This is Part 4. There is also Part 1, Part 2, Part 3, Part 5, Part 6, Part 7 and Part 8.
This post continues listing the Text-to-Image scripts included with Visions of Chaos and some example outputs from each script.
Name: PixelDraw
Author: dribnet
Original script: https://colab.research.google.com/github/dribnet/clipit/blob/master/demos/PixelDrawer.ipynb
Time for 512×512 on a 3090: 1 minutes 59 seconds
Maximum resolution on a 24 GB 3090: Huge. 4096×4096 and beyond.
Maximum resolution on an 8GB 2080: Unable to run on 8GB VRAM
Description: Generates “pixel art” images. I had a lot of requests to add support for this one.
a cartoon of a peacock
a cloudy sunset
a gorilla
a morning landscape
a watercolor painting of a castle
an art deco painting of Al Pacino
Hell
Shrek
Name: DirectVisions
Author: Jens Goldberg
Original script: https://colab.research.google.com/drive/127lKSsQjx-UDDUSvIkLL6mREfZ0KQu5D
Time for 512×512 on a 3090: 2 minutes 39 seconds
Maximum resolution on a 24 GB 3090: Huge. 4096×4096 and beyond.
Maximum resolution on an 8GB 2080: 4096×4096
Description: Interesting detailed images. Can create huge resolution results.
a color pencil sketch of a western town
a detailed painting of a cephalopod
a digital rendering of an ugly face
a pencil sketch of Buzz Lightyear
a rough seascape by Pinchus Kremegne
a stock photo of a president
a sunset
an alien city
an alien forest by Helen Berman
an evening landscape
Name: Pixel Direct
Author: Unknown
Original script: https://colab.research.google.com/drive/1F9ZOZnpV3uBPRDSESaAXYwzNZJQRJT75
Time for 512×512 on a 3090: 1 minutes 03 seconds
Maximum resolution on a 24 GB 3090: Huge. 4096×4096 and beyond.
Maximum resolution on an 8GB 2080: 2048×2048 1 minute 51 seconds
Description: Another “Pixel Art” script. More abstract results than the PixelDraw script above.
a bronze sculpture of a nightmare creature
a cartoon of Al Pacino
a nightclub
a silk screen of a bouquet of flowers
an etching of a worried woman
an illustration of of a thunder storm
Name: FourierVisions
Author: Unknown
Original script: https://colab.research.google.com/drive/1nGNBjhbYnDHSumGPjpFHjDOsaZFAqGgF
Time for 512×512 on a 3090: 1 minutes 40 seconds
Maximum resolution on a 24 GB 3090: Huge. 4096×4096 and beyond.
Maximum resolution on an 8GB 2080: 1024×1024 4 minutes 07 seconds
Description: Detailed images. The default script generates washed out pastel images, but with some gamma and brightness tweaks they can be improved (still not ideal, but better). Allows very large resolution images.
a cathedral
a charcoal drawing of zombies
a detailed painting of a sunset by Thomas Cantrell Dugdale
a ghost made of mist
a kitchen
a movie monster
a pencil sketch of a sad clown
a werewolf
an evil clown by Viktor Oliva
an ink drawing of an ugly monster
Name: PyramidVisions
Author: Unknown
Original script: https://colab.research.google.com/drive/1dpAS_wK34y7c6s-CatAFmBtbkjGT_erM
Time for 512×512 on a 3090: 3 minutes 08 seconds
Maximum resolution on a 24 GB 3090: Huge. 4096×4096 and beyond.
Maximum resolution on an 8GB 2080: 1024×1024 10 minutes 48 seconds
Description: Very detailed images. Not the fastest script, but gives some very nice results. Lower VRAM requirements so good for lesser spec GPUs. Definitely one of the better scripts worth exploring.
a desert oasis
a lush rainforest
a marble sculpture of an angry person
a minimalist painting of the Amazon Rainforest
a nightmare creature
a pastel of a computer made of paper
an abstract sculpture of a sad clown
an acrylic painting of an alien forest | vivid colors
Medusa
vector art of an ugly woman
Name: Visions of AI v1
Author: Jason Rampe
Original script: Included with Visions of Chaos. No colab.
Time for 512×512 on a 3090: 1 minutes 32 seconds
Maximum resolution on a 24 GB 3090: 768×768 or 1120×480.
Maximum resolution on an 8GB 2080: 256×256 1 minute 33 seconds
Description: My first attempt at actually creating a Text-to-Image script. Based on the excellent example from Jonathan Whitaker‘s AIAIArt Lesson 3 tutorial. Gives some very nice fine detail in some areas, but suffers the non coherance of other scripts in that it creates multiple copies of the subject throughout the image. After actually trying to write my own script I only have more respect for those who can do this. Hopefully I can improve these results for a version 2. In the meantime, here are some sample from the current Visions of AI script.
a cartoon of the human condition by Judy Takács
a cubist painting of an evening landscape
a digital rendering of frogs
a fire breathing dragon
a hyperrealistic painting of a movie monster
a morning landscape
a shark
a woodcut of an ugly man
an airbrush painting of C-3PO
Frankenstein
Name: Visions of AI v2
Author: Jason Rampe
Original script: Included with Visions of Chaos. No colab.
