sprite-gen is a command-line tool that takes a single character illustration and automatically produces sprite atlases and animation loops you can drop straight into a game. It has an image generation AI draw one row per pose, removes the background with chroma keying, cuts out the individual frames, and packs everything into an atlas accompanied by a manifest that records frame coordinates ─ all chained together through a single set of commands. It is developed by aldegad (South Korea) and released as open source under the Apache 2.0 license. It also ships as a skill that can be invoked from coding agents such as Claude Code and Codex, which makes it easy to fold into an existing game development workflow.
Key Features
- Component-row atlas pipeline: Processing runs through
prepare→gen-set→extract→compose-atlas. It generates one image row per pose, strips the background, cuts out the frames, and outputs a transparent atlas along with amanifest.jsonthat records frame coordinates - Automatic breathing animation (Breathe): It bakes a squash-and-stretch breathing motion into idle poses. The processing is deterministic, so the same input always yields the same result and repeated runs will not drift
- Seamless loops from video: It detects the cycle in a clip exported from a video generation AI (such as Grok Imagine) and cuts out a seamless loop while keeping the background transparent. A September 2026 update added processing that detects and corrects the chroma-key color cast that video models tend to paint onto the subject
- Batch color variations (Recolor): Simply specify a palette map and it produces variation sheets in bulk ─ recolored enemies, for example. This step is deterministic as well
- Web-based curation view: You can line up the generated candidates, compare them, reject the ones you do not like, and make small adjustments. Being able to watch the loop play before committing makes the hit-or-miss nature of AI output far easier to handle
- Game engine export: Post-processing commands support export formats for existing tools such as Aseprite, Phaser, and Flame
Pricing
| Plan | Price | What you get |
|---|---|---|
| Open source (the only offering) | Free | Apache 2.0 license. Full functionality with no restrictions |
sprite-gen itself has no paid plan, and there is no official pricing page. However, the part that actually draws the poses and video is delegated to external image and video generation AI services (Codex, Grok Imagine, and so on), so you will need accounts and pay usage fees on those services separately. In practice, your real cost depends on which generation AI you use and how much you run it.
Pricing information is current as of September 2026. Please check the official repository for the latest details.
Pros & Cons
✅ Pros
- One source illustration gets you both the atlas image and the frame-coordinate manifest, so you avoid the common situation where the assets exist but the work of wiring them into the game is still ahead of you
- Finishing steps that quietly eat time ─ breathing animation, mass-producing recolors ─ are automated, so even a small team can assemble presentable assets
- The curation view lets you filter results, easing the burden of digging through a pile of unusable AI output by hand
- Export support for Aseprite, Phaser, and Flame means you can slot it into a project that is already underway
- Being open source, you can read how the processing works and adapt it for your own engine
⚠️ Cons
- Image and video generation itself depends on external APIs. Even though sprite-gen is free, the generation AI side costs money in proportion to how much you use it
- It is offered only as a CLI and an agent skill ─ there is no GUI application. That is a hurdle if you are not comfortable working in a terminal
- It is a young project created in May 2026, and there are still few real-world case studies
- The developer describes quality checking for cyclic motion such as walking and running as experimental
Comparison with Similar Services
| Service | Form factor | How it differs from sprite-gen |
|---|---|---|
| sprite-gen | Local CLI / agent skill | Generates atlas and manifest from a single image. Free and open source, though generation AI costs are separate |
| PixelLab | Web service | Specializes in pixel art, with skeleton-based animation generation and Aseprite integration |
| Scenario | SaaS | Built for studios, reproducing a house art style by training custom models |
| Rosebud AI | Web editor | Generates sprites inside an editor that runs entirely in the browser |
Note: the characteristics of the three comparison services are drawn from summaries of comparison articles and have not been individually verified against each vendor’s official pages.
Who Is It For
- Indie game developers who can draw the source illustration but lack the time to mass-produce pose variations such as walking and attacking
- Anyone building games with Claude Code or Codex who wants asset generation to ride along in the same workflow
- Anyone who needs to prepare palette-swap variations in bulk, such as recolored enemies
- Developers on a custom engine who want frame coordinates as machine-readable data, not just images
- Anyone who would rather use a tool they can read and run locally than a closed SaaS
Summary
How close can you get to “draw it once and let the machine handle the rest”? That single question is what sprite-gen is working on. Background removal, frame extraction, recording frame coordinates, breathing motion, mass-producing recolors ─ steps that are simple in isolation but heavy once they pile up ─ are dispatched as a chain of commands.
That said, the drawing itself is done by an external generation AI, so output quality comes down to that model’s capability and your prompts. It is also a young project under active development (v2.3.x as of September 2026). The easiest way to judge whether it fits your production flow is to feed in one illustration you already have and see the atlas come out the other end.