Since the release of Cryptopunks and Bored Apes, most blue chip NFT collections have used the same recurring concept for their creative baseline: An illustrated portrait (also known as a PFP, for profile picture or picture for proof) altered into a few thousand variants made slightly different by using a random distribution of customization layers. These unique traits of various rarity levels can affect face shapes, eyes, skins, outfits, accessories, backgrounds, etc.
One of the purposes of The Smurfs’ Society Legendary collection was to set new standards for PFP collections. But how do you transcend a genre while respecting its codes? And how do you push the qualitative boundaries while voluntarily keeping an automated/procedural process that implies limited human impact when pairing traits?
The concept: Using the iconic Smurfs IP as a canvas to create an unprecedented collection
The Smurfs' relatively homogenous and easily recognizable form made for an ideal backdrop for our creativity. From day one, we made some bold choices in order to deliver exceptional quality with our collection:
The full-body frame: A major differentiator compared to traditional PFP collections: The Smurfs NFTs are full-body instead of classic PFPs that are only focused on the face/torso. This allows for a lot more details and a full range of poses, adding huge variation and personality to the collection.
250 unique characters: Another major difference and a colossal amount of production work compared to classic NFTs: The Smurfs’ Society Legendary collection offers 250 unique characters. Each of them with its own pose, characteristics, and accessories. For comparison, CloneX has only 8 DNAs, and even these have pretty much the same body pose.
The full 3D render: Even though the Smurfs are originally a 2D comic character, the most recent films and TV shows are in 3D, so it felt more modern to work in full 3D. We recreated all of the characters entirely from scratch, accounting for dozens of hours of work for each Smurf.
A wide range of traits: Hundreds of possible traits in 7 categories, including backgrounds, environments, elements, eye colors, special FX, skin, and outfit colors… With more than 62,000,000 possible combinations.
Artists collabs: Rather than designing the collection with a single famous artist or relying solely on an in-house team, we decided to collaborate with a variety of contemporary artists. From backgrounds and skins to actual Smurf characters, our guest artists add another layer of rarity and diversity to the collection.
Implementing these details took an incredible amount of time and effort. But it’s time we know was well invested in making a truly unprecedented collection, simultaneously drawing on an instantly recognizable character while making room for exceptional creativity.
How we chose and calibrated the rarity traits in order to create a coherent collection: It all starts with a Spreadsheet.
As surprising as it may seem, most of the hard creative work was accomplished at the brainstorming and testing level: defining the different categories of traits in the collection, choosing how many variations each category would have, and setting the creative range we would push ourselves to achieve.
We had two teams working at the same time: one in charge of the modelization and the pose of the characters and another in charge of creating the patterns and colors that would customize the 50 variants for each Smurf.
Julien Probst, a co-founder of Iron Velvet Studio, obsessed with Web Design & 3D rendering, and the mastermind behind our trait engineering, was in charge of the latter: “I basically spent 2 months testing renders, skins, and textures to see how they would fit together. I would regularly share experiments with the team to select the most exceptional traits and include them in the final renders.”
The next step was to list all the selected traits in a huge spreadsheet and give them a rarity score that would define their probability of being distributed in the final set of PFPs. There are several reasons behind this evaluation, driven by objective criteria rather than subjective reasons:
-
Consistency with the Smurfs’ Lore (Smurfs are traditionally blue with white hats and pants)
-
Perceived rarity (A gold trait is culturally seen as more precious)
-
Production cost (some traits like the Wave background are long and costly to render)
-
Community’s anticipation (Some items and accessories were long awaited)
-
Special added value (Some famous artists have designed rare traits, for instance)
-
Artistic direction consistency (at the end of the day, it’s a creative choice)
That’s a lot of criteria for a lot of traits. Imagine that for each given Smurf, there are 271,320 possible combinations of traits. Our job was to get the engine to generate only 50 variants that all look amazing!
Once all the traits were chosen and their rarity level defined, we still had to execute the creation of the legendary collection.
The tools and processes used to generate the final images of the collection.
Once all the rendering characteristics were defined, it was time to convert the dream into (virtual) reality. We’ve used several different tools to create the collection:
-
The Smurfs are modelized (creation of three-dimensional representations) and posed (prediction of the transformation of an object from a user-defined reference pose), with Autodesk Maya.
-
The texturing (creation of texture or skin patterns) is done with Adobe Substance 3D Painter.
-
The Final Render (applying textures onto the models, light setup, special effects, and generation of the randomized traits) was made with Blender Cycles, while traits were randomly combined with the FZRandomizer plugin.
-
Finally, the files are sent to a render farm. 3D file generation is a resource-intensive process: We used external servers with high computing power to generate thousands of images in the collection.
Each step is important. But the real magic happens in Blender, where we literally load all the material, 3D models, and textures and implement rarity tickets to each item using the open-source plugin FZRandomizer. Then we create a timeline in the project where the software automatically creates thousands of new frames with a different PFP using the probability tickets.
We had to run several batches of tests to verify that the trait repartition conformed to our plans and that the quality of the renders was perfect: Over 25,000 lines were generated in CSV files during the entire process in order to define the right formula. This helped spot potential problems and set exceptions to the trait distribution process:
-
When several traits don’t go well together for esthetic reasons (styles or colors that don’t match) or for 3D conflicts (a background and a VFX, for instance)
-
When traits from more than one collab were combined in one single PFP (Druillet + Marest, for instance)
Once everything was ready, We exported CSV files (you know, these unreadable comma-separated lists) for our tech team to transform into metadata for the NFTs and individual render files for the render farm to process the final PFP images, which take about 25 full days (TBC).
We hope everyone can see and understand how much work has gone into this collection. There has been an incredible amount of planning behind each Smurf variant, making them truly unique works of art. But even after the final images have been processed, we’re still not finished! Once we’ve ensured that our collection is well-calibrated and looks amazing, we have to make sure the PFP files will be fairly and randomly distributed to our PFP holders. But that’s a story for our next article!