| [ | |
| { | |
| "index": 0, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a polar bear standing on iceberg", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/13_polar bear.jpg", | |
| "caption": "a polar bear" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "polar bear", | |
| "polar bear" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 1, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a sea turtle swimming in the sea.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/18_sea turtle.jpg", | |
| "caption": "a sea turtle" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "sea turtle", | |
| "sea turtle" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 2, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a raccoon standing in a forest\n", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/08_raccoon.jpg", | |
| "caption": "a raccoon" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "raccoon", | |
| "raccoon" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 3, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a dog wearing a red collar running", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/12_dog.jpg", | |
| "caption": "a dog" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "dog", | |
| "dog" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 4, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a sphynx cat sitting on a sofa", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/25_Sphynx cat.jpg", | |
| "caption": "a sphynx cat" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "sphynx cat", | |
| "sphynx cat" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 5, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a colorful butterfly flying in the garden", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/15_butterfly.jpg", | |
| "caption": "a butterfly" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "butterfly", | |
| "butterfly" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 6, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a grasshopper jumping on the grass", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/29_grasshopper.jpg", | |
| "caption": "a grasshopper" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "grasshopper", | |
| "grasshopper" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 7, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a deer standing in a forest", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/31_deer.jpg", | |
| "caption": "a deer" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "deer", | |
| "deer" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 8, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a dolphin jumping out of the water", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/39_dolphin.jpg", | |
| "caption": "a dolphin" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "dolphin", | |
| "dolphin" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 9, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute kitten wearing a bowtie", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/02_kitten.jpg", | |
| "caption": "a kitten" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "kitten", | |
| "kitten" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 10, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "an elephant standing in the savannah", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/38_elephant.jpg", | |
| "caption": "an elephant" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "elephant", | |
| "elephant" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 11, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a shark swimming in the blue sea", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/06_shark.jpg", | |
| "caption": "a shark" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "shark", | |
| "shark" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 12, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "an eagle flying in the blue sky", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/14_eagle.jpg", | |
| "caption": "an eagle" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "eagle", | |
| "eagle" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 13, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a panda sitting on a bamboo mat", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/33_panda.jpg", | |
| "caption": "a panda" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "panda", | |
| "panda" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 14, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a rooster standing on a wooden fence", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/09_rooster.jpg", | |
| "caption": "a rooster" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "rooster", | |
| "rooster" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 15, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute kitten sitting on a mat.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/00_kitten.jpg", | |
| "caption": "a kitten" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "kitten", | |
| "kitten" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 16, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a lizard on a rocky background", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/36_lizard.jpg", | |
| "caption": "a lizard" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "lizard", | |
| "lizard" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 17, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a penguin standing on an iceberg", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/05_penguin.jpg", | |
| "caption": "a penguin" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "penguin", | |
| "penguin" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 18, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a white tiger standing in the grass.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/40_white tiger.jpg", | |
| "caption": "a white tiger" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "white tiger", | |
| "white tiger" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 19, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a french bulldog in a cozy sweater", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/43_French bulldog.jpg", | |
| "caption": "a french bulldog" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "french bulldog", | |
| "french bulldog" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 20, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute hamster wearing a tiny hat", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/20_hamster.jpg", | |
| "caption": "a hamster" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "hamster", | |
| "hamster" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 21, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a crab walking on the sandy beach", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/04_crab.jpg", | |
| "caption": "a crab" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "crab", | |
| "crab" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 22, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a stork standing on a grassy field", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/01_stork.jpg", | |
| "caption": "a stork" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "stork", | |
| "stork" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 23, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a wolf howling in the forest\n", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/27_wolf.jpg", | |
| "caption": "a wolf" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "wolf", | |
| "wolf" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 24, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a sly fox in a forest setting", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/10_fox.jpg", | |
| "caption": "a fox" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "fox", | |
| "fox" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 25, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute puppy in a red bow\n", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/42_puppy.jpg", | |
| "caption": "a puppy" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "puppy", | |
| "puppy" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 26, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute cat sitting on a mat", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/44_cat.jpg", | |
| "caption": "a cat" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "cat", | |
| "cat" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 27, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute dog in a red collar", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/07_dog.jpg", | |
| "caption": "a dog" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "dog", | |
| "dog" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 28, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a heron standing by a pond", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/animal/11_heron.jpg", | |
