[Run on threadripper2]
[Run on threadripper2]
In[]:=
StableDiffusionSynthesize=ResourceFunction["https://www.wolframcloud.com/obj/nikm/DeployedResources/Function/StableDiffusionSynthesize/"];
threadripper2:
In[]:=
$NetModelPath="/home/sw/Dropbox/GeneralBox/Research/LLMScience/FromNik";
In[]:=
sdOptions=Normal@<|TargetDevice->"GPU",AssociationMap[FileNameJoin[{$NetModelPath,"Parameters/Stable-Diffusion-V1",#<>".wlnet"}]&,{"Unet","Encoder","Decoder"}]|>;
Green-eyed cat
Green-eyed cat
In[]:=
(*greeneyedcat*)latent=;textEmbeddings=;greenEyedCat=textEmbeddings->latent;
Test run
Test run
In[]:=
BlockRandom[SeedRandom[26];StableDiffusionSynthesize["A cat in a party hat",sdOptions]]
Out[]=
The “Pet Store”
The “Pet Store”
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,22,30}]
Out[]=
26
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,31,36}]
Out[]=
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,37,42}]
Out[]=
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,43,50}]
Out[]=
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,51,60}]
Out[]=
56
In[]:=
Table[ClickToCopy[BlockRandom[SeedRandom[n];StableDiffusionSynthesize["A cat in a party hat",sdOptions]],n],{n,61,70}]
Out[]=
61
In[]:=
BlockRandom[SeedRandom[777];StableDiffusionSynthesize[Table["A cat in a party hat",5],sdOptions]]
Out[]=
Diffusion Steps
Diffusion Steps
In[]:=
BlockRandom[SeedRandom[26];StableDiffusionSynthesize["A cat in a party hat",sdOptions]]
Out[]=
Modifying the Net
Modifying the Net
Modifying the Embedding Vector
Modifying the Embedding Vector
Random Embedding Vectors
Random Embedding Vectors
Random Sequences on the Input to the Text Embedder
Random Sequences on the Input to the Text Embedder
[[ Attempting to batch ]]
Conclusion: it is possible to do 11 diffusions in parallel
Conclusion: it is possible to do 11 diffusions in parallel
Many “non-human” results look basically like 2D CA patterns
Many “non-human” results look basically like 2D CA patterns
The signature of the creator : human forms start showing up....
All the things a human-trained brain could think about .... Though it might not think/choose to
All the things a human-trained brain could think about .... Though it might not think/choose to
Things one could imagine imagining .... but don’t have words to describe...
[ What prompts is it closest to? ]
What human-trained brains could imagine if they chose to...
What human-trained brains could imagine if they chose to...
[ Do they have free will about their embedding vectors .... or do they simply evolve on paths defined by other things? ]
Random positions in meaning space reflect what is imaginable to us humans....
Random positions in meaning space reflect what is imaginable to us humans....
Faces look so weird to us because we’re more sensitive to those...
Faces look so weird to us because we’re more sensitive to those...
Richard’s experiment
Richard’s experiment
Deterministic Embedding Vectors
Deterministic Embedding Vectors
Images are different:
Simple Initial Condition
Simple Initial Condition
Interpolated Embedding Vectors
Interpolated Embedding Vectors
E.g. red green apple