Illness is an Audit
Illness is an unwelcome audit. A week of forced deceleration across the entire Valeon stack, and what you discover is that most of what felt urgent turns out not to be, and most of what keeps things standing is invisible until it stops being. Progress slowed — the visible kind, at least. The internal work continued.
The majority of effort across plutarc, VocaSync, and the core Valeon infrastructure went into hardening. Improving bot and worker resilience, tightening the processes that sit beneath the surface, building out employee tooling so that support staff can audit, diagnose, and help users navigate tickets with actual visibility rather than inference. None of the products have staff yet. That is a future problem being solved in the present, which is the right time to solve it. The Valeon authors dashboard received the same treatment — incremental, internal, deliberately unglamorous.
VocaSync also received comprehensive documentation written in the plutarc style: sequential, navigable, honest about edge cases. Anyone who has used an audio processing service and found the documentation either absent or written as a monument to the developer rather than an instrument for the user will understand why this matters. Documentation is not an afterthought. It is the first layer of feature transparency, and it has been overdue on VocaSync.
The more substantial changes happened in object0 and gramatic.
object0's original interface was functional in the way a command-line wrapper with a coat of paint is functional. It did the job. It did not feel like a product. The interface has now received the plutarc treatment — custom theming, deliberate component choices, a visual coherence that the initial implementation simply did not have. The goal was always for object0 to feel like a proper desktop application rather than a web form that happened to touch an S3 bucket. It is considerably closer to that now. Further refinements are planned, but the foundation is right and the worst of the old ugliness is gone.
gramatic received comprehensive documentation alongside six new additions: three models and three tools, all oriented toward processing and transforming existing images as much as generating new ones.
On the model side: Proteus (1 credit) handles natural-language anime and illustration — vivid characters, clean linework, stylised scenes, responding well to plain descriptive prompts. NoobAI XL (2 credits) is a specialist anime model aimed at peak character fidelity, tuned for Danbooru-style tag prompts rather than natural language, and the quality ceiling it reaches for that particular use case is noticeably higher. Kontext Edit (4 credits) is the most significant of the three: it takes an existing image and a text instruction and edits — scene changes, style shifts, object replacement — while preserving what should be preserved. The distinction between generation and editing is one that matters more as workflows mature, and Kontext Edit is the beginning of that capability on gramatic.
On the tool side: Restore Faces (1 credit) repairs and sharpens faces in degraded or low-quality images without losing likeness. Colorize (1 credit) converts greyscale and black-and-white photographs to realistic colour. Ghiblify (2 credits) transforms a photograph into a soft, painterly Studio Ghibli-style illustration. These three tools operate on existing images rather than generating from scratch, which is a meaningfully different use pattern — iteration on something real rather than conjuring something new.
The most architecturally interesting addition is prompt templates. Templates allow users to define reusable generation prompts with named tokens — placeholders that are populated at request time by passing parameters in the request body. In isolation this is a convenience feature. Paired with the new batch endpoints, it becomes something more: a single template with multiple parameter sets will run generation for every combination automatically, without manual re-queuing. If you are producing variants at scale — for testing, for content pipelines, for systematic style exploration — the batch endpoint is where gramatic stops being a generation tool and starts behaving like infrastructure.
The Canva integration, currently pending marketplace review, closes a loop that should matter to anyone producing content at volume. The pattern is: generate in gramatic, send to Canva for layout and composition, send back to gramatic for a polish pass through a secondary model with image input capability. The round-trip is the point. Neither tool is trying to replace the other. What you get is generation fidelity and design precision in the same workflow, without the usual export-reimport-start-again cycle. Thumbnail production for channels running at any real volume is the obvious immediate application.
The showcase page was built using gramatic itself, which felt like the right kind of proof. It is also worth noting that some of the generated examples in the showcase are deliberately unflattering. Where a model produces poor results in a specific use case, the output is left as-is with an advisory rather than quietly removed. The honest version of a showcase includes the failures. That is what the limitations section is for.
More to follow as the marketplace review resolves and further refinements ship.