| "caption": "a heron" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "heron", | |
| "heron" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 29, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/05_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 30, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman smiling in a flower-filled garden", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/03_woman.jpg", | |
| "caption": "a woman" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "woman", | |
| "woman" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 31, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a cute boy smiling in a crib.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/00_boy.jpg", | |
| "caption": "a boy" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "boy", | |
| "boy" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 32, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a boy smiling in a sunny park", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/14_boy.jpg", | |
| "caption": "a boy" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "boy", | |
| "boy" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 33, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/06_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 34, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a girl smiling in a flower-filled garden", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/19_girl.jpg", | |
| "caption": "a girl" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "girl", | |
| "girl" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 35, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman in a red dress smiling", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/09_woman.jpg", | |
| "caption": "a woman" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "woman", | |
| "woman" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 36, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/07_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 37, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman standing in a park", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/13_woman.jpg", | |
| "caption": "a woman" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "woman", | |
| "woman" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 38, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man wearing a hat standing in the forest", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/17_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 39, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/02_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 40, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman standing in a garden", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/15_woman.jpg", | |
| "caption": "a woman" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "woman", | |
| "woman" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 41, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman standing in a flower-filled garden", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/12_woman.jpg", | |
| "caption": "a woman" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "woman", | |
| "woman" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 42, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/08_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 43, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/18_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 44, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/11_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 45, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/01_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 46, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/10_man.jpg", | |
| "caption": "a man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "man", | |
| "man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 47, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "an old man sitting on a bench.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/16_old man.jpg", | |
| "caption": "an old man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "old man", | |
| "old man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 48, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a little girl smiling in a garden.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/human/04_little girl.jpg", | |
| "caption": "a little girl" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "little girl", | |
| "little girl" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 49, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a man wearing a watch, standing. ", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/12_watch.jpg", | |
| "caption": "a watch" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "watch", | |
| "watch" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 50, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "anime spider-man in a city background", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/71_anime Spider-Man.jpg", | |
| "caption": "an anime spider-man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "anime spider-man", | |
| "anime spider-man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 51, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person wearing a cap smiling", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/32_cap.jpg", | |
| "caption": "a cap" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "cap", | |
| "cap" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 52, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "anime girl in a cute pose", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/66_anime girl.jpg", | |
| "caption": "an anime girl" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "anime girl", | |
| "anime girl" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 53, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person holding a mug in the kitchen", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/27_mug.jpg", | |
| "caption": "a mug" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "mug", | |
| "mug" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 54, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing among cherry blossoms", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/03_cherry blossoms.jpg", | |
| "caption": "a cherry blossoms" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "cherry blossoms", | |
| "cherry blossoms" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 55, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person rowing a boat on the lake", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/62_boat.jpg", | |
| "caption": "a boat" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "boat", | |
| "boat" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 56, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing by a stop sign", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/51_stop sign.jpg", | |
| "caption": "a stop sign" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "stop sign", | |
| "stop sign" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 57, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a guy playing guitar on the street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/39_guitar.jpg", | |
| "caption": "a guitar" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "guitar", | |
| "guitar" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 58, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing in front of a house", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/14_house.jpg", | |
| "caption": "a house" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "house", | |
| "house" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 59, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing in front of wooden house", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/45_wooden house.jpg", | |
| "caption": "a wooden house" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "wooden house", | |
| "wooden house" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 60, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person in a yellow taxi", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/44_yellow taxi.jpg", | |
| "caption": "a yellow taxi" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "yellow taxi", | |
| "yellow taxi" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 61, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "anime girl in a cute pose", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/70_anime girl.jpg", | |
| "caption": "an anime girl" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "anime girl", | |
| "anime girl" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 62, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person wearing a ring smiling\n", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/16_ring.jpg", | |
| "caption": "a ring" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "ring", | |
