with a seductive facial expression, gritty, intricate details, (intricate blue-- eyes), posing in a dark and moody industrial setting, leather & lace clothing, spiked dog collar, hard lighting, (detailed skin and hair texture), highly contrasted, high-quality photo shot on iPhone Full negative prompt: epiCNegative, nsfw, (nipples:1), nudity, anime, extra hand, extra leg, extra arm, extra head, extra fingers, extra body parts, bad anatomy, distorted face, extra limbs, gross proportions, malformed limbs, missing arms, missing legs, mutated hands, fused fingers, too many fingers, long neck, extra head, cloned head, extra body, cloned body, cropped, low resolution , username, blurry, trademark, watermark, title, multiple view, Reference sheet, curvy, plump, fat, strabismus, clothing cutout, side slit, out of focus, overexposed, blurry, logo, watermark, signature, jpeg artifacts, illustration, 3d, sepia, painting, cartoons, sketch, multiple pictures, fused bodies, ugly, duplicate, mutilated, out of frame, (mutation:1.3), (deformed:1.3), (blurry), (bad anatomy:1.1), (bad proportions:1.2), (long neck:1.2), (worst quality:1.4), (low quality:1.4), (monochrome:1.1) 0%| | 0/80 [00:00<?, ?it/s] 1%|▏ | 1/80 [00:01<01:48, 1.37s/it] 2%|▎ | 2/80 [00:03<02:05, 1.60s/it] 4%|▍ | 3/80 [00:05<02:13, 1.74s/it] 5%|▌ | 4/80 [00:06<02:16, 1.80s/it] 6%|▋ | 5/80 [00:08<02:12, 1.76s/it] 8%|▊ | 6/80 [00:10<02:10, 1.77s/it] 9%|▉ | 7/80 [00:12<02:11, 1.80s/it] 10%|█ | 8/80 [00:14<02:09, 1.80s/it] 11%|█▏ | 9/80 [00:15<02:09, 1.82s/it] 12%|█▎ | 10/80 [00:17<02:06, 1.81s/it] 14%|█▍ | 11/80 [00:20<02:25, 2.11s/it] 15%|█▌ | 12/80 [00:23<02:31, 2.23s/it] 16%|█▋ | 13/80 [00:24<02:20, 2.10s/it] 18%|█▊ | 14/80 [00:26<02:10, 1.98s/it] 19%|█▉ | 15/80 [00:28<02:03, 1.90s/it] 20%|██ | 16/80 [00:30<02:01, 1.90s/it] 21%|██▏ | 17/80 [00:31<01:55, 1.84s/it] 22%|██▎ | 18/80 [00:33<01:53, 1.83s/it] 24%|██▍ | 19/80 [00:35<01:52, 1.84s/it] 25%|██▌ | 20/80 [00:37<01:48, 1.82s/it] 26%|██▋ | 21/80 [00:38<01:43, 1.76s/it] 28%|██▊ | 22/80 [00:40<01:40, 1.74s/it] 29%|██▉ | 23/80 [00:42<01:37, 1.72s/it] 30%|███ | 24/80 [00:44<01:38, 1.75s/it] 31%|███▏ | 25/80 [00:45<01:35, 1.73s/it] 32%|███▎ | 26/80 [00:47<01:35, 1.76s/it] 34%|███▍ | 27/80 [00:49<01:35, 1.79s/it] 35%|███▌ | 28/80 [00:51<01:31, 1.77s/it] 36%|███▋ | 29/80 [00:52<01:28, 1.74s/it] 38%|███▊ | 30/80 [00:54<01:26, 1.73s/it] 39%|███▉ | 31/80 [00:56<01:24, 1.72s/it] 40%|████ | 32/80 [00:57<01:20, 1.68s/it] 41%|████▏ | 33/80 [00:59<01:18, 1.68s/it] 42%|████▎ | 34/80 [01:01<01:16, 1.67s/it] 44%|████▍ | 35/80 [01:02<01:16, 1.70s/it] 45%|████▌ | 36/80 [01:04<01:16, 1.75s/it] 46%|████▋ | 37/80 [01:06<01:10, 1.65s/it] 48%|████▊ | 38/80 [01:07<01:06, 1.58s/it] 49%|████▉ | 39/80 [01:09<01:04, 1.58s/it] 50%|█████ | 40/80 [01:11<01:06, 1.65s/it] 51%|█████▏ | 41/80 [01:12<01:03, 1.62s/it] 52%|█████▎ | 42/80 [01:14<01:01, 1.62s/it] 54%|█████▍ | 43/80 [01:15<00:59, 1.61s/it] 55%|█████▌ | 44/80 [01:17<00:57, 1.59s/it] 56%|█████▋ | 45/80 [01:18<00:56, 1.61s/it] 57%|█████▊ | 46/80 [01:20<00:55, 1.62s/it] 59%|█████▉ | 47/80 [01:22<00:52, 1.61s/it] 60%|██████ | 48/80 [01:23<00:52, 1.65s/it] 61%|██████▏ | 49/80 [01:25<00:49, 1.61s/it] 62%|██████▎ | 50/80 [01:27<00:49, 1.64s/it] 64%|██████▍ | 51/80 [01:28<00:45, 1.59s/it] 65%|██████▌ | 52/80 [01:30<00:43, 1.55s/it] 66%|██████▋ | 53/80 [01:31<00:43, 1.61s/it] 68%|██████▊ | 54/80 [01:33<00:42, 1.62s/it] 69%|██████▉ | 55/80 [01:34<00:39, 1.57s/it] 70%|███████ | 56/80 [01:36<00:39, 1.65s/it] 71%|███████▏ | 57/80 [01:38<00:35, 1.56s/it] 72%|███████▎ | 58/80 [01:39<00:33, 1.53s/it] 74%|███████▍ | 59/80 [01:40<00:30, 1.46s/it] 75%|███████▌ | 60/80 [01:42<00:29, 1.49s/it] 76%|███████▋ | 61/80 [01:43<00:27, 1.44s/it] 78%|███████▊ | 62/80 [01:45<00:26, 1.49s/it] 79%|███████▉ | 63/80 [01:46<00:25, 1.49s/it] 80%|████████ | 64/80 [01:48<00:23, 1.48s/it] 81%|████████▏ | 65/80 [01:49<00:22, 1.53s/it] 82%|████████▎ | 66/80 [01:51<00:20, 1.47s/it] 84%|████████▍ | 67/80 [01:52<00:18, 1.45s/it] 85%|████████▌ | 68/80 [01:53<00:16, 1.40s/it] 86%|████████▋ | 69/80 [01:55<00:16, 1.47s/it] 88%|████████▊ | 70/80 [01:57<00:14, 1.45s/it] 89%|████████▉ | 71/80 [01:58<00:13, 1.45s/it] 90%|█████████ | 72/80 [01:59<00:11, 1.45s/it] 91%|█████████▏| 73/80 [02:01<00:09, 1.36s/it] 92%|█████████▎| 74/80 [02:02<00:07, 1.32s/it] 94%|█████████▍| 75/80 [02:03<00:07, 1.40s/it] 95%|█████████▌| 76/80 [02:05<00:05, 1.36s/it] 96%|█████████▋| 77/80 [02:06<00:04, 1.40s/it] 98%|█████████▊| 78/80 [02:07<00:02, 1.36s/it] 99%|█████████▉| 79/80 [02:08<00:01, 1.17s/it] 100%|██████████| 80/80 [02:09<00:00, 1.00s/it] 100%|██████████| 80/80 [02:09<00:00, 1.62s/it] Decoding latents in cuda:0... done in 0.99s Move latents to cpu... done in 0.02s 0: 640x640 1 face, 8.3ms Speed: 3.2ms preprocess, 8.3ms inference, 23.7ms postprocess per image at shape (1, 3, 640, 640)