Upload create-mtg-training-data.py

create-mtg-training-data.py

@@ -0,0 +1,388 @@
+#a program to build a training data set for Stable Diffusion for Magic: The Gathering
+#it will pull all data from a saved JSON file, parse it, and save an image file and corresponding text file for each card
+
+import requests
+import json
+import os
+import random
+import time
+
+
+#set up the directory to save the images and data to
+directory = './mtg-training-data/'  #this is the directory where the images and data will be saved
+if not os.path.exists(directory):
+    os.makedirs(directory)  #make the directory if it doesn't exist already
+        
+#our json file location
+json_filename = './mtg-training-data/mtgdata.json'
+
+#save a normal, single-faced card
+def save_card_main(card):
+    #get the image url and filename
+    img_url = card['image_uris'].get('art_crop')
+    filename = which_filename(card)
+    #clean up any characters that might make the filename invalid
+    filename = filename.replace('/', '-')
+    filename = filename.replace(':', '-')
+    filename = filename.replace('?', '')
+    filename = filename.replace('!', '')
+    filename = filename.replace('\'', '-')
+    filename = filename.replace('\"', '')
+    filename = filename.replace(' ', '_')
+    filename = filename.replace('(', '')
+    filename = filename.replace(')', '')
+    filename = filename.replace(',', '')
+    filename = filename.replace('’', '')
+    filename = filename.replace('–', '-')
+    filename = filename.replace('—', '-')
+    filename = filename.replace('…', '')
+    #save the image unless it already exists
+    if not os.path.isfile('{}{}'.format(directory, filename)):
+        #add a 0.1 second delay if we get blocked by the server
+        #time.sleep(0.1)
+        save_image(img_url, filename)
+    #get the data
+    data = return_data(card)
+    #save the data
+    save_data(data, filename)
+    
+
+#determine the filename we're saving the image to; format is cardname_setcode_illustration_id.jpg
+def which_filename(card):
+    if card.get('flavor_name'):
+        name = card['flavor_name']
+    #if there's no illustration_id, use multiverse_ids as a backup
+    if card.get('illustration_id'):
+        id = card['illustration_id']
+    else:
+        id = card['multiverse_ids'][0]
+    id = str(id)
+    filename = card['name'] + '_' + card['set'] + '_' + id + '.jpg'
+    return filename
+
+#save the image
+def save_image(img_url, filename):
+    response = requests.get(img_url)
+    with open('{}{}'.format(directory, filename), 'wb') as f:
+        f.write(response.content)
+    report to stdout that we just saved the url to the filename
+    print('Saved {} to {}'.format(img_url, filename))
+
+
+#save the data as a text file with the same name as the image
+def save_data(data, filename):
+    filename = filename.replace('.jpg', '.txt')
+    with open('{}{}'.format(directory, filename), 'w', encoding="utf8") as f:
+        f.write(str(data))
+    #report to stdout what we just saved and to what filename
+    print('Saved {} to {}'.format(data, filename))
+    
+#get the data we want from the card and return it
+#from the json, we're getting card name, artist, the year of release, color, type, rarity, set name, set code, mana cost, any watermarks, power/toughness, any keywords, universes beyond info
+def return_data(card):    
+    glue = 'MTG card art'
+    name = card['name']
+    #look for a flavor name, change name to flavor name if it exists
+    if card.get('flavor_name'):
+        name = card['flavor_name']
+    #replace ' with ’ in name for placeholder purposes
+    name = name.replace('\'', '’')
+    artist = 'by ' + card['artist']
+    #replace ' with ’ in artist for placeholder purposes
+    artist = artist.replace('\'', '’')
+    year = get_year(card['released_at'])
+    if card.get('colors'):
+        colors_english = mtg_color_to_english(card['colors'])
+        colors = clean_colors(card['colors'])
+    else:
+        colors_english = 'Colorless'
+        colors = 'C'
+    card_type = card['type_line']