| "ring" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 63, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing by an airplane", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/37_airplane.jpg", | |
| "caption": "an airplane" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "airplane", | |
| "airplane" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 64, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a teddy bear sitting on a cozy sofa", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/13_teddy bear.jpg", | |
| "caption": "a teddy bear" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "teddy bear", | |
| "teddy bear" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 65, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person wearing sunglasses casually", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/58_sunglasses.jpg", | |
| "caption": "a sunglasses" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "sunglasses", | |
| "sunglasses" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 66, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person looking at a clock on the wall", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/56_clock.jpg", | |
| "caption": "a clock" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "clock", | |
| "clock" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 67, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person holding a pineapple smiling", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/43_pineapple.jpg", | |
| "caption": "a pineapple" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "pineapple", | |
| "pineapple" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 68, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person sitting in front of a vintage television", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/49_vintage television.jpg", | |
| "caption": "a vintage television" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "vintage television", | |
| "vintage television" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 69, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person using a vintage computer", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/35_vintage computer.jpg", | |
| "caption": "a vintage computer" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "vintage computer", | |
| "vintage computer" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 70, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "anime man in a cool pose", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/73_anime man.jpg", | |
| "caption": "an anime man" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "anime man", | |
| "anime man" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 71, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person in a t-shirt smiling", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/41_t-shirt.jpg", | |
| "caption": "a t-shirt" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "t-shirt", | |
| "t-shirt" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 72, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman holding a teapot in the kitchen", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/52_teapot.jpg", | |
| "caption": "a teapot" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "teapot", | |
| "teapot" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 73, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a robot standing in a city street", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/18_robot.jpg", | |
| "caption": "a robot" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "robot", | |
| "robot" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 74, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person carrying a backpack walking", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/28_backpack.jpg", | |
| "caption": "a backpack" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "backpack", | |
| "backpack" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 75, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person holding a magic cube", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/20_Magic Cube.jpg", | |
| "caption": "a magic cube" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "magic cube", | |
| "magic cube" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 76, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person wearing headphones, standing", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/55_headphones.jpg", | |
| "caption": "a headphones" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "headphones", | |
| "headphones" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 77, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person riding a bicycle outdoors", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/17_bicycle.jpg", | |
| "caption": "a bicycle" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "bicycle", | |
| "bicycle" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 78, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a steam locomotive chugging on tracks", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/19_steam locomotive.jpg", | |
| "caption": "a steam locomotive" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "steam locomotive", | |
| "steam locomotive" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 79, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person sitting in an armchair", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/11_armchair.jpg", | |
| "caption": "an armchair" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "armchair", | |
| "armchair" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 80, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person eating a donut casually.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/38_donut.jpg", | |
| "caption": "a donut" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "donut", | |
| "donut" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 81, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a woman holds a leather handbag.", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/57_leather handbag.jpg", | |
| "caption": "a leather handbag" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "leather handbag", | |
| "leather handbag" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 82, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a rider on a motorcycle speeding", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/00_motorcycle.jpg", | |
| "caption": "a motorcycle" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "motorcycle", | |
| "motorcycle" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 83, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a teddy bear sitting on a sofa", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/46_teddy bear.jpg", | |
| "caption": "a teddy bear" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "teddy bear", | |
| "teddy bear" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 84, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "anime samurai in a traditional pose", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/69_anime samurai.jpg", | |
| "caption": "an anime samurai" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "anime samurai", | |
| "anime samurai" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 85, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person sipping a cocktail in bar", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/63_cocktail.jpg", | |
| "caption": "a cocktail" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "cocktail", | |
| "cocktail" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 86, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a classical bust on a pedestal", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/25_classical bust.jpg", | |
| "caption": "a classical bust" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "classical bust", | |
| "classical bust" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 87, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person in a hot air balloon flying", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/09_hot air balloon.jpg", | |
| "caption": "a hot air balloon" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "hot air balloon", | |
| "hot air balloon" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 88, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "a person standing in front of a hut", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/42_hut.jpg", | |
| "caption": "a hut" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "hut", | |
| "hut" | |
| ] | |
| ] | |
| } | |
| ] | |
| }, | |
| { | |
| "index": 89, | |
| "input_images": [], | |
| "position_delta": [ | |
| 0, | |
| -32 | |
| ], | |
| "prompt": "pixelated warrior standing in a pixel world", | |
| "modulation": [ | |
| { | |
| "type": "adapter", | |
| "src_inputs": [ | |
| { | |
| "image_path": "assets/XVerseBench/object/67_pixelated warrior.jpg", | |
| "caption": "a pixelated warrior" | |
| } | |
| ], | |
| "use_words": [ | |
| [ | |
| 0, | |
| "pixelated warrior", | |
| "pixelated warrior" | |
| ] | |
| ] | |
| } | |
| ] | |
| } | |
| ] |