Time for 512×512 on a 3090: 2 minutes 35 seconds
Maximum resolution on a 24 GB 3090: 768×768 or 1120×480.
Maximum resolution on an 8GB 2080: 256×256 2 minutes 36 seconds
Description: An attempt to improve the coherency of the previous script. The first 30 iterations zoom into the image every 10 frames. This results in larger shapes/blobs for the rest of the script to work from. The idea is that it will give larger subjects compared to the v1 script. Kind of works. Gives blurrier results. To be fixed in the next version?
a morning landscape by William Gear
a raytraced image of a nightclub lens flare
a tentacle monster by Carlo Crivelli
a woodcut of a worried woman by Li Keran
an illustration of of a cave made of cheese
Cthulhu
cyberpunk art of a futuristic city
goldfish
reflective spheres
the Australian outback
Name: Multi-Perceptor CLIP Guided Diffusion
Author: Varkarrus
Original script: https://colab.research.google.com/drive/1y3Vt39A5KSNFRa6Z2bCqDHxteZSVH9NC
Time for 512×512 on a 3090: 3 minutes 08 seconds
Maximum resolution on a 24 GB 3090: 896×512 or 1152×384 (dimensions must be divisible by 128).
Maximum resolution on an 8GB 2080: 128×128 1 minute 56 seconds
Description: Builds upon previous CLIP Guided Diffusion scripts. Like the previous script by Dango233 it uses three CLIP models simultaneously to “rate” the generated images, and I have added options to use up to six different CLIP models. The resulting image accuracy compared to the prompt, and the resulting image coherence seem to be much better than previous CLIP Guided Diffusion scripts that could almost have random outputs sometimes. This script is superb and highly recommended. Great lighting, textures and brushstrokes. Normally with these blog posts I do a batch run of random prompts overnight and then pick the best 10 images. In this case I had nearly 50 images in my “good” folder after going through the batch results. So, for this script I am showing 20 sample images.
a cute creature | TriX 400 TX
a digital painting of Frankenstein by Kanzan Shimomura
a morning landscape by János SaxonSzász
a nightmare creature
a photorealistic painting of a teddy bear
a portrait of a young girl
a space nebula | IMAX
a worried man
a zombie by Nathaniel Hone
an acrylic painting of a spider by Abram Arkhipov
an airbrush painting of a monkey by Jeremy Henderson
an alien landscape
an ugly creature made of insects
an ultrafine detailed painting of a sad person | ZBrush
Arnold Schwarzenegger | trending on ArtStation
concept art of Robocop
dinosaurs
Dracula | CGSociety
flesh made of insects
God by William Simpson
Name: Pixel MultiColors
Author: Remi Durant
Original script: https://colab.research.google.com/drive/17c-13cl_VQKpHq2rDrnFVi6ZT-CHeZNn
Time for 512×512 on a 3090: 0 minutes 44 seconds
Maximum resolution on a 24 GB 3090: 4096×4096.
Maximum resolution on an 8GB 2080: 2048×2048 7 minutes 45 seconds
Description: Very noisy/pixelated/abstract results. The default script gives dark images which some tweaks to brightness and contrast can help. Maybe a little bit of blur could help too in a future revision. It is fast though, and can support huge image sizes.
a charcoal drawing of a cute creature made of metal
a farm
a forest path by Walter Leighton Clark
a lighthouse
a surrealist painting of a beachside resort
a well kept garden
an abstract sculpture of Pikachu
an art deco painting of a volcano
an ink drawing of tentacles
an octopus Rendered in Cinema4D
Name: Ultraquick CLIP Guided Diffusion
Author: @sadly_existent
Original script: https://colab.research.google.com/github/sadnow/360Diffusion/blob/main/360Diffusion_AlphaTesting.ipynb
Time for 512×512 on a 3090: 1 minute 57 seconds
Maximum resolution on a 24 GB 3090: Locked to either 256×256 or 512×512.
Maximum resolution on an 8GB 2080: Unable to run on 8GB VRAM
Description: Another CLIP Guided Diffusion script. Can give some interesting results.
a cave
a color pencil sketch of Cthulhu
a detailed painting of Shrek
a flemish baroque of the human condition by George Barret Jr
a low poly render of halloween
a photorealistic painting of a worried woman made of paper by Ann Thetis Blacker
a surrealist painting of a worried man
a surrealist sculpture of an angry man 8K 3D
Robocop
zombies
Name: ruDALL-E
Author: @sadly_existent
Original script: https://colab.research.google.com/drive/1wGE-046et27oHvNlBNPH07qrEQNE04PQ
Optimized script: https://colab.research.google.com/drive/1euIMG8E6kSFA2nU58LqrVsq6nbXjqELY
Time for 256×256 on a 3090: 1 minute 05 seconds
Maximum resolution on a 24 GB 3090: Locked to 256×256.
