example config: admin_id comment

pre doesn't allow to copy prompt by clicking on it

README.md

@@ -1,35 +1,20 @@
-# ai621
+# ai621111 (A1111 webui fork)
 
-This is the code currently running over at @ai621bot on telegram.
-Don't worry, it's gonna be rewritten soon.
+This is fork of ai621 (https://git.foxo.me/AI621/ai621) bot that contains some additional features/fixes.
 
 ## How does it work?
 
-This is only the "bot" part. The actual generation is done by Easy Diffusion. To run this, you have to install and run an Easy Diffusion (https://github.com/easydiffusion/easydiffusion) instance and define the IP/hostname in the config files so that the bot can query and send data to port 9000 using the embedded api.
+This is only the "bot" part. The actual generation is done by AUTOMATIC1111 Stable Diffusion web UI. To run this, you have to install and run an AUTOMATIC1111 Stable Diffusion web UI (https://github.com/AUTOMATIC1111/stable-diffusion-webui) instance and define the IP/hostname in the config files so that the bot can query and send data to port 7860 using the embedded api.
 
 ## How do i run it?
+- Create bot account. To get a bot account, you need to talk with `@BotFather`.
+- Create two Telegram channels - one for output, and one for log. Add your bot as administrator, enable rights to post messages.
+- Start your A1111 web UI instance. Don't forget to add `--api` to `set COMMANDLINE_ARGS=` in `webui-user.bat`!
+- Obtain a Telegram api_id: https://docs.telethon.dev/en/stable/basic/signing-in.html
+- Create a Python virtual env to not pollute your system: https://docs.python.org/3/library/venv.html#creating-virtual-environments
+- Install all the dependencies needed by the bot using `pip install -r requirements.txt` after activating the venv
+- Copy example config file, rename it to `config.py` and edit it following the instructions.
+- Copy example database and rename it to `ai621.db`.
+- Run the bot with `python3 bot.py`
 
-1. Obtain a Telegram api_id: https://docs.telethon.dev/en/stable/basic/signing-in.html
-2. Create a Python virtual env to not pollute your system: https://docs.python.org/3/library/venv.html#creating-virtual-environments
-3. Install all the dependencies needed by the bot using "pip install -r requirements.txt" after activating the venv
-4. Edit the config files following the instructions
-5. Run the bot with "python3 bot.py"
-
-## User data
-
-Data is stored inside "ai621.db": three tables (user, prompt and pending_prompt) define all the data. Additionally, uploaded images (for the img2img mode) are stored inside the "userdata/" folder which is inside of the work folder.
-
-## Can you define multiple video cards / nodes to speed generation up?
-
-No. Or at least, not anymore. You will notice from the example configuration that three nodes were defined. This was working properly, until i accidentally overwrote the files that were responsible for that with an older version that didn't have that function anymore. I didn't bother.
-
-## What are the log and raw channels?
-
-ai621 posts all generated images to a channel called "ai621 raw", and then forwards them to the user at the end of the prompt processing. This is so that everybody can see and get inspired by other prompts. This is what the raw channel is.
-The log channel, instead, is where all prompts and stored along with buttons to be able to moderate users (delete prompts and ban users).
-
-## What are "cycles"?
-
-Every day each user has a defined number of cycles, also known as "steps". Since steps can be defined in a sorta stable amount of time, this is used to make sure that all users can fairly use the bot.
-
-Some specific prompts also ramp up the amount of cycles used as a "penalty", like using blacklisted tags or bad quality prompts.
+Additional info about the bot can be found in the upstream repo (https://git.foxo.me/AI621/ai621).

bot.py

@@ -113,6 +113,12 @@ async def edit_pending(ev):
 		await ev.respond('Something went wrong. Try /start again.')
 		return
 		
