Since DragonBones is basically dead and the other 2 popular options are pretty expensive. I threw together a quick manual prototype. You align the layers and the bone pivots manually and save it to a json.
only the initial prototype
Iām excited to see your progress, looks amazing so far.
Here are some dumb super quick procedural generated animations for idle.
completely forget where i got the asset from. i gotta try a different asset perhaps a nude human character or simple clothes.
the slight gap in the sections appear to be an artwork issue as there is no underskin/undercoat layers, so when I manually split the asset into layers it creates a gap. i think it is from the master chief like power armor design.
so if i test a plain human who doesnt have weird armor sections the splitting into layers on that art work should be better
Iām sure weād all be interested to know more about how youāre actually doing all this.
But first, let me check, is this the progression?
- Make an external system where parts can be āpinnedā together and then given movement sequences
- Then output that to json
- Then write a GD extension to allow using the json
How are you doing 1.? As in what are you using? Are the gifs from step 1? Or are they already jsonified in your own editor? Or some external editor?
super long responseā¦
so i used a python script to split the 2D asset into around 13-17 pieces ( i forgot the total right now) . If i was to manually create a mask and separate it there wouldnt be art issues but I am looking for the automated solution.
so 2d asset to python for splitting body parts.
then the manual rigging and resemble app is written in javascript and rendered in html.
run a quick local python server to host the website app and then manually resemble the pieces and manually move the bone pivots around. kinda like you would do in blender for 3d but super basic.
then the next function of the web app is to create and export the json.
Gdevelop has nothing to do with anything technically other than my overall project protoype is in gdevelop. I doubt an gdevelop extension could achieve the part above I described.
Then I just used javascript to procedural animate the json and png files and create a gif to test.
If i was rendering this in gdevelop I would need a bunch more javascript to render it in the 2d pixijs engine in realtime.
this was more proof of concept.
So it is either a bunch of workarounds or buying Spine $$$$. My project not only do I want some procedural animation but I want to be able to see equipment change in real time, aim a weapon and have recoil. None of that can be done really with spritesheets unless you create a ton of them.
the solution is 2D bone rigging. The weapon or a helmet or a glove is just 1 piece attached to a section on the bone rig which can be changed real time to anything.
I am actually considering a move to Godot since it has an entire built in bone rigging and animation system for 2D. But I dont know C# or gdscript at all. And I certainly dont want to convert my javascript and I got no real time to learn. I did consider Fable 5 or codex just to refactor the entire thing to C#. Which is basically what those LLM were designed for.
My other option is to convert to PlayCanvas which still uses Javascript but it would basically be a 2.5D game where all the assets and rigging are in 3D you are just doing a camera trick to get a 2D isometric look and feel. PlayCanvas at least uses WebGPU to render so it could handle the 3D load. And I severely doubt my project even small could be rendered in Gdevelop WebGl only.
Right now I am just using Gdevelop as a rapid prototyping tool. Even if I stay in 100% 2D I will probably convert to CT.js which still uses pixijs like gdevelop. But it has webgpu as renderer and much better shaders, custom shaders and post processing. Probably a better scene editor and it has its own no code or limited code language. Maybe not as simple as gdevelop events but 95% easier than learning gdscript in Godot.
Now if you have money it is probably easier to use Spine and even switch to Unity or Unreal because even though those engine are complex you can basically buy any asset or game mechanic you need, even sub engine which dont require blueprints or unity no code, you are basically just manipulating fields.
even though its 3D it is what the SurrounDead lone developer did. It is an Unreal game and everything is an off the shelf solution. Users wanted to tow cars the dev bought the tow mechanic. While the HumanitZ devs do not have towing in their Unreal game because it is not native and kinda big deal to create from scratch. I think the SurrounDead dev sold over 50k units and is a full time active military member. But you are basically spending $1000s.
So i will probably finish using gdevelop to prototype some more systems to completion then move to CT.js if I stay 2D or if I want to go 2.5D move to PlayCanvas.
My little manual webapp is engine independent. It just spits out the png and the jsons. My goal is to always try and make every mechanicās logic completely isolated from rendering. So if you changes renders you only need to refactor small amount of code and not your entire mechanic,
Thank you very much for this. I am not an advanced user but I understood what youāve done and your motivations.