+    rarity = card['rarity']
+    set_name = card['set_name']
+    set_code = ' '
+    set_code = card['set']
+    set_type = card['set_type']
+    plane = ' '
+    plane = determine_plane(set_code)
+    if card.get('watermark'):
+        watermark = card['watermark']
+    else: watermark = 'WATERMARK_PLACEHOLDER'
+    mana_cost = clean_data(card['mana_cost'])
+    if card.get('security_stamp'):
+        security_stamp = parse_security_stamp(card['security_stamp'])
+    else: security_stamp = 'SECURITY_STAMP_PLACEHOLDER'
+    if card.get('power'):
+        power = card['power']
+    else: power = 'POWER_PLACEHOLDER'
+    if card.get('toughness'):
+        toughness = card['toughness']
+    else: toughness = 'TOUGHNESS_PLACEHOLDER'
+    power_toughness = power + '/' + toughness
+    if card.get('keywords'):
+        keywords = card['keywords']
+    else: keywords = 'KEYWORDS_PLACEHOLDER'
+    if card.get('promo_types'):
+        promo_types = card['promo_types']
+    else: promo_types = 'PROMO_TYPES_PLACEHOLDER'
+    if card.get('story_spotlight'):
+        story_spotlight = card['story_spotlight']
+    else: story_spotlight = 'SPOTLIGHT_PLACEHOLDER'
+    #at one point I put extra spaces in as a quick fix to make the txt files come out right - I have no idea if they are still needed
+    extra_space = ' '
+    #put it all together
+    data = glue, name, artist, year, colors_english, colors, card_type, rarity, set_name, set_code, plane, set_type, watermark, extra_space, mana_cost, extra_space, security_stamp, power_toughness, keywords, promo_types, story_spotlight
+    #then clean it up
+    data = clean_data(data)
+    return data
+   
+#determine the plane from the set code
+def determine_plane(set_code):
+    #convert set_code to uppercase
+    set_code = set_code.upper()
+    if set_code == 'ALA' or set_code == 'CON' or set_code == 'ARB':
+        plane = 'Alara'
+    elif set_code == 'AKH' or set_code == 'HOU':
+        plane = 'Amonkhet'
+    elif set_code == 'STX':
+        plane = 'Arcavios'
+    elif set_code == 'SNC':
+        plane = 'Capenna'
+    elif set_code == 'ELD':
+        plane = 'Eldraine'
+    elif set_code == 'CNS' or set_code == 'CN2':
+        plane = 'Fiora'
+    elif set_code == 'IKO':
+        plane = 'Ikoria'
+    elif set_code == 'ISD' or set_code == 'DKA' or set_code == 'AVR' or set_code == 'SOI' or set_code == 'EMN' or set_code == 'MID' or set_code == 'VOW':
+        plane = 'Innistrad'
+    elif set_code == 'IXL' or set_code == 'RIX':
+        plane = 'Ixalan'
+    elif set_code == 'KLD' or set_code == 'AER':
+        plane = 'Kaladesh'
+    elif set_code == 'KHM':
+        plane = 'Kaldheim'
+    elif set_code == 'SOK' or set_code == 'BOK' or set_code == 'CHK':
+        plane = 'Kamigawa, past'
+    elif set_code == 'NEC':
+        plane = 'Kamigawa, present'
+    #for some reason Kylem causes the script to crash. I have no idea why. I got tired of trying to figure it out, and it's 2am, so fuck Kylem
+    #elif set_code == 'BBD':
+    #    plane == 'Kylem'
+    elif set_code == 'LOR' or set_code == 'MOR':
+        plane = 'Lorwyn, LorShad'
+    elif set_code == 'SHM' or set_code == 'EVE':
+        plane = 'Shadowmoor, LorShad'
+    elif set_code == 'MMQ':
+        plane = 'Mercadia'
+    elif set_code == 'NPH':
+        plane = 'New Phyrexia'
+    elif set_code == 'MRD' or set_code == 'DST' or set_code == '5DN' or set_code == 'SOM' or set_code == 'MBS':
+        plane = 'Mirrodin'
+    elif set_code == 'ARN':
+        plane = 'Rabiah'
+    elif set_code == 'TMP' or set_code == 'STH' or set_code == 'EXO' or set_code == 'NEM':
+        plane = 'Rath'
+    elif set_code == 'RAV' or set_code == 'GPT' or set_code == 'DIS' or set_code == 'RTR' or set_code == 'GTC' or set_code == 'DGM' or set_code == 'GRN' or set_code == 'RNA' or set_code == 'WAR':
+        plane = 'Ravnica'
+    elif set_code == 'M10' or set_code == 'M11' or set_code == 'M12' or set_code == 'M13' or set_code == 'M14' or set_code == 'M15':
+        plane = 'Shandalar'