+	if field == 'negative_prompt':
+		if data == 'no_negative_prompt':
+			conn.execute('UPDATE pending_prompt SET negative_prompt = NULL WHERE user_id = ?', (ev.input_sender.user_id,))
+			conn.execute(f"UPDATE user SET pending_action = NULL WHERE id = ?", (ev.input_sender.user_id,))
+			return await edit_prompt(ev)
+	
 	if field == 'number':
 		try:
 			data = int(data)
@@ -298,6 +304,10 @@ async def queue_info(ev):
 	waiting_usrs = conn.execute("SELECT count(*) FROM prompt WHERE is_done IS NULL").fetchone()
 	avg_speed = conn.execute("SELECT avg(aa) FROM (SELECT (inference_steps*number)/abs(completed_at-started_at) AS aa FROM prompt WHERE started_at IS NOT NULL AND completed_at IS NOT NULL ORDER BY id DESC LIMIT 10)").fetchone()
 
+	avg_comp_time = avg_comp_time if avg_comp_time[0] is not None else [0]
+	min_max_wait = min_max_wait if min_max_wait[0] is not None else [0, 0]
+	avg_speed = avg_speed if avg_speed[0] is not None else [0]
+
 	await ev.respond("\n".join([
 		f"👯‍♂️ {waiting_usrs[0]} people are waiting in the queue",
 
@@ -377,6 +387,7 @@ async def queue(ev, new_message=False):
 		raise events.StopPropagation
 		
 	avg_comp_time = conn.execute('SELECT avg(aa) FROM (SELECT abs(queued_at-started_at) AS aa FROM prompt WHERE queued_at IS NOT NULL AND completed_at IS NOT NULL ORDER BY id DESC LIMIT 10)').fetchone()
+	avg_comp_time = avg_comp_time if avg_comp_time[0] is not None else [0]
 
 	behind_you = conn.execute('SELECT count(*) FROM prompt WHERE is_done IS NULL AND id > ?', (prompt['id'],)).fetchone()[0]
 	front_you = conn.execute('SELECT count(*) FROM prompt WHERE is_done IS NULL AND id < ?', (prompt['id'],)).fetchone()[0]
@@ -455,6 +466,7 @@ desc = {
 
 desc_buttons = {
 	'seed': [Button.inline(f"Get from last prompt", f"msg_but:last_prompt"),],
+	'negative_prompt': [Button.inline(f"🚫 Remove negative prompt", f"msg_but:no_negative_prompt")],
 	'number': [Button.inline(f"1️⃣", f"msg_but:1"), Button.inline(f"2️⃣", f"msg_but:2"), Button.inline(f"3️⃣", f"msg_but:3"), Button.inline(f"4️⃣", f"msg_but:4")],
 	'detail': [Button.inline(f"S (6.0)", f"msg_but:6"), Button.inline(f"M (10)", f"msg_but:10"), Button.inline(f"L (18)", f"msg_but:18")],
 	'inference_steps': [Button.inline(f"S (20)", f"msg_but:20"), Button.inline(f"M (40)", f"msg_but:40"), Button.inline(f"L (60)", f"msg_but:60")],
@@ -546,8 +558,8 @@ async def edit_prompt(ev):
 
 🖼 Starting image: {'None' if p['image'] == 'no_image' else ('https://e621.net/posts/'+str(p['image_e6']) if p['image_e6'] else 'User uploaded image')}
 🌱 Seed: {p['seed']}
-👀 Prompt: <pre>{p['prompt']}</pre>
-⛔ Negative prompt: <pre>{p['negative_prompt'] or 'no negative prompt set.'}</pre>
+👀 Prompt: <code>{p['prompt']}</code>
+⛔ Negative prompt: <code>{p['negative_prompt'] or 'no negative prompt set.'}</code>
 
 <b>Do not touch these if you don't know what you're doing</b>
 🔢 Number of images: {p['number']}
@@ -615,6 +627,6 @@ if __name__ == '__main__':
 			continue
 		client.queue.put_nowait((task['id'], task['id'],))
 
-	client.start()
+	client.start(bot_token=bot_token)
 	client.flood_sleep_threshold = 24*60*60
 	client.run_until_disconnected()

config.example.py

@@ -7,10 +7,12 @@ from process_queue import *
 
 coloredlogs.install(level='INFO')
 