I had a look at ct.js and saw it had something at my level, catnip, and will have a play with it. I see their export options are currently limited to web and desktop with Android export in progress.
I hope you find what youāre looking for, good luck.
sorry long post agAINā¦
Well gdvelop has built in Electron wrapper which allows gdevelop to export to desktop and gdevelop has some old steam std hack/implementation to get steam hook ups.
However with all of the web only engines you are free to use electron and even better TAURI to wrap your game. You just do it manually but there are loads of tutorials and walkthroughs.
You can prompt chatgpt to learn the difference but Electron has a huge overhead. Which is problem for larger games. Gdevelop is taking a performance hit using WebGL then another performance hit using Electron to go to desktop.
I will have to check my notes and bookmarks but I know someone created an entirely new steam sdk hookup for game wrappers that is much improved from the old one that gdevelop uses. It is pretty recent like last 12-18 months. It really isnt gdevelop fault there are still many tutorials manually using the same mechanic but at least someone created a new version. steamworks sdk is just awful.
ah found it:
The other thing I wouldnt be afraid of is if you pay for Claude or ChatGpt you have access to their Agents Claude Code and Codex. Besides Unity and Unreal all of the engines I discussed are open source and have full documentation. So you can literally upload the engine source code, if they only web documention you use a webcrawler to grab everything and zip it into a file you can upload.
If using Javascript you want to prompt the Ai Agents to uses extensive and detailed jsdoc and make sure javascript is Es6+ standard. This really helps the Ai not imagine anything especially when you add the engine source code and documentation. Engines that use TypeScript are even better because those AI Agents are even better at TypeScript because its impossible to invent anything. Supposedly that are also terrific at C# but I have no experience.
Ai Coding Agents kind of get a bad name from people āvibe codingā on Youtube hey Claude create me a game like GTA 5 etc.
You must break down your game into sub systems which are fully modular and isolate logic from render. Also have a master project bible helps AI to stay on track. You then want to use AI to iterate systems with detailed project plans. The more debugging and the more validation you build in the better.
They are kinda scary good at the languages I mentioned. And if you have a process it does work.
I am coding a procedural dungeon generator and I also wanted a Tarkov like Tetris Inventory system. The popular engines no problem there are free ones to ones that cost $100+. Why reinvent the wheel.
Well I came up with a plan that will probably need 10-15 iterations and I am letting Codex do it. And it is a billion times fast me and probably 100 times better than me. I think within 1 day I did something like 5-6 iterations. Yeah that would have taken me 4-6 weeks.
Plus with game engines they handle a lot. The more creative type of stuff. If you just say need a system inventory or say RPG character creation system. I dont think a human can come close to the new Ai Agents. Probably why Claude Fable 5 was banned from the USA govnt they dont want any NON US citizens to have access to it whether they live in USA or not and USA citizens no access outside of USA. No idea how they are going to do that which ius why Fable 5 was immediately turned off.
The thing is Ai Agents handle certain languages extremely well and you can guide them to write really good comments and documents explaining the code. So it is still better to have the AI Agent write the code and then a human to go through it, read the comments and the documents to verify. Instead of writing everything from scratch.
That inventory system is probably 50k lines of code. This is why even for Indy devs Unity and Unreal are widely used. You just buy the system. And is it any different than using the code AI created? I mean there is absolutely no way if an indy dev buys a tarkov tetris system they are going to verify all the code manually. So they are blindly trusting someone who created it and sells it on the marketplace because they got 20 good reviews. How does anyone know if what they just bought is as good as Ai Agent system when done properly?
Where does the AI fail video games? Well the JUICE the TWEEN etc. But making systems that are mostly math yeah the paid stuff is insane.
Aka my SurroundDead example you can buy everything off the shelf but you still gotta provide the juice to make that game click. But if you need a procedural tree create yea Ai agent can easily do that.
When I am using Codex for the inventory system I have it setup for human visual validation in a Go No Go style. If it has 30 steps but on step 15 I need to visually valid a deb screen, an image, a json output or whatever, the Agent does not continue until I valid it and type Go. If type No Go then there is fallback to fix the errors, go through another Go No Go before ever getting to the next step.
big difference between free ai and paid ai and even paid ai vs Agent Ai. Codex Agent > ChatGpt 5.5 for example if they the same monthly plan gives you access to both.