Ai journey w/ Generated Imagery

Intro of 3 of the most common AI tools that are able to create realistic images given a prompt:

DALL-E
Midjourney
Stable Diffusion

Transcript

1. AI JOURNEY W/ GENERATED IMAGERY
2. TODAY Intro of 3 of the most common AI tools that are able to create realistic images give a prompt: •DALL-E •Midjourney •Stable Diffusion child riding a whale, photographic --ar 2:3 --v 5
3. ROELAND WEVE • Graduated AI in 2007 (UvA) • Since 2011 CTO/Co-founder at Pinch
4. REMINDER • Missed a slide or prompt? Feel free to reach out after the presentation to see it again • Although I’m a pretty lazy prompt engineer, my prompts are only few words these days… full-length portrait of neon glowing cyborg,peter gric style,neontubes,neon,wires,hyperrealistic,real,lifelike, ultrarealistic,intricate details,8k --ar 2:3 --v 5
5. DALL-E
6. JANUARY 2021 
 https://twitter.com/gdb/status/1346554999241809920 Joined waiting list in April ’22 and July fi nally got access
7. DALL-E • Easy to easy use browser based • OpenAI account 
 (same company as ChatGPT) • Credit system (115 credits / $15) • In- and outpainting • Development and quality behind
8. DALL-E OUTPAINTING “A guy wearing red pants”
9. DALL-E OUTPAINTING “A guy wearing a red t-shirt”
12. MIDJOURNEY From v2/3 to v5.1
13. • Really (dangerously) good quality • Supports different aspect ratio’s • New updates regularly (3d, drawing, web) • $10 - $30 - $60 p/m MIDJOURNEY (FEB 2022) • Works with Discord • No in/outpainting support (yet) • Sometimes it’s down or slow • Open (all the prompts are public, unless 
 you pay to be private), but closed models • Annoying “NSFW” fi lter (13+ rating) CONS PROS
14. MIDJOURNEY - JULY ‘22 “a photo of a robot sweating in the sahara”
15. “a wooden robot 
 with glass eyes” MIDJOURNEY
16. “a robot that looks like Donatello from the Teenage Mutant Ninja Turtles” MIDJOURNEY
17. “half mechanical sonic the hedgehog on fi re and glowing eyes, hyperrealistic, real, lifelike, ultrarealistic, intricate details, 8k” MIDJOURNEY
18. MIDJOURNEY TESTP FLAG robot Wonder Women with glowing eyes, hyperrealistic, real, lifelike, ultrarealistic, intricate details,8k --testp --upbeta --v 3
19. majestic spider- woman, fi nely detailed body, cinematic lighting, intricate fi ligree pink metal design, 4k, 8k, unreal engine, octane render --testp -- ar 6:19 --v 3 MIDJOURNEY
20. MIDJOURNEY
21. MIDJOURNEY V3 Donatello from teenage mutant ninja turtles wearing shiny cybernetic suit, golden hour, new york streets, hyperrealistic, real, lifelike, ultrarealistic, intricate details, 8k, ultra-detailed photography --ar 2:3 --v 3 a robot that looks like Donatello from the Teenage Mutant Ninja Turtles
22. “donatello from teenage mutant ninja turtles wearing shiny cybernetic suit, golden hour, new york streets, hyperrealistic, real, lifelike, ultrarealistic, intricate details, 8k, ultra-detailed photography --ar 2:3 --v 4 --q 2” MIDJOURNEY V4
23. MIDJOURNEY V5 half mechanical donatello from teenage mutant ninja turtles, golden hour, new york streets, hyperrealistic, real, lifelike, ultrarealistic, intricate details, 8k, ultra-detailed photography --ar 2:3
24. full-length portrait of neon glowing cyborg, peter gric style, neon tubes, neon wires, hyperrealistic, real, lifelike, ultrarealistic, intricate details,8k --ar 2:3 --v 5 MIDJOURNEY CYBORG
25. MIDJOURNEY PHOTO REALISTIC V5 improved a lot on hands, photo quality and delivering 
 realistic images. “pope francis wearing pink suit --v 5”
26. MIDJOURNEY ARTISTIC
27. It’s fun to join challenges. Nothing win, but it’s interesting what you can come up with. Fashion show MIDJOURNEY “CHALLENGES” woman walking fashion show wearing huge fi re out fi t --ar 2:3 --v 5
28. MIDJOURNEY CHALLENGE: TRAIN #promptmadness2023 
 “insane train monster --v 5”
29. #promptmadness2023 “obama wearing teal spandex suit --v 5” MIDJOURNEY CHALLENGE: TEAL