+    elif set_code == "KTK" or set_code == "FRF" or set_code == "DTK":
+        plane = 'Tarkir'
+    elif set_code == 'THS' or set_code == 'BNG' or set_code == 'JOU' or set_code == 'THB':
+        plane = 'Theros'
+    elif set_code == 'HML':
+        plane = 'Ulgrotha'
+    elif set_code == 'ZEN' or set_code == 'WWK' or set_code == 'ROE' or set_code == 'BFZ' or set_code == 'OGW' or set_code == 'ZNR':
+        plane = 'Zendikar'
+
+#Dominaria
+    elif set_code == 'LEA' or set_code == 'LEG' or set_code == 'DOM' or set_code == 'DMU':
+        plane = 'Dominaria'
+    elif set_code == 'ATQ' or set_code == 'DRK':
+        plane = 'Dominaria, Terisiare'
+    elif set_code == 'ICE' or set_code == 'ALL' or set_code == 'CSP':
+        plane = 'Dominaria, Terisiare, Ice Age'
+    elif set_code == 'BRO':
+        plane = 'Dominaria, past'
+    elif set_code == 'MIR' or set_code == 'VIS' or set_code == 'WTH' or set_code == 'PCY':
+        plane = 'Dominaria, Jamuraa'
+    elif set_code == 'TSP' or set_code == 'PLC' or set_code == 'FUT':
+        plane = 'Dominaria, Time Spiral'
+    elif set_code == 'INV' or set_code == 'PLS' or set_code == 'APC':
+        plane = 'Dominaria, Phyrexian Invasion'
+    elif set_code == 'ODY' or set_code == 'TOR' or set_code == 'JUD' or set_code == 'ONS' or set_code == 'LGN' or set_code == 'SCG':
+        plane = 'Dominaria, Otaria'
+    else:
+        plane = ' '
+    return plane
+
+#clean the colors up, remove the square brackets, quotes, commas and spaces
+def clean_colors(colors):
+    colors = str(colors)
+    colors = colors.replace('[', '')
+    colors = colors.replace(']', '')
+    colors = colors.replace('\'', '')
+    colors = colors.replace(',', '')
+    colors = colors.replace(' ', '')
+    #then add a space back at the start
+    #now rearrange the colors so they're in WUBRG order until there are no more rearrangements to be made
+    while True:
+        colors = colors.replace('UW', 'WU')
+        colors = colors.replace('BW', 'WB')
+        colors = colors.replace('RW', 'WR')
+        colors = colors.replace('GW', 'WG')
+        colors = colors.replace('BU', 'UB')
+        colors = colors.replace('GU', 'UG')
+        colors = colors.replace('RU', 'UR')
+        colors = colors.replace('RB', 'BR')
+        colors = colors.replace('GB', 'BG')
+        colors = colors.replace('GR', 'RG')
+        if 'UW' not in colors and 'BW' not in colors and 'RW' not in colors and 'GW' not in colors and 'BU' not in colors and 'GU' not in colors and 'RU' not in colors and 'RB' not in colors and 'GB' not in colors and 'GR' not in colors:
+            break
+    return colors
+#figure out what the security stamp is telling us
+def parse_security_stamp(security_stamp):
+    if security_stamp:
+        if security_stamp == 'acorn':
+            return 'acorn'
+        if security_stamp == 'triangle':
+            return 'Universes Beyond'
+        if security_stamp == 'arena':
+            return 'Arena'
+        if security_stamp == 'circle':
+            ##return 'Signature Spellbook'
+            #return nothing for now
+            return None
+        if security_stamp == 'heart':
+            return 'My Little Pony'
+        if security_stamp == 'oval':
+            return None
+    else: return None
+
+
+#remove all characters other than WUBRG, then convert letter by letter
+def mtg_color_to_english(colors):
+    #remove all characters other than WUBRG
+    colors = ''.join([c for c in colors if c in 'WUBRG'])
+    #for each letter in colors, add the corresponding color to colors_english, plus a comma and space
+    colors_english = ''
+    for c in colors:
+        if c == 'W':
+            colors_english += 'White, '
+        if c == 'U':
+            colors_english += 'Blue, '
+        if c == 'B':
+            colors_english += 'Black, '
+        if c == 'R':
+            colors_english += 'Red, '
+        if c == 'G':
+            colors_english += 'Green, '
+        if c == '':
+            colors_english += 'Colorless, '
+    #remove the last comma and space
+    colors_english = colors_english[:-2]
