[ LOADING SYSTEM ASSETS... ]
Anomaly detected on Base
3,000 generative PFPs born from digital DNA.
STATUS: INCUBATINGDNA: 98.4% STABLE
â—Ź REC
Connect Wallet to Register
Supply3,000
Price0.0033 ETH
ChainBASE
PROJECT_DATA
Sulkicks is a PFP project that integrates drawing, handcrafted 3D modeling, and artificial intelligence into a single creative process. Each piece begins as a hand-drawn sketch and is transformed into a 3D sculpture—the genetic seed used to train a custom AI model (LoRA).
From 20 seed sculptures, the system generates 3,000 algorithmic variations, expanding the project’s visual universe. Unlike traditional PFPs built from predefined layers, Sulkicks is grounded in a concept of digital DNA, allowing each character to emerge as a unique organism with its own identity within the collection.
GENETIC_ENGINEERING
In this process, AI is not the end but the means: a tool to amplify human sensitivity and extend the medium’s expressive possibilities. The LoRA model learns, encodes, and expands the genetics of the original seeds, functioning as a digital recombination laboratory.
(Sulkicks bases made in manual 3D modeling)
01. Innovation
We merge traditional art with cutting-edge technology.
02. Uniqueness
Each Sulkick is a one-of-a-kind digital organism.
03. Community
Join an ecosystem of creators and collectors.
SEQUENCE_INITIALIZED
Production Process
01
Stage 1: Hand Drawing
Character design begins with manual sketches that enable freer, more spontaneous expression. Unlike 3D—which can become rigid—starting with drawing helps capture each character’s sensitivity and personality from the outset.
AUTO_SEQUENCE
01 / 13
02
Stage 2: 3D Sculpting
Each sketch is transformed into a handcrafted 3D sculpture, with a focus on detail, textures, and lighting variations. The result is a final render that becomes a training datapoint for the dataset. The goal is for characters to be very different from one another yet coherent in style, maintaining diversity within a strong visual identity.
Seed Specimens
03
Stage 3: AI Training (LoRA)
The elements and features present in the 3D renders become valuable training data. The LoRA model extracts shapes, proportions, and details capable of generating new combinations that did not exist directly. Every detail operates as digital DNA—a genetic trait that combines with others to form the core of the seed model, guiding the creative direction while preserving the artist’s sensitivity.
DNA Dataset
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
3D Sculpting Input
MIX
Generative Result
Trained Model Output
Anomaly Detection // Human Intervention
Base Model
Generic / Boring / AI cliché
Human Work
Trained Model
Expressive / Sensitive / Alive
These images were generated with the same prompt and the same seed. The difference is that one is loaded with human work and creativity, while the other is pure automation with the aesthetics of corporate base models.
04
Stage 4: Generative Prompt with Wildcards
With the 20 seeds trained, the system runs in a node-based environment that dynamically generates prompts using wildcards.
Modular structure
segmented by species, accessories, garments, hairstyles, backgrounds, morphologies, etc.
Scale of variability
each section includes hundreds of variables and a huge volume of traits, enabling near-infinite variability in the creation of each piece.
Probabilities and rarity
the system is programmed with probability weights and rarity levels, so very rare and hard-to-obtain Sulkicks can appear.
Unique generation
each image is produced with a new prompt assembled from the wildcards, avoiding the exact repetitions typical of other PFPs.
BATCH #07 / RESULT #1
GENERATIVE PROMPT
RECOMPILING...
SPECIES:
HEAD:
FACEGEAR:
EYEWEAR:
HEADWEAR:
NECK:
TORSO:
BACK:
BACKGROUND:
LOADING GENERATIVE ASSETS...
0%
05
Stage 5: Curation
From the thousands of generated results, a meticulous manual selection is performed to arrive at the 3,000 Sulkicks that make up the final collection. This process ensures quality, coherence, and artistic judgment, discarding variations that do not meet the project’s standards. This human-in-the-loop approach guarantees that only the strongest and most expressive characters reach the collection on OpenSea.
3,000
FINAL PIECES
.png)
Why It's Different
Digital DNA vs. fixed layers
identity does not arise from stacking layers but from recombining AI-learned visual genetics.
Real variability
the combination of wildcards + traits + probabilities and rarity produces authentic, non-repetitive diversity.
Sensitive authorship
AI expands the artist’s decisions rather than replacing them, maintaining aesthetic direction and original sensitivity.