-api_id = YOUR TG API ID HERE
-api_hash = YOUR TG API HASH HERE
+# https://docs.telethon.dev/en/stable/basic/signing-in.html
+api_id = <YOUR TG API ID HERE>
+api_hash = <YOUR TG API HASH HERE>
 
-temp_folder = TEMP FOLDER OF THE GENERATIONS
+# Get it from @BotFather bot.
+bot_token = <YOUR BOT TOKEN HERE>
 
 client = TelegramClient('bot', api_id, api_hash)
 
@@ -22,16 +24,23 @@ conn.row_factory = sqlite3.Row
 client.conn = conn
 client.process = None
 
-client.log_channel_id = ID OF LOG CHANNEL
-client.main_channel_id = ID OF RAW CHANNEL
+# To get channel ID: Send something in private(!) channel, tap/click on it,
+# choose "Copy post link", paste it somewhere. It should look like this:
+# https://t.me/c/12345/2
+# 12345 part is ID - add "-100" prefix before it (like this: -10012345)
+# Replace placeholder with result:
+client.log_channel_id = <ID OF LOG CHANNEL>
+client.main_channel_id = <ID OF RAW CHANNEL>
+
 client.queue = PriorityQueue()
 client.media_lock = asyncio.Lock()
 client.process_queue = False
 
-client.admin_id = USER ID OF THE ADMIN
+# Send message to @my_id_bot, it will reply with your Telegram ID.
+client.admin_id = <USER ID OF THE ADMIN>
 
 workers = [
-	Worker('http://127.0.0.1:9000', client, 'armorlink'),
+	Worker('http://127.0.0.1:7860', client, 'armorlink'),
 	#Worker('http://127.0.0.1:9001', client, 'armorlink-low'),
 	#Worker('http://local.proxy:9000', client, 'g14')
 ]

process_queue.py

@@ -12,21 +12,10 @@ from time import time
 from telethon.errors.rpcerrorlist import MessageNotModifiedError, UserIsBlockedError
 
 log = logging.getLogger('process')
-temp_folder = '/home/ed/temp/'
 
+scale_factor = 1.5
 default_vars = {
-	"use_cpu":False,
-	"use_full_precision": False,
-	"stream_progress_updates": True,
-	"stream_image_progress": False,
-	"show_only_filtered_image": True,
-	"sampler_name": "dpm_solver_stability",
-	"save_to_disk_path": temp_folder,
-	"output_format": "png",
-	"use_stable_diffusion_model": "fluffyrock-576-704-832-960-1088-lion-low-lr-e61-terminal-snr-e34",
-	"metadata_output_format": "embed",
-	"use_hypernetwork_model": "boring_e621",
-	"hypernetwork_strength": 0.25,
+	"sampler_name": "DPM++ 2M Karras",
 }
 
 class Worker:
@@ -42,6 +31,25 @@ class Worker:
 		self.log = logging.getLogger(name)
 		self.prompt = None
 			
+	async def update_progress_bar(self, msg):
+		start = time()
+		last_edit = 0
+		while True:
+			async with httpx.AsyncClient() as httpclient:
+				step = await httpclient.get(self.api + '/sdapi/v1/progress')
+				try:
+					data = step.json()
+				except:
+					continue
+				if data['state']['sampling_step']%10 == 0:
+					self.log.info(f"Generation progress of {self.prompt}: {data['state']['job']} | {data['state']['sampling_step']}/{data['state']['sampling_steps']}")
+								
+				if time() - last_edit > 3:
+					await msg.edit(f"Generating prompt #{self.prompt}, {int(data['progress'] * 100)}% done. {time()-start:.1f}s elapsed, {int(data['eta_relative'])}s remaining.")
+					last_edit = time()
+							
+			await asyncio.sleep(2)
+
 	def start(self, future=None):
 	
 		if not self.client.process_queue:
@@ -67,12 +75,13 @@ class Worker:
 			self.task.add_done_callback(self.start)
 
 	async def process_prompt(self, prompt, queue_item):
+		endpoint = 'txt2img'
 	
