Original software · Technical expertise · Ongoing creation
AI IMAGE WORKSPACE
ZR Asset Manager
An asset management and creation workbench that turns AI-generated content into reusable creative projects. An infinite canvas with 20+ node types, covering end-to-end AI image processing: generation, editing, background removal, object removal, inpainting, outpainting and upscaling.
At a glance
AI layered PSD: split and fill in layers automatically, then export an editable PSD
Extras like Real-ESRGAN upscaling, TTS and voice cloning
MCP integration with 40+ tools; AI clients can drive the canvas directly
A creative pipeline on an infinite canvas
Image, audio, text and generation nodes connect into a flow on one canvas, lay themselves out, and rerun whenever you want.
ZR Asset Manager infinite canvas workbench structure: asset list on the left, node canvas in the middle, properties panel on the right. The image below is a generated-image example, not a screenshot.
Interface mockup. Assets, canvas and properties sit side by side in three panels, and every call leaves a full audit record.
End-to-end AI image processing
From a single prompt to a deliverable asset, every step in between has a canvas node.
Image generationMultiple models · 1K/2K/4K
Image editing and retouchingInpainting · masks
Smart background removalIncludes color background removal
Object removal and repairRemove clutter in one stroke
OutpaintingExtend the frame freely
UpscaleReal-ESRGAN 2x/3x/4x
Color grading and croppingBatch style consistency
Grid slicingSlices go straight to the library
Image-to-promptImage to text
Prompt optimizationAuto-enhance before generating
AI layered PSD
Exclusive · built in-house
Split a flat image into an editable layered file automatically: cut out the foreground, split layers by transparent connected regions, fill in what each layer had hidden, rebuild the background, then export a standard PSD. The designer gets a source file they can keep working on, not just a picture.
Cut the subject out automatically
Split layers by transparent connected regions
Fill in what's hidden, layer by layer
Fill in the whole background
Export editable PSD
More than image processing
MCP integration, 40+ tools
AI clients such as Claude and Codex can drive the canvas and start generation jobs directly, with every call written to the audit log.
TTS and voice cloning
Text-to-speech and voice cloning nodes, multi-angle camera orbits, pixel art generation and other extensions keep arriving.
Every call is auditable
Every generation, edit and export is written to the log, so you can look up who touched which image and when.
Project gallery
Projects can be shared publicly, with comments, reporting and backend moderation. A good workflow deserves to be reused by more people.
Connect any model
OpenAI GPT models, Doubao Seedream, Google Imagen / Gemini, and any OpenAI-compatible endpoint.
Web and desktop
Use it right in the browser, or install the Windows desktop build. Switch creative environments as the situation calls for it.
OpenAI GPT modelsDoubao SeedreamGoogle ImagenGeminiOpenAI-Compatible1K / 2K / 4KAdaptive ratio
FAQ
Do I need to supply my own model API?
Both work: use the built-in model presets, or connect your own OpenAI-compatible endpoint and API key and keep control of models and usage.
How much control do I have over layered PSD results?
The layering process runs step by step with a preview at each step, and once exported as a standard PSD you can keep editing it in Photoshop or similar. For complex images, run the background removal node first to confirm the subject edges.
How do AI clients connect to MCP?
Generate an MCP access credential in account settings, point clients like Claude or Codex at the service address as described in the docs, and you can call 40+ tools. Every action is written to the audit log.
A high-performance desktop infinite canvas for graphics editing, node orchestration and AI workflows. Design drafts and workflow drafts as dual engines, plus a Figma-style unified tabbed workspace — let ideas happen on the same canvas.
At a glance
Dual-canvas core: design canvas plus node workflow, with automatic layout
Developer center: CLI and node SDK, with hot-reload debugging for custom nodes
Community: find drafts, plugins, nodes and Skills in one place
Six core capabilities built for the creative workflow
Dual-canvas core
Design drafts come with two views, quick mode and canvas mode; workflow drafts focus on node orchestration, with one-click auto layout.
Unified tabbed workspace
Figma-style frameless window, fixed Home and Community entries, document tabs restored after restart. Run several projects at once without losing your place.
Developer center
Build custom nodes with the CLI and node SDK, debug with local hot reload, and publish to the community after static checks and sandbox validation.
Community
Find drafts, plugins, nodes and Skills in one place, with favorites, likes and author pages, so good tools stop being scattered.
Multilingual support
The app and content site support multiple languages, including help, tutorials and legal text. Available languages depend on the current release.
Enterprise support and compliance
Built-in live chat and a structured reporting flow; motion degrades for accessibility, and the small details hold to workspace standards.
See what it looks like
Workspace home mockup: workspaces and shared lists on the left, a wall of recent project cards on the right
Recent projects at a glance. Design drafts and workflow drafts both start from the same home screen.
Node canvas mockup: layer list on the left, node graph in the middle, properties panel on the right
Two engines, a design canvas and a node workflow. Connect nodes to orchestrate, and the layout tidies itself.
Developer center mockup: CLI terminal output on the left, registered node list on the right
Bring your own nodes to the canvas: write, debug and publish, three CLI steps.
FAQ
Which operating systems are supported?
The current release targets Windows first, and the installer is a Chinese WiX package. It's built on Tauri 2; for other platforms, watch for release updates or ask on live chat.
Which languages are supported?
The interface and product copy support multiple languages. Follow the system language or choose one in Settings; available languages depend on the current release.
Can my team use it together?
Yes. Workspaces support shared projects and a shared component library, with version history you can step back through. For team and self-hosted deployment, ask on live chat.
Can I build my own nodes?
Yes. Use the official CLI to scaffold a node template, debug with hot reload in watch mode, and once it clears static checks and sandbox validation, publish it to the community for everyone to use.
An asset management and creation workbench that turns AI-generated content into reusable creative projects. An infinite canvas with 20+ node types, covering end-to-end AI image processing: generation, editing, background removal, object removal, inpainting, outpainting and upscaling.
At a glance
AI layered PSD: split and fill in layers automatically, then export an editable PSD
Extras like Real-ESRGAN upscaling, TTS and voice cloning
MCP integration with 40+ tools; AI clients can drive the canvas directly
A creative pipeline on an infinite canvas
Image, audio, text and generation nodes connect into a flow on one canvas, lay themselves out, and rerun whenever you want.
ZR Asset Manager infinite canvas workbench structure: asset list on the left, node canvas in the middle, properties panel on the right. The image below is a generated-image example, not a screenshot.
Interface mockup. Assets, canvas and properties sit side by side in three panels, and every call leaves a full audit record.
End-to-end AI image processing
From a single prompt to a deliverable asset, every step in between has a canvas node.
Image generationMultiple models · 1K/2K/4K
Image editing and retouchingInpainting · masks
Smart background removalIncludes color background removal
Object removal and repairRemove clutter in one stroke
OutpaintingExtend the frame freely
UpscaleReal-ESRGAN 2x/3x/4x
Color grading and croppingBatch style consistency
Grid slicingSlices go straight to the library
Image-to-promptImage to text
Prompt optimizationAuto-enhance before generating
AI layered PSD
Exclusive · built in-house
Split a flat image into an editable layered file automatically: cut out the foreground, split layers by transparent connected regions, fill in what each layer had hidden, rebuild the background, then export a standard PSD. The designer gets a source file they can keep working on, not just a picture.
Cut the subject out automatically
Split layers by transparent connected regions
Fill in what's hidden, layer by layer
Fill in the whole background
Export editable PSD
More than image processing
MCP integration, 40+ tools
AI clients such as Claude and Codex can drive the canvas and start generation jobs directly, with every call written to the audit log.
TTS and voice cloning
Text-to-speech and voice cloning nodes, multi-angle camera orbits, pixel art generation and other extensions keep arriving.
Every call is auditable
Every generation, edit and export is written to the log, so you can look up who touched which image and when.
Project gallery
Projects can be shared publicly, with comments, reporting and backend moderation. A good workflow deserves to be reused by more people.
Connect any model
OpenAI GPT models, Doubao Seedream, Google Imagen / Gemini, and any OpenAI-compatible endpoint.
Web and desktop
Use it right in the browser, or install the Windows desktop build. Switch creative environments as the situation calls for it.
OpenAI GPT modelsDoubao SeedreamGoogle ImagenGeminiOpenAI-Compatible1K / 2K / 4KAdaptive ratio
FAQ
Do I need to supply my own model API?
Both work: use the built-in model presets, or connect your own OpenAI-compatible endpoint and API key and keep control of models and usage.
How much control do I have over layered PSD results?
The layering process runs step by step with a preview at each step, and once exported as a standard PSD you can keep editing it in Photoshop or similar. For complex images, run the background removal node first to confirm the subject edges.
How do AI clients connect to MCP?
Generate an MCP access credential in account settings, point clients like Claude or Codex at the service address as described in the docs, and you can call 40+ tools. Every action is written to the audit log.
