Open Discussion
I just finished watching the recent RL webinar, and it got me thinking. Flux has come a long way — we’ve got Flux Dev, Flux Krea, even ControlNets starting to show up. It’s exciting tech and definitely worth keeping an eye on. But after experimenting with both, I keep coming back to SDXL, especially since I am running everything on consumer GPUs.
For me, it comes down to three things:
- VRAM → Flux is powerful, but it’s heavy, built for server infrastructure, except the 5090. On a 12–24GB consumer card, you bump into walls fast. SDXL isn’t lightweight, but it’s much more practical to run locally without always relying on the cloud (opinion).
- Licensing → Flux licensing is… rough. SDXL is still wide open for commercial use, and that matters if you plan to build on it professionally.
- Ecosystem → SDXL’s ecosystem is massive: LoRAs, ControlNets, refiners, VAEs, and a deep community of tools. With the right nodes, SDXL can get very close to Flux (and sometimes surpass it) without enterprise hardware.
I’m not trying to convince anyone to switch — