icon
News Community Store
Back to news hub

Tripo launches P2.0 and Aegentic Canvas at G-STAR, P2.0 supports 25,000 quads

Tripo launches P2.0 and Aegentic Canvas at G-STAR, P2.0 supports 25,000 quads

Key takeaways

  • P2.0 outputs quad-based 3D models up to 25,000 quads
  • Tripo announced P2.0 and Aegentic Canvas at G-STAR
  • Tripo supports biped, quadruped, fish, insect and flying creature rigs
  • Paid users’ outputs won’t be used for external publication or model training (NDA option)

Tripo showed P2.0 and its AI agent-driven Aegentic Canvas at G-STAR, pitching faster, editable game-ready 3D models that developers can refine inside a linked production workflow. The company says P2.0 outputs models built from quadrilateral faces—up to 25,000 quads—to make generated assets easier to edit, rig and animate for game use, while Aegentic Canvas ties model generation, part decomposition, texturing and rigging into a repeatable, agent-orchestrated environment.

At G-STAR we met Tripo’s head of product Jack Pan and Korea/Japan head Elli Wang to walk through how the tools are organized. Tripo Studio focuses on taking an idea image and turning it into a 3D model, and handles topology, UV unwraps, textures, rigging and animation in a single platform. Aegentic Canvas is presented as a production surface where AI agents infer the creator’s intent, assemble a work sequence, and let users control or store that sequence for reuse. Pan emphasized that Canvas splits models into parts—like head and torso—so teams can swap, tweak or reapply individual elements rather than redoing a whole mesh.

The company separates its model-generation tech into H-series and P-series families, and explained where each fits in a game pipeline. H-series targets tasks needing very high fidelity and fine detail, which typically produce higher polygon models that may be heavy for real-time engines. P-series is designed to reduce runtime and editing burden: it produces models composed of rectangular faces from the start, with fewer polygons and a topology conducive to downstream edits. Developers can choose H or P depending on whether they need dense sculpt-level detail or an editable, engine-friendly base mesh.

P2.0, the new P-series iteration Tripo brought to G-STAR, was described as a version aimed at delivering a practical balance of speed and editable quality. The company advertises model generation in under five seconds for a primary model, and clarified that the five-second metric refers to producing a first-pass, quad-based model that’s ready for follow-up work rather than a final, fully polished asset. The team said they paid specific attention to edge-loop flow: allocating more faces to areas that need more deformation or bone attachment, so rigs and animations behave predictably. Pan noted the work remains iterative—animatable topology and edge-loop placement are active research and will be refined across future versions.

Aegentic Canvas got a lot of time in the demo because it’s where Tripo tries to reduce the real pain points teams report. Canvas presents the generation pipeline as discrete, chainable steps: image input, part segmentation, per-part modeling, UVs, texture baking, rigging and animation. AI agents propose a task order based on an interpretation of the creator’s intent, but users can pause, edit, reorder or save that workflow for later. Saved workflows can be reused as templates: you can swap the initial image and run the same sequence against new inputs, or ask the agent to process a batch of images at once. The Canvas approach is explicitly meant to let teams make one workflow and apply it to many characters or items while keeping a consistent production method.

Style consistency across assets is handled in Canvas by letting a user upload a reference style image as an independent work unit. All generated outputs that should match the same visual language can be anchored to that reference, so a desert background, for example, can act as the stylistic anchor for NPCs, animals and props created within that session. The platform also supports applying an existing workflow and reference to other images—Tripo pointed out that you can take the exact steps used to create one character and apply them to another image, or ask the agent to run those steps across ten new images in one batch. Pan demonstrated that approach on-stage: pick a base image, lock the style reference, then expand parts and reuse the sequence to produce similar-looking assets quickly.

On rigging and animation support, Tripo said it covers more than just bipedal humans. Current support includes quadrupeds, fish, insects and flying creatures, and the team plans to expand toward other structures like vehicles as feature development continues. For non-human rigs the company intends to cover not only the main vehicle body but interactive elements—doors that open, and the animations of a person getting in and out. A planned feature will let creators upload reference videos alongside models so the generated rig and animation can follow the video’s motions. For asking the system to produce a specific motion, Tripo uses natural language prompts: creators can type a description of the desired action and the agent will apply that animation to the rigged model. Image-driven workflows remain available too: upload an image, let Canvas split components and link each to modeling and rigging steps.

Compatibility and toolchain integration were a clear focus. Tripo provides export choices so the generated model can feed into the developer’s existing pipeline, and offers plugins for tools such as Blender and 3ds Max to call Tripo functionality without leaving those applications. The intent is to let creators continue in their preferred environment while leveraging Tripo’s generation and Canvas orchestration. The company also highlighted real-world usage beyond games: Tripo cooperates with virtual YouTuber creators in Japan, and a Japanese creator published a workflow combining Tripo with other AI tools to build VTuber avatars. Elli Wang said Korean VTuber cases are fewer so far, but Tripo is watching cross-regional usage patterns and promoting similar workflows where they make sense.

On the question of time and cost, Pan was careful with numbers but provided a practical frame: Tripo aims to let the same person produce more output in the same time window. He contrasted a traditional game-model pipeline—where a single character model could take weeks to months—with Tripo-enabled workflows that can shorten that span to hours or days when combined with other tools. Pan stopped short of promising precise cost savings, but emphasized that the time reduction is the route through which teams should expect efficiency gains. The G-STAR demo area let visitors upload any image and generate a 3D model on the spot, illustrating how quickly a concept can be turned into a base mesh and a set of reusable steps.

Security and IP protection were topics developers raised at the show, and Tripo addressed them directly. The company’s policy states that paid users’ generated outputs will not be published externally or used for training other models. Customers using Tripo via API can sign NDAs that explicitly deny external publication and prohibit using those outputs for model training. Pan confirmed contractual and terms-of-service language govern how customer assets are handled, and that paid-user content will not be used for external promotion or learning datasets. Tripo runs support and feedback channels inside the product and via Discord, and in Korea the team runs Kakao Open Chat and a Naver Cafe to solicit direct input from local users; Pan said the company also performs in-person visits to Korean customers to collect on-the-ground feedback.

On market posture and competition, Elli Wang described a regional approach: Japan and Korea both use Tripo in games, but the patterns differ—Japan has stronger IP and animation-related usage while Korea is an active game market with its own adoption patterns. The team said they are tailoring marketing and product strategies to those market differences. When asked how Tripo intends to keep up with competing AI creation tools and engine-level generation features from larger vendors, Pan and Wang pointed to continuous technical releases and fast product iteration informed by user feedback. They also described pursuing collaborations with big tech and open-source projects where it makes sense, and iterating on tooling based on direct industry input rather than only on lab development.

Finally, the Tripo team encouraged hands-on engagement: they plan workshops, hackathons, offline events and more local activities to meet creators, and invited Korean developers to use the product and feed back issues or feature requests. Pan asked users to be vocal about friction points so the team can prioritize fixes, and Wang emphasized community channels as primary feedback routes. If you dropped an unreleased art board into their demo booth at G-STAR, the company said the NDA and paid-user protections are there to keep that art private while you test the generation pipeline.

Why it matters

Takeaway: Tripo’s P2.0 focuses on producing quad-based, editable base meshes (up to 25,000 quads) and Aegentic Canvas ties generation, part-splitting, texture work and rigging into a reusable, agent-managed workflow—both are pitched to shorten model turnaround from weeks or months to hours or days while offering contractual protections for uploaded assets.

Original source

Google News: ThisIsGame

Related headlines