A new-era AI code editing tool and editing framework. A VS Code style prompt engineering workbench: Monaco editor, file tree, dependency view and a full Git workflow, so prompts get managed as engineering.
At a glance
Monaco prompt editor plus a drag-and-drop workbench layout
Full Git workflow: stage, diff, commit, branch, publish to GitHub
Dependency views and variable tables make prompt structure clear at a glance
A familiar editor, new objects
File tree on the left, Monaco editor in the middle, source control on the right — only this time, what you're managing is prompts.
ccks editor mockup: file tree on the left, prompt editor in the middle, Git source control panel on the right
Interface mockup. The workspace layout supports drag-to-rearrange, panel minimizing, light and dark themes, and switching between Chinese and English.
Treat prompts as engineering
Monaco prompt editor
The same editing experience as VS Code: syntax highlighting, multiple cursors, find and replace. Long prompts stay easy to work with.
Full Git workflow
Init, stage, diff, commit, pull and push, branch management, right through to publishing on GitHub, all from the interface.
Dependency view
See how prompts reference and combine with each other, so you know the blast radius before you refactor.
Project import and export
Import and export a whole prompt project as a ZIP, and drag external files straight in. Migration costs nothing.
Variable tables and validation
Template variables managed in one place, edited in a table and checked against Zod rules, so prompt parameters stop scattering.
Customizable workspace
Activity Bar, drag-and-drop layout, minimizable panels, light and dark themes, and Chinese/English switching. The workspace follows your habits.
Use cases
Team prompt library
Collect prompts scattered across documents and chat logs into one repository: consistent format, reviewed changes, traceable versions. A new hire can pick up the team's whole AI asset base in a day.
Agent and workflow development
Manage system prompts, tool descriptions and few-shot examples for agent projects. The dependency view shows what references what, and Git branches let you run A/B iterations safely.
FAQ
How is ccks different from a regular code editor?
ccks is designed specifically for prompt engineering: variable tables, dependency views and template validation are things a general editor does not have, while the VS Code-style editing and Git experience stays.
Where is the data stored?
Works with both MySQL and SQLite: use MySQL for centralized team deployments, or run SQLite as a single file for personal use. You can also push to your own remote Git repository at any time.
Can it be self-hosted?
Yes. ccks is a standard Next.js app and can be deployed to your own server or an internal network. Talk to us on live chat for the details.
An AI image workspace for your pocket. Pick a model, set the quality, choose a ratio, and turn an idea into a picture in a few steps; work is filed into your space automatically, and every image remembers its own settings.
At a glance
Model presets, with quality and size set on one screen
1K / 2K / 4K output, aspect ratios from 1:1 to 21:9
Traceable creation history, with one-tap regeneration from the original parameters
Three tabs, one creative flow
Space holds your work, Create starts a job, Me manages your account — the restraint a mobile app should have, one tab per job.
YanArt Space screen mockup: a 3x3 grid of work
Space · Everything you generate is filed by time automatically; long-press to share or edit it again.
YanArt Create screen mockup: model, image quality, size and ratio settings
Create · Model, quality, size and aspect ratio all fit on one screen; change a parameter and generate right away.
YanArt image detail screen mockup: preview and parameter settings
Details · Every image keeps its own settings, reusable in one tap, so you stop screenshotting prompts to remember them.
An image generation flow rethought for the phone
Every setting on one screen
Model, quality, size and ratio all fit on one screen, set within thumb reach, no flipping back and forth.
Multiple models
Presets for a range of image models, each with its own style and cost, and you can switch models in settings as they get updated.
1K / 2K / 4K output
From social media images to print-grade assets: three resolution tiers and ratios from 1:1 to 21:9, right the first time.
Traceable creation history
Every generation keeps a full parameter record. To reproduce an image from two weeks ago, open the details and run it again with the original settings.
Work archived to your space automatically
Everything you generate goes straight to your library, browsable in a masonry grid, with sharing and export. Your phone album stops filling up with half-finished work.
Usage stays visible
Remaining credits and usage records are laid out on the Me page, so you know what a run costs before you start it.
Use cases
Quick creation and illustrations
A picture that came to you on the commute, the image you're missing before you post, an illustration you suddenly need for a bedtime story — take out your phone and generate it, no waiting until you're back at a computer.
Mobile companion to desktop work
Try models and ratios quickly in YanArt, then take the parameters you liked back to ZR Asset Manager or Pixel Spark for retouching and batch production.
FAQ
Which mobile platforms are supported?
An Android package is available now; an iOS version is planned. Contact live chat for the current build and how to install it.