 		# First of all, check if the user can still be messaged.
 		
 		async with httpx.AsyncClient() as httpclient:
 			try:
-				await httpclient.get(join(self.api, 'ping'), timeout=5)
+				await httpclient.get(self.api + '/internal/ping', timeout=5)
 			except Exception as e:
 				print(str(e))
 				log.error('Server is dead. Waiting 10 seconds for server availability...')
@@ -88,23 +97,23 @@ class Worker:
 		
 		# Prepare the parameters for the request
 		params = default_vars.copy()
-		params['session_id']		 	= str(prompt['id'])
 		params['prompt']			 	= prompt['prompt'] or ''
-		params['negative_prompt']	 	= prompt['negative_prompt'] or 'boring_e621_v4'
-		params['num_outputs']		 	= int(prompt['number'])
-		params['num_inference_steps']	= prompt['inference_steps']
-		params['guidance_scale']     	= prompt['detail']
+		params['negative_prompt']	 	= str(prompt['negative_prompt']) 
+		params['n_iter']		 	= int(prompt['number'])
+		#params['batch_size']        = int(prompt['number'])
+		params['steps']	= prompt['inference_steps']
+		params['cfg_scale']     	= prompt['detail']
 		params['width']					= prompt['resolution'].split('x')[0]
 		params['height']				= prompt['resolution'].split('x')[1]
 		params['seed']					= str(prompt['seed'])
-		params['vram_usage_level'] = 'low' if '-low' in self.name else ('medium' if '1024' in prompt['resolution'] else 'high')
 
 		self.prompt = prompt['id']
 
-		if prompt['hires'] == 'yes':
-			params['use_upscale'] = 'RealESRGAN_x4plus_anime_6B'
+		#if prompt['hires'] == 'yes':
+		#	params['use_upscale'] = 'RealESRGAN_x4plus_anime_6B'
 
 		if prompt['image'] != 'no_image':
+			endpoint = 'img2img'
 			img = Image.open(prompt['image'])
 			img = img.convert('RGB')
 			if prompt['crop'] == 'no':
@@ -115,60 +124,43 @@ class Worker:
 			imgdata = BytesIO()
 			img.save(imgdata, format='JPEG')
 
-			params['init_image'] = ('data:image/jpeg;base64,'+b64encode(imgdata.getvalue()).decode()).strip()
-			params['sampler_name'] = 'ddim'
-			params['prompt_strength'] = prompt['blend']
+			params['init_images'] = [('data:image/jpeg;base64,'+b64encode(imgdata.getvalue()).decode()).strip()]
+			params['sampler_name'] = 'DDIM'
+			params['denoising_strength'] = prompt['blend']
 
 		async with httpx.AsyncClient() as httpclient:
 
 			self.conn.execute('UPDATE prompt SET started_at = ? WHERE id = ?', (time(), prompt['id']))
 
-			start = time()
 			failed = False
 
 			self.log.info('POST to server')
-			res = await httpclient.post(join(self.api, 'render'), data=json.dumps(params))
-			res = res.json()
-
-			last_edit = 0
-
-			while 1:
-				step = await httpclient.get(join(self.api, res['stream'][1:]))
-				
+			progress_task = asyncio.create_task(self.update_progress_bar(msg))
 			try:
-					data = step.json()
-				except:
-					continue
-					
-				if 'step' in data:
-					if int(data['step'])%10 == 0:
-						self.log.info(f"Generation progress of {prompt['id']}: {data['step']}/{data['total_steps']}")
-							
-					if time() - last_edit > 10:
-						await msg.edit(f"Generating prompt #{prompt['id']}, step {data['step']} of {data['total_steps']}. {time()-start:.1f}s elapsed.")
-						last_edit = time()
-						
-				elif 'status' in data and data['status'] == 'failed':
-					await self.client.send_message(184151234, f"While generating #{prompt['id']}: {data['detail']}...")
-					await self.client.send_message(prompt['user_id'], f"While trying to generate your prompt we encountered an error: {data['detail']}\n\nThis might mean a bad combination of parameters, or issues on our sifde. We will retry a couple of times just in case.")
+				res = await httpclient.post(self.api + f'/sdapi/v1/{endpoint}', data=json.dumps(params), timeout=300) 
+			except Exception as e:
+				await self.client.send_message(self.client.admin_id, f"While generating #{prompt['id']}: {e}")
+				await self.client.send_message(prompt['user_id'], f"While trying to generate your prompt we encountered an error.\n\nThis might mean a bad combination of parameters, or issues on our side. We will retry a couple of times just in case.")
 				failed = True
 				self.client.conn.execute('UPDATE prompt SET is_error = 1, is_done = 1, completed_at = ? WHERE id = ?',(time(), prompt['id']))
-					break
-				elif 'status' in data and data['status'] == 'succeeded':
+				return
+			res = res.json()
+
 			self.log.info('Success!')
 			images = []
-					for img in data['output']:
-						imgdata = BytesIO(b64decode(img['data'].split('base64,',1)[1]))
-						#imgdata.name = img['path_abs'].rsplit('/', 1)[-1]
+			for img in res['images']:
+				imgdata = BytesIO(b64decode(img.split(",",1)[0]))
 				imgdata.name = 'image.png'
 				imgdata.seek(0)
 				images.append(imgdata)
-					break
-				else:
-					print(data)
 
