Product & community case study

Loomi — AI Creative Canvas

A tree-based creative software concept for material-guided, iterative exploration.

Role: Prototype & Concept — Team とあるパンダ
Period: 2026
Status: Completed
Stack: AI/ML · Product Design · Prototype · Multimodal

Loomi concept interface showing one source draft and three material-guided visual directions — User-supplied concept interface: one source image branches into frosted-glass, stained-glass and dark reflective directions. This illustrates the interaction design, not a measured product outcome.

One-line summary

A tree-based, AI-assisted creative canvas that retrieves material attributes such as texture and brightness from a database, applies them to new drafts, and preserves alternative directions for iterative exploration.

At a glance

01Problem: Creators collect fragments of inspiration, but most generative tools jump directly from a prompt to a finished output.
02Context: An AI-and-art hackathon concept developed with Team とあるパンダ.
03My contribution: Co-developed the product concept, interaction model and prototype narrative.
04Outcome: A product concept and prototype narrative connecting a reusable material database, attribute-guided remixing, candidate comparison and a tree of iterative creative branches.

Three-minute walkthrough

The missing middle in AI creation tools

Creative work often begins with colours, textures, images, phrases and unfinished fragments. A prompt-to-output interface skips the slower work of collecting, comparing and recombining those materials, even though that process is where a creator's own direction develops.

Loomi problem framing — Figure 1. Creators may have abundant inspiration but no connected workspace for turning it into an evolving body of work.

The complete workspace: materials, visual relationships, sketch and directions

The studio concept brings the material library, visual relationships, current sketch and candidate results into one workspace. Selected references contribute different qualities, while the sketch remains the structural anchor. Material controls specify the role and influence of each reference; the creator can adjust the intent, compare directions and use the next result as a starting point for further exploration.

Loomi overall studio concept with a material library, visual relationship map, current sketch, candidate results and material-guided generation controls — Figure 2. User-supplied overall workspace concept. Materials guide texture, colour and mood while the sketch anchors structure; the interface joins reference selection, visual relationships and iterative result controls. This is a concept visual, not evidence of deployed usage.

A creative exploration tree, not a disposable sequence of prompts

Loomi treats a draft as the starting node of an exploration tree. A remix creates alternative directions; selecting one opens another round of refinement while keeping the other branches available. The creator can compare directions and return to an earlier idea instead of losing it when the next image is generated.

Loomi exploration-tree concept interface with a source draft, material variations and further branches under the selected stained-glass direction — Figure 3. User-supplied concept interface. The tree preserves the relationship between a draft, its alternatives and subsequent refinements; the indented branches make the iteration history visible.

A material library that becomes more useful over time

Loomi stores inspirations as reusable material in a database. The concept retrieves attributes such as texture, brightness, colour, line and composition, then uses those qualities to guide inspiration for a new work. Inspiration Cards expose possible relationships between fragments, so the model proposes connections that a creator can accept, ignore or reinterpret.

Material Library and Inspiration Cards — Figure 4. Saved material is organised and connected without forcing the user to commit to a finished generation.

Transfer selected qualities, rather than copying an entire reference

The remix concept lets the creator select inspiration material and choose which qualities to carry into a draft: colour, line, texture, light or composition. The interface also offers an influence strength and an optional instruction. A painted mask limits the affected area; without a mask, the change applies to the whole draft. This connects material-library attributes to an explicit creative choice rather than treating a reference image as an all-or-nothing style instruction.

Loomi remix concept interface with a draft, selectable material references, colour line texture light composition options, influence strength and optional instructions — Figure 5. User-supplied concept interface, preserved in Japanese. The menu separates colour, line, texture, light and composition; the controls specify scope, influence and instructions before proposing three alternatives.

The model proposes; the creator decides

Draft integration brings selected materials into a working canvas, where AI-generated variations support comparison rather than silently becoming the final artwork. The intended division of labour keeps taste, selection and authorship with the creator.

Draft integration and interactive variations — Figure 6. Variations are presented as material for judgment and refinement, not as an automatic final result.

What the concept demonstrated

The hackathon output is a product concept and prototype narrative, not evidence of adoption. It demonstrates how one workspace could let a material library, a set of ideas and the user's working relationship with AI grow together.

The concept interfaces make the intended interaction concrete, but they do not establish database implementation details, a numerical definition of brightness or texture, retrieval accuracy, generation quality or deployed usage. The material attributes described here are product inputs, not validated performance metrics.

Three Loomi growth loops — Figure 7. The concept connects the growth of materials, ideas and human-AI collaboration.