Where are generated images stored?
Your work is saved in your account space and can be exported to your photo library or shared at any time. Once deleted, no copy is kept.
How does it relate to ZR Asset Manager and Pixel Spark?
The same AI imaging stack in different forms: YanArt handles quick creation on mobile, ZR Asset Manager handles asset storage and retouching, and Pixel Spark handles workflow orchestration on the infinite canvas.
A showcase and sharing site for AI work. Generating is just the start — the work needs somewhere to be seen: collecting, arranging, search and review, one complete content pipeline.
At a glance
Responsive masonry gallery with progressive loading
Third-party token submission, wired straight into your generation pipeline
Review routing, tag search and object storage
A gallery site you can run for years
More than hanging images on a wall. Submissions, review, categories, search and rights info: one complete content pipeline.
Masonry gallery
A responsive masonry layout with progressive loading, so thousands of images still scroll smoothly, mobile included.
Third-party token submission
External tools and automated flows can submit directly with a token, no account needed, which suits wiring into a generation pipeline.
Review routing
Submissions enter a review queue and only go public once approved, so compliance and site tone stay under control.
Tags and search
Archive by tag, model and style, so finding an image from three months ago no longer means scrolling.
Backed by object storage
Images are served from object storage over a CDN, so the site itself stays light and bandwidth cost and load speed are easy to control.
Parameters and sources on record
Each piece can carry its model, parameters and author, so the creative context stays with the display.
Want a gallery site of your own?
Also A Service · custom builds available
The structure behind AI Gallery — submissions, review, masonry display, object storage, tag search — can be rebuilt around your brand and business: new domain, new visuals, new review rules. The content stays yours.
Log in to submit on the site; for automated cases you can request a third-party token and submit programmatically. Submissions enter a review queue first and go public once approved.
Can you build one under our brand?
Yes. That falls under our custom software development work, and everything from visuals and domain to review rules can be adjusted to fit. Describe your case on live chat and we'll come back with a plan and a quote.
Complete software delivery from architecture to deployment, across desktop, web and cross-platform applications.
Software development
Complete software delivery, from architecture design to production ops. Desktop is our strength — our flagship product Pixel Spark is built on Tauri 2 + Rust.
Desktop appCross-platform Tauri / Electron desktop apps with custom windows, system integration and auto-update
Full-stack webNext.js / Nuxt full-stack sites and admin panels, with database design and deployment delivered together
Cross-platform appsMini programs, mobile H5 and uniapp cross-platform builds
InfrastructureDocker containerized deployment, object storage, message queues and ops monitoring
Tech stack we use
We pick the stack for the project, not for our comfort zone. Below are the ones we use most day to day; less common directions are open for discussion too.
Bring generative AI into real business workflows, from product positioning and prototypes to working software.
AI software design
Product design and engineering that puts generative AI into real business processes. We have built complete AI image pipelines, infinite canvases and asset management platforms, so we know which traps to avoid up front.
AI image processingFull pipeline design and implementation for generation, editing, background removal, outpainting, upscaling and layered PSD
AI workbenchInteraction and architecture for infinite canvas, node orchestration and batch task queues
Model integrationModel presets and OpenAI-compatible API support, with billing and auditing covered end to end
From 0 to 1Positioning, prototypes, MVPs and roadmaps for AI products
Skills and toolchains for agents, connecting business systems to callable and composable AI tools.
AI skill development
Skill and toolchain development for the agent era. Our products already expose 40+ tools over MCP to AI clients such as Claude and Codex, and that experience transfers straight to your business.
MCP tool developmentWrap your business systems as standard tools any AI client can call
Agent workflowsMulti-step agent design and tool orchestration on frameworks like LangGraph
Prompt engineeringPrompt asset management, evaluation and iteration method (see ccks)
Plugin ecosystemCustom nodes, a plugin SDK and a developer platform
Build corporate and brand websites that can be maintained over time, from information architecture to deployment.
Delivery scenarios
These six areas are the work we do most and know best. If yours is not on the list, talk to us anyway — we may have built something similar, or we will just tell you we are not the right fit.
Corporate and brand sites
From information architecture to deployment, we build sites you can maintain for years, not one-off templates.
Changchun PixelFlash Technology Co., Ltd., a product team growing through in-house software and development services.
Company profile
Changchun PixelFlash Technology Co., Ltd. builds software, designs AI software and develops AI skills, and runs two lines of business: In-house software + development services.
