Lovart
Design agent from LiblibAI (哩布哩布) that plans an entire deliverable set from one brief then builds it on an editable canvas. Ships as 星流 Agent inside mainland China.
Overview
LiblibAI, known domestically as 哩布哩布, was founded by Chen Mian, previously head of global commercialisation for ByteDance's CapCut. Lovart is the company's design agent, and the framing matters: this is not a text-to-image tool with a chat wrapper. Give it a brief and it plans a set of deliverables, then generates a logo, poster variants, packaging mockups, social crops and short video as a connected body of work. Everything lands on ChatCanvas, an infinite canvas where the assets stay editable, so correcting one element is a conversation rather than a regeneration of the whole batch. The company sits under 奇点星宇 and Bloomberg reported on 5 August 2026 that it was exploring a Hong Kong listing at a valuation in the billions.
The split product structure is the piece worth understanding if you work across the border. Lovart serves the international market in English. 星流 Agent, launched on 3 July 2025, is the mainland build, and it is not a translation. It handles Chinese font rendering and batch poster generation for domestic formats, which is exactly where imported tools fail: Chinese typography has layout rules that Western models were never trained to respect, and the difference between a usable 宋体 headline and a mangled one is immediately obvious to any Chinese reader. Pricing diverges too. Overseas starts at 15 USD a month for 2,000 credits with a Pro tier around 72 USD for roughly 11,000. Domestically it is 49 yuan basic and 99 yuan advanced, and the 99 yuan tier reportedly delivers over 70 percent more usage than the 15 USD entry plan.
Judge it on what it removes from a designer's week, not on any single render. For campaign work where you need thirty consistent assets rather than one striking image, the planning layer earns its keep. Against that, it is closed at every level: no weights, no self-hosting, no visibility into or choice over the third party image and video models each step calls, and no way to export the agent's plan into another tool. The two regional products do not share accounts or assets, so a distributed team picks one side and lives with it. And be sceptical of the growth numbers circulating in English, which range from 30 million to 300 million USD ARR depending on the source and on which corporate entity is being described.
Key Features
- ✓ Decomposes a single brief into a full deliverable set: logo, poster variants, packaging mockups, social cuts and short video, rather than returning one image per prompt
- ✓ ChatCanvas, an infinite canvas where generated assets stay editable in place and you correct one element conversationally instead of regenerating the batch
- ✓ Batch editing across a poster series, including one-click recolour and layout variants driven by natural language
- ✓ Chinese typography handled properly in the 星流 Agent build, with batch poster generation tuned to mainland formats and font conventions
- ✓ Two separately hosted products from one codebase: Lovart internationally in English, 星流 Agent domestically since 3 July 2025
- ✓ Wraps third party image and video models behind the agent, with credit cost varying by which model a step selects
Where it holds
- • Plans a campaign rather than producing a moodboard. The unit of output is a coherent deliverable set, which is a real change in how the work is structured
- • ChatCanvas keeps assets editable after generation, so a single wrong element does not cost you the whole batch
- • Genuine localisation on both sides. The 星流 build handles Chinese fonts and mainland poster formats properly, which Western models still mangle
- • Commercially solid. LiblibAI closed a large round in October 2025 and Bloomberg reported in August 2026 that the parent was exploring a Hong Kong listing, so it is unlikely to vanish
Where it breaks
- • Credit economics bite at the top. Pro runs 72 USD per month for around 11,000 credits, video generation drains them quickly and nothing rolls over
- • Entirely proprietary. No weights, no self-hosting, no export of the agent's plan and no choice over which underlying image or video models a step calls
- • Lovart and 星流 Agent are separate accounts on separate infrastructure. A team split between Shenzhen and overseas cannot share one workspace, and assets do not migrate between them
- • Revenue and valuation figures in English coverage vary by an order of magnitude between sources, from 30 million to 300 million USD ARR depending on who is counting and what entity they mean. Treat the growth narrative carefully
My Take
Most image tools give you one picture at a time. This takes a brief, plans the deliverable set, then produces logo, poster series, packaging mockup and short video on one canvas you can keep editing, which changes the unit of work rather than the quality of a single render. The pricing gap is the part English reviews miss entirely: 15 USD a month for 2,000 credits overseas against 49 RMB basic and 99 RMB advanced for the domestic 星流 Agent, with the RMB tier buying noticeably more usage. Chinese typography on the 星流 side is also handled properly, which sounds minor until you have watched a Western model butcher a 宋体 headline.
Quick Info
- Pricing:
- freemium
- Openness:
- Proprietary
- Starting at:
- Freemium with credit metering. Free tier for trial, paid plans from 15 USD per month for 2,000 credits, rising to a Pro tier at 72 USD per month for roughly 11,000 credits. Unused credits do not roll over. The mainland build, 星流 Agent, is priced separately in RMB at 49 yuan for the basic tier and 99 yuan for the advanced tier, and users report the 99 yuan tier buying over 70 percent more usage than the 15 USD overseas entry plan. Video generation drains credits far faster than image work on either side.
- Added:
- Aug 2026
- Updated:
- Aug 2026
Use Cases
Judge it on your own work
The notes above say where Lovart holds and where it breaks. The fastest check is your own workload.
Visit website ↗Alternatives to Lovart
Adobe Firefly
freemiumAdobe's creative AI workspace, now a router for 30-plus models including its own indemnified Firefly Image Model 5, wired into Photoshop, Express and an enterprise-only API.
Recraft
freemiumImage model built for design output rather than art, with true SVG vector generation, locked brand styles and an API covering vectorize, upscale and background removal.
Meitu AI / 美图AI
freemiumMeitu’s suite of AI-powered image and video editing tools, specializing in portrait enhancement, commercial photography, and design automation.