After a few crazy weeks of training, testing and late-night builds… it’s finally live 🥹
Wireframe → Hand-Painted Texture GenAI Pipeline
https://t.co/OrHjgkpS9X
I built a GenAI workflow that turns 3D wireframes into hand-painted stylized textures inside Photoshop, using a custom-trained AI model 🚀
Huge thanks to:
@Scenario_gg & @emmanuel_2m. Without this platform, I honestly wouldn’t have been able to build the dataset
@Alibaba_Qwen for the base model (Qwen-Image-Edit)
@ModelScope2022 for the training platform!
Huggingface link on the ArtStation post!
This is all about helping artists iterate faster, not replacing them ❤️👊
In Q3 2022 we had 4 months of runway left. We burned the ships, pivoted hard from a failed product, let go of half the team ❤️🩹, and went all-in on… AI.
Fast forward to 2026: I'm live demoing @Scenario_gg to 200+ people at @supercell's HQ in San Francisco.
Our product is now used by some of the world's top gaming studios and creative teams. 3x YoY growth - and accelerating every month.
Two takeaways for anyone building:
1. Don't die.
2. Never give up.
Qwen-Image-Edit-2511 is here!
We’ve pushed the boundaries of consistency and control. A massive leap over the 2509 version, specifically built for real-world workflows.
Model:
🤗 Hugging Face: https://t.co/5TWncOSkXk
🧭 ModelScope: https://t.co/n97aasRwad
After a few intense days of testing, training, and way too many late-night experiments… it’s finally here: Wireframe and Concept to Texture 360° 🎨
Full breakdown + examples: https://t.co/nya5r1kPMQ
My new GenAI LoRA that turns mesh + concept art into a consistent hand-painted base texture — in under a minute!
But the big upgrade?
It now supports full 360° turnarounds thanks to ultra-stable multi-angle outputs 🔄🚀
This is all about helping artists speed up the first pass, not replacing their craft. More time for polish, less time fighting the blank canvas!❤️👊
Again, huge thanks to:
@Scenario_gg & @emmanuel_2m. Without this platform, I honestly wouldn’t have been able to build the new version of this dataset!
@Alibaba_Qwen for the base model (Qwen-Image-Edit 2905)
@FurkanGozukara for the training parameters, tips, and in-depth YouTube tutorials!
@Dimsane @Ludwig_Dresch Dimitri, I spent some time thinking about what you said and I find it genuinely interesting. I think you have a point that I might still be struggling to fully understand
If you don’t mind, I’d love to talk to you via DMs, but only if that feels comfortable for you
Ludwig! How are you?
First of all wow, what an incredible list of games you’ve worked on. I love Spyro with all my heart. I used to play it as a kid, and I absolutely loved the remaster
But back to your point: I’m sorry you feel that way, and I mean that 100%
If it helps at all, I’m also making real art! I practice traditional fundamentals to improve as a real artist, and I call it “real” because, as I said in another comment (and in more depth in my post), for me AI is not art
Something I personally enjoy a lot is creating real art by drawing small robots, but what I also especially love is improving through courses. Some of my favorites are Thiago Klafke’s Environment Art Mastery and Marc Brunet’s Art School. But I also want to save up to take the CGMA course: Creating Stylized Game Assets
I do real art, not at your level, but I’m trying to improve so I can reach that level someday, it will take me years, but I’m in no rush
@JogurtA@Magsolid1 Hi Gui! What you’re saying is completely understandable, and you’re absolutely right. But as I mention in the post, I didn’t use anyone’s work. The dataset I created to train the model is made from personal content and AI-generated material :)
Hi Dimitri and Ludwig! Sorry for the late reply (I was sleeping hehe)
I don’t want to assume, but I think I understand what you mean. I do believe I created something, which is the AI model (the LoRA)
But what I think you’re referring to is the thing the AI “does”. I want to make it clear that, for me, the AI itself doesn’t create anything
Near the end of my post I talk in depth about my philosophy regarding AI, but in short: AI is not art. It never will be. It’s a mathematical equation that predicts colors from a noise image. What’s artistic about that? For me, nothing
It’s the same as a Smart Material in Substance Painter, and just like a Smart Material, it doesn’t give you a perfect or final result. It only helps in the first phase so you can focus on polishing details afterward
Humans make art, AI doesn’t. I invite you to check out what I wrote in the post! :)
And of course, as I also mention in the post, my DMs are open if at any moment you guys feel like sharing your own thoughts with me. I promise I’ll respect your opinions and really try to listen so I can understand your point of view!
Hi Magsolid! :)
Good point. In general, it’s true that an AI can generate content relatively quickly (not necessarily good, but fast). However, in this case I was referring to the training process. I should have clarified that better
I trained “my own AI” (a LoRA) so it could produce this type of image consistently, and that took several days (especially because many models didn’t give me the quality I wanted, and it was all trial and error). I go into more detail in the ArtStation post if you’re interested! :))
PS: Cool art! I loved the pixel art ones the most (I have a weakness for that style)
Hi!!
I didn’t fully understand what you were referring to, but in my ArtStation post I talk about the everything in depth. If you were referring to the awful topology of the model, I’m 100% with you, in the post I explain that it’s an AI-generated model! (but works to show the process)
PS: I LOVED your Dragon Ball artworks. I’m from Argentina, so Goku is basically a national icon here
Hey Averse! Good question! :)
Let me explain, aside from the level of consistency required for 3D game assets, the issue is that concept art is never identical to the final 3D model. Sometimes the structure changes, but more importantly, the style itself shifts. It becomes a sort of 2.5D hand-painted look (in this case)
To achieve these results, you could build a workflow using an advanced ControlNet, but also a LoRA for the style! And that’s why I preferred to simplify the work into a single good LoRA that can do everything at once with a simple workflow!
After a few crazy weeks of training, testing and late-night builds… it’s finally live 🥹
Wireframe → Hand-Painted Texture GenAI Pipeline
https://t.co/OrHjgkpS9X
I built a GenAI workflow that turns 3D wireframes into hand-painted stylized textures inside Photoshop, using a custom-trained AI model 🚀
Huge thanks to:
@Scenario_gg & @emmanuel_2m. Without this platform, I honestly wouldn’t have been able to build the dataset
@Alibaba_Qwen for the base model (Qwen-Image-Edit)
@ModelScope2022 for the training platform!
Huggingface link on the ArtStation post!
This is all about helping artists iterate faster, not replacing them ❤️👊
@Kiko52888 Thanks! This model was made specifically for the hand-painted style, but it shouldn’t be an issue to train a new style. You just need the right dataset. If you read the post, I explain everything in depth!
You think I didn’t consider that? I actually have two very good reasons for that
1. I didn’t consider that, at all 😂
It’s a great idea!
2. At first, I wanted to avoid using ComfyUI as much as possible, since the workflow is meant for artists with no AI experience, but it really is a good idea! I see some difficulties, but I want to see how I could implement it. Thanks! :)
Hey my dude, I want you to feel heard. I’m not going to say “I understand how you feel”, because that would be condescending of me, but I do want you to know that this model is absolutely not made to replace artists. Not at all!
I also want you to know that my DMs are always open if you’d like to talk more. I promise I’ll genuinely listen :)
P.S. I’m not totally sure how X works (this was my first post), so I’m not sure if my DMs are actually open. If they aren’t, let me know and I’ll set them up so we can talk if you feel comfortable