On the in-house side: from our flagship infinite canvas Pixel Spark, to ZR Asset Manager, an AI image toolkit covering generation, editing, layering and upscaling, to the prompt engineering workspace ccks, and on to YanArt on mobile and the live AI Gallery — five product lines that connect to each other and make up a complete creative stack.
On the services side, we take on custom development for corporate websites, admin back ends, WeChat Mini Programs, desktop clients, mobile apps, AI integration and data visualization. Unlike teams that only deliver, we are first of all heavy users of our own products: every capability we offer outside is polished and proven on our own product lines first, and every solution we hand a client is backed by a product that actually runs.
Our products support multiple languages for creators and teams in China and abroad. Available languages depend on the current release of each product.
Explore more
ZR-AI single sign-on portal
One account across every product line: a single login and authorization hub with third-party OAuth support.
One account · service authorization
MCP toolset and agent skill packs
Canvas and image processing wrapped as MCP tools, so AI clients can call and orchestrate them directly.
From early product experiments to desktop, mobile and publishing tools, every release adds practical experience.
How we got here
2025
Product line incubation
Several product lines, ZR Asset Manager and ccks among them, are incubating in parallel. The AI image pipeline gets refined through real creative work, while client projects build up engineering experience alongside it.
2026-08
Pixel Spark ships
The flagship product moved from 0.0.5 to 0.1.0 quickly: dual canvas core, unified tag workbench, developer center and a four-language content site all shipped.
2026-09
Mobile apps and showcase sites
YanArt, the mobile AI image app, and AI Gallery, the showcase site, both went live. Creation moved from the desktop to the road, and from there to being seen by more people.
Future
Still growing
The product lineup has taken shape, but it is far from finished. Our own products keep going deeper, our services keep going wider — each line feeds the other.
Use what we build, attend to the details, and keep data and projects transparent and controllable.
Three things we believe
Good enough for us first
We only release products we use ourselves every day. If a feature can't meet that bar, we'd rather not ship it.
The details show we mean it
A one-pixel stroke, a 140-millisecond transition, a fallback for one failed call — the craft is in the details.
Transparent and under your control
Call logs can be audited, data can be exported, and delivered source code belongs to the client. Users have full visibility and control over their own data and projects.
Agree on scope, milestones and acceptance criteria, then deliver working software in clear stages.
Project-based delivery
Clear scope, clear milestones, clear acceptance criteria. We demo a running build at each stage and hand over full source code and documentation after acceptance. Good for standalone projects with clear boundaries.
How we deliver
01 / Requirements
Understand your problem
Tell us what you need on live chat or by email. We reply within 2 business days with a requirements list and a first read on feasibility.
02 / Solution design
Sketch the solution
Tech choices, UI prototypes, milestones and a quote. Once the scope is agreed, we sign and start.
03 / Build and delivery
Deliver in small steps
We demo a running build at each milestone, so problems surface early and get fixed early. After acceptance you get the source code and documentation.
04 / Ongoing support
Launch is just the start
Bug fixes, feature iterations and ops support, so the product grows with the business.
Scope and ownership
Clear boundaries up frontThe plan, scope and acceptance criteria go in writing before work starts. Mid-project changes follow the same process; nothing runs on a verbal tally.
The source code and docs are yoursOnce the project is accepted we hand over the full source code, deployment docs and ops notes. No technical lock-in.
If we can't do it, we'll say soWe don't take on hardware drivers, blockchain token issuance, or work that needs a special licence, and we won't say yes first and figure it out later.
Monthly or quarterly collaboration for product improvements, fixes, operations and technical advice.
Long-term technical support
Ongoing monthly or quarterly work: feature iterations, bug fixes, deployment and ops, and technical advice — like having a product and engineering team on call.
On business days someone usually picks up the same day, and you get a clear answer within 2 business days at the latest. If the requirements are complex, we'll book a 30-minute call to pin down the scope before writing up a plan.
First conversation is free
Work starts after the plan and scope are confirmed in writing
Take a look first
Not sure whether to reach out? Look at what is already live: ZR Asset Manager, ccks and AI Gallery are all open to use right now.
You do not need a full spec. A few sentences on what problem you want to solve, who will use it and when you want it live is enough; we will work out the rest together. The first conversation is free.
How do you price a project?
Project work is quoted once, by scope and milestones; long-term support is a monthly or quarterly subscription. We give a clear cost breakdown at the proposal stage and confirm any new requirements separately along the way. No vague quotes.
Who do I call if something breaks after delivery?
Every project ships with a warranty period we agree up front: defects fixed free. After it ends you can move to long-term support, with the same team still on it. No disappearing act after delivery.