-				await asyncio.sleep(2)
 
+				
+			progress_task.cancel()
+			try:
+				await progress_task
+			except asyncio.CancelledError:
+				pass
 			self.conn.execute('UPDATE prompt SET is_done = 1, completed_at = ? WHERE id = ?',(time(), prompt['id']))
 			await msg.delete()
 
@@ -176,12 +168,14 @@ class Worker:
 				asyncio.create_task(self.send_submission(prompt, images))
 				
 	async def send_submission(self, prompt, images):
+		await self.client.send_message(prompt['user_id'], 'Uploading images... This will take a while')		
 	
 		tg_files = []
 
 		for fn in images:		
 			img = Image.open(fn)
-			img.thumbnail((1280,1280))
+			width, height = img.size
+			img = img.resize((int(width * scale_factor), int(height * scale_factor)))
 			imgdata = BytesIO()
 			img.save(imgdata, format='JPEG')
 			siz = imgdata.seek(0,2)
@@ -189,10 +183,10 @@ class Worker:
 			tg_files.append(await self.client.upload_file(imgdata, file_size=siz, file_name=fn.name))
 						
 		results = await self.client.send_file(self.client.main_channel_id, tg_files, caption=
-			"\n".join([f"#{prompt['id']} · 🌀 {prompt['inference_steps']} · 🌱 {prompt['seed']} · 💎 {prompt['detail']}" + (f" · 🎚 {prompt['blend']} (seed from image)" if prompt['image'] != 'no_image' else ''),
+			"\n".join([f"#{prompt['id']} · 🌀 <code>{prompt['inference_steps']}</code> · 🌱 <code>{prompt['seed']}</code> · 💎 <code>{prompt['detail']}</code>" + (f" · 🎚 {prompt['blend']} (seed from image)" if prompt['image'] != 'no_image' else ''),
 			#((f"🖼 https://e621.net/posts/{prompt['image_e6']}" if prompt['image_e6'] else 'user-uploaded image') if prompt['image'] != 'no_image' else ''),
-			('👍 ' if prompt['negative_prompt'] else '')+(f"Prompt from https://e621.net/posts/{prompt['prompt_e6']}" if prompt['prompt_e6'] else prompt['prompt']),
-			(f"\n👎 {prompt['negative_prompt']}" if prompt['negative_prompt'] else '')])[:1000], parse_mode='HTML')
+			('👍 ' if prompt['negative_prompt'] else '')+(f"Prompt from https://e621.net/posts/{prompt['prompt_e6']}" if prompt['prompt_e6'] else f'<code>{prompt["prompt"]}</code>'),
+			(f"\n👎 <code>{prompt['negative_prompt']}</code>" if prompt['negative_prompt'] else '')])[:1000], parse_mode='HTML')
 
 		await self.client.forward_messages(prompt['user_id'], results)
 		if prompt['hires'] == 'yes':