+    return colors_english
+        
+#get the year from the release date
+def get_year(date):
+    year = date[:4]
+    return year
+
+#special function for double-faced cards - we are going to save the front and back separately
+#scryfall treats them as a single card, but we want to treat them as two separate cards, so we need to split them up
+def split_dfc(card):
+    face = 0
+    while face < 2:
+        card['illustration_id'] = card['card_faces'][face]['illustration_id']
+        card['name'] = card['card_faces'][face]['name']
+        if card['card_faces'][face].get('flavor_name'):
+            card['flavor_name'] = card['card_faces'][face]['flavor_name']
+        card['mana_cost'] = card['card_faces'][face]['mana_cost']
+        card['type_line'] = card['card_faces'][face]['type_line']
+        card['oracle_text'] = card['card_faces'][face]['oracle_text']
+        card['colors'] = card['card_faces'][face]['colors']
+        if card['card_faces'][face].get('power'):
+            card['power'] = card['card_faces'][face]['power']
+        if card['card_faces'][face].get('toughness'):
+            card['toughness'] = card['card_faces'][face]['toughness']
+        card['artist'] = card['card_faces'][face]['artist']
+        card['image_uris'] = card['card_faces'][face]['image_uris']
+        save_card_main(card)
+        face += 1
+         
+#clean up the data we're saving so the AI can use it
+#remove curly brackets, convert long dash to hyphen
+def clean_data(data):
+    data = str(data)
+    data = data.replace('{', '')
+    data = data.replace('}', '')
+    data = data.replace('—', '-')
+    data = data.replace('—', "-")
+    #remove placeholder text
+    data = data.replace('WATERMARK_PLACEHOLDER', '')
+    data = data.replace('SECURITY_STAMP_PLACEHOLDER', '')
+    data = data.replace('POWER_PLACEHOLDER', '')
+    data = data.replace('TOUGHNESS_PLACEHOLDER', '')
+    data = data.replace('KEYWORDS_PLACEHOLDER', '')
+    data = data.replace('PROMO_TYPES_PLACEHOLDER', '')
+    data = data.replace('SPOTLIGHT_PLACEHOLDER', '')
+    #remove double commas
+    data = data.replace(', ,', ',')
+    #remove ( and )
+    data = data.replace('(', '')
+    data = data.replace(')', '')
+    #remove square brackets
+    data = data.replace('[', '')
+    data = data.replace(']', '')
+    #remove single quotes
+    data = data.replace('\'', '')
+    #remove double quotes
+    data = data.replace('"', '')
+    #remove any " , " that might be left over
+    #data = data.replace(' , ', '')
+    #remove double commas
+    data = data.replace(',,', '')
+    #remove /,
+    data = data.replace('/,', '')
+    #remove " ,"
+    data = data.replace(' ,', ' ')
+    #remove double spaces
+    data = data.replace('  ', ' ')
+    #replace ǵ with g
+    data = data.replace('ǵ', 'g')
+    #replace õ with o
+    data = data.replace('õ', 'o')
+    #replace é with e
+    data = data.replace('é', 'e')
+    #replace ï with i
+    data = data.replace('ï', 'i')
+    #replace ñ with n
+    data = data.replace('ñ', 'n')
+    #replace ë with e
+    data = data.replace('ë', 'e')
+    #replace û with u
+    data = data.replace('û', 'u')
+    #replace ó with o
+    data = data.replace('ó', 'o')
+    #replace ń with n
+    data = data.replace('ń', 'n')
+    #put ' back in for the placeholder `
+    data = data.replace('`', '\'')
+    return data
+
+#open the JSON file and load the data
+with open(json_filename, 'r', encoding="utf8") as f:
+    card_data = json.load(f)
+    #loop through each card in the data, one by one, and save the image and data
+for card in card_data:
+    #figure out if we're working with a double-faced card or a split card - we have a special function for those
+    if card['layout'] == 'transform' or card['layout'] == 'modal_dfc' or card['layout'] == 'reversible_card' or card['layout'] == 'double_faced_token':
+        split_dfc(card)
+    #if the layout is art_series, skip it, we don't want to save those
+    elif card['layout'] == 'art_series':
+        continue
+    else: save_card_main(card)
+
+
+