Maximum resolution on an 8GB 2080: Cannot run on 8GB VRAM
Description: Russian version of DALL-E. Only takes text prompts in Russian, so I do some auto English to Russian translations. Locked to small 256×256 images at this stage, but can create some interesting results.
a hyperrealistic painting of Chewbacca by Edith Grace Wheatley
a low poly render of Pikachu
a man
a rose
a stock photo of puppies
egyptian art of a portrait of a woman
Harry Potter
Indiana Jones
Robocop made of gold
Yoda
Name: ruVQGAN+CLIP
Author: nev
Original script: https://colab.research.google.com/drive/1wAnIHocDYFAbWtA7rk8C7cFEUdRyLzwZ
Time for 512×512 on a 3090: 1 minute 28 seconds
Maximum resolution on a 24 GB 3090: 1120×480.
Maximum resolution on an 8GB 2080: 256×256 1 minute 27 seconds
Description: Creates fairly blurry results. Even with post process sharpening. If anyone could get these results crisper it would be really improve the output.
a 3D render of a wizard by Gertrude Greene
a cubist painting of a Pokemon character
a cute creature
a matte painting of halloween by Carlos Trillo Name
a photorealistic painting of an alien landscape by Jacob Ochtervelt
a rough seascape filmic
a sea monster
a woodcut of a skull by Gu Hongzhong trending on ArtStation
Cthulhu
trypophobia
Name: Multi-Perceptor VQGAN+CLIP
Author: Remi Durant
Original script: https://colab.research.google.com/drive/1peZ98vBihDD9A1v7JdH5VvHDUuW5tcRK
Time for 512×512 on a 3090: 2 minute 30 seconds
Maximum resolution on a 24 GB 3090: 1120×480.
Maximum resolution on an 8GB 2080: Unable to run on 8GB VRAM
Description: As with the previous Multi-Perceptor CLIP Guided Diffusion scripts this one allows two different CLIP models to be used to rate the VQGAN output images. VQGAN is not going to beat diffusion for image coherance, but this script can give some very nice lighting and fine details in images.
a bronze sculpture of an evil clown made of clay by Dionisio Baixeras Verdaguer
a fantasy land by Shigeru Aoki
a hyperrealistic painting of puppies
a midnineteenth century engraving of the Sydney Opera House
a statue of reflective spheres
a surrealist painting of a tropical beach
an alien city CGSociety
an oil painting of a fire breathing dragon
computer rendering of a well kept garden by Norman Garstin ZBrush
war CryEngine
Name: Hypertron
Author: Philipuss
Original script: https://colab.research.google.com/drive/10fa8X6EsfZfda1dfhJ_BtfPZ7Te1WGoX
Time for 512×512 on a 3090: 2 minute 00 seconds
Maximum resolution on a 24 GB 3090: 1120×480.
Maximum resolution on an 8GB 2080: 256×256 1 minute 35 seconds
Description: Another VQGAN based script. Has various “flavors” to give different results. Works OK. Can give the “image in a sea of purple/grey” that previous MSE based scripts suffered from. Still worth a try.
a black and white photo of a fireman
a cute monster by Józef Mehoffer
a matte painting of a forest clearing
a pop art painting of a human
a renaissance painting of a ghost by Jan van de Cappelle film
a sea monster made of metal
a tattoo of a zombie
a watercolor painting of a dragon Flickr
an art deco painting of a haunted house by Mary Cameron
concept art of a mountainscape by Maximilian Cercha
Name: CLIP Guided Diffusion Secondary Model Method
Author: Katherine Crowson
Original script: https://colab.research.google.com/drive/1mpkrhOjoyzPeSWy2r7T8EYRaU7amYOOi
Time for 512×512 on a 3090: 2 minute 28 seconds
Maximum resolution on a 24 GB 3090: 1792×768 or 2048×640.
Maximum resolution on an 8GB 2080: Unable to run on 8GB VRAM
Description: A new diffusion based script from Katherine Crowson including a new “secondary model” she trained. Capable of some unique results with good textures and lighting.
a detailed painting of Fozzy Bear by LeConte Stewart
a flemish baroque of a happy person trending on pixiv
a flock of birds
a Ghostbuster CGSociety
a kitchen made of cheese
a nightmare creature
a photorealistic painting of The Grinch
a portrait of a woman
an art deco painting of a sad clown
an oil painting of a nightmare
Any Others I Missed?
Do you know of any other colabs and/or github Text-to-Image systems I have missed? Let me know and I will see if I can convert them to work with Visions of Chaos for a future release. If you know of any public Discords with other colabs being shared let me know too.
Jason.
Softology's Blog 