30. #promptmadness2023 
 man wearing balloon out fi t, year 2020, new york --ar 2:3 --v 5 MIDJOURNEY CHALLENGE: BALLOONS
31. Chubby Trump wearing a kimono, Cinematic, portrait Photography, Shot on 50mm lens, Ultra-Wide Angle, Depth of Field, hyper-detailed, beautifully color-coded, insane details, intricate details, beautifully color graded, Unreal Engine, Cinematic, Color Grading, Editorial Photography, Photography, Photoshoot, Shot on 70mm lens, Depth of Field, Bokeh, DOF, Tilt Blur, Shutter Speed 1/1000, F/22, White Balance, 32k, Super-Resolution, Megapixel, Backlight, Cinematic Lighting, Studio Lighting, Soft Lighting, Volumetric, Contre-Jour, Beautiful Lighting, Accent Lighting, Global Illumination, Optics, Scattering, Glowing, Shadows, Rough, Shimmering, Diffraction Grading, Chromatic Aberration, Scan Lines, Ambient Occlusion, FKAA, TXAA, RTX, SSAO, Shaders, OpenGL-Shaders, GLSL-Shaders, Post Processing, Post-Production, Cel Shading, Tone Mapping, CGI, VFX, SFX, hypermaximalist, elegant, hyper realistic, super detailed, dynamic pose, photography, 8k --ar 2:3 MIDJOURNEY CHALLENGE: KIMONO
32. /describe (upload photo): “dentist dana lathamhahn shares a cute snapshot, in the style of new leipzig school, handsome, frontal perspective, smilecore, tenwave, creased crinkled wrinkled” MIDJOURNEY “DESCRIBE”
33. MIDJOURNEY “BLEND”
34. MIDJOURNEY “REMIXING”
35. MIDJOURNEY “REMIXING”
36. MIDJOURNEY “REMIXING”
37. MIDJOURNEY REFERENCE IMAGE Upload your own image and use that as reference image. /imagine URL prompt “https://s.mj.run/QM5AfnfJsDY old man with brown hair and short beard“ With the “iw” parameter you can tweak the weight of the image.
38. MIDJOURNEY https://s.mj.run/QM5AfnfJsDY toddler with brown hair --ar 2:3
39. MIDJOURNEY https://s.mj.run/QM5AfnfJsDY old man with brown hair and short beard --ar 2:3 
 https://s.mj.run/QM5AfnfJsDY old man with brown hair and short beard smiling
40. MIDJOURNEY Goofy face Duck face Pixar
41. MIDJOURNEY /imagine [URL] lion face Photo of Julia Roberts What about 
 Julion Roberts?
42. MIDJOURNEY Sheep face Photo of Ed Sheeran What about 
 Ed Sheepan?
43. MIDJOURNEY muskusrat face Photo of Elon Musk What about 
 Elon Muskrat?
44. MIDJOURNEY tomato Photo of a Tom Cruise Nice to meet you Tomato Cruise!
45. MIDJOURNEY
46. MIDJOURNEY chicken Photo of a kangaroo Mix and match 
 two animals, 
 let’s start with
 Kangarooster
47. MIDJOURNEY crocodile Photo of a duck Meet Crocoduck!
48. MIDJOURNEY kitten Photo of a penguin Meet Penguitten!
49. MIDJOURNEY bee Photo of an elephant Meet Beelephant
50. MIDJOURNEY
51. MIDJOURNEY • Use exact words (no typos) • Words in the beginning of the prompts have more importance. Repeating words might fi x this (cow cow cow cow) • Prompts usually don’t work across MJ versions • There are people collecting all the different styles in Google Docs •https://bit.ly/mj-styles • If you need high quality photos there are plenty of upscales around. Don’t use Photoshop for this.
52. STABLE DIFFUSION
53. STABLE DIFFUSION (AUG ’22) • Can run locally (depending on hardware) • Lots of extensions, models, freedom (no fi lters) • No downtime • Free to use • Supports in- and outpainting • Aspect ratio support limited • SD models not good with hands and faces, but many models available (https://civitai.com/) Paid services: https://beta.dreamstudio.ai/generate
54. BUT ALSO… • Many many options • Can be overwhelming • WebUI launched by AUTOMATIC1111 • Youtube videos can help you out
55. One of the big advantages of SD is that you can generate unlimited images (no extra costs) By using Deforum you can give prompts per keyframe, you can give the animation movements and many more options to generate videos. STABLE DIFFUSION
56. STABLE DIFFUSION Video contains 1200 frames, every 300 frames I changed the prompt: 
 
 300: “portrait photograph of silvermist fairy, xxxxx” 600: "portrait photograph of vidia fairy, xxx “ 900: "portrait photograph of rosetta fairy, xxxx” zoom: 0:(1.0025+0.002*sin(1.25*3.14*t/1200)) translation_x: 0:(5*sin(2*3.14*t/1200)) translation_y: 0:(0) translation_z: 0:(1.75) transform_center_x: 0:(0.5) transform_center_y: 0:(0) 0: “portrait photograph of tinker bell, young beautiful woman, perfect symmetrical face, wings, (((fairy tale))) environment, Disney, ultra realistic, concept art, elegant, ((intricate)), ((highly detailed)), depth of fi eld, ((professionally color graded)), 8k, art by artgerm and greg rutkowski and alphonse mucha 
 --neg (((backlight))), dark face, white fur, gold fabric, ((cropped head)), ((out of frame)), deformed, cripple, ugly, additional arms, additional legs, additional head, two heads, multiple people, group of people”
57. STABLE DIFFUSION: CONTROLNET & SCRIBBLE
58. STABLE DIFFUSION: CONTROLNET & SCRIBBLE
59. STABLE DIFFUSION: CONTROLNET & OPENPOSE
60. STABLE DIFFUSION: COMBINATION • Deforum / Controlnet • EBSynth • FFmpeg
61. STABLE DIFFUSION: COMBINATION #2 • Get one frame from the video, use it for img2img (reference image). • Go nuts with prompts and transitions.
62. STABLE DIFFUSION: MUSIC • Upload mp3 to www.framesync.xyz • Choose on frame for img2img • Write a creative prompt
63. STABLE DIFFUSION: DEPTH AND BLENDER Midjourney Depth map with Stable Diffusion
64. STABLE DIFFUSION: DEPTH AND BLENDER Blender
65. Besides DALL-E, Midjourney and Stable Diffusion there are many many many other generative AI tools out there. DALL-E is a bit behind at the moment, Midjourney is really good for realistic art. Stable Diffusion is a little bit hard at the beginning, but really cool for videos. Besides the basic prompting there are many many possibilities! RECAP
66. SOCIALS • instagram.com/robots.rule • twitter.com/robots_rule • linkedin.com/in/roelandweve
67. THANKS FOR LISTENING ENJOY YOUR WEEKEND! +31 20 22 10 717 Sloterkade 41-H, 1058 HE, Amsterdam www.pinch.nl Calle de Bolsa 3, planta 4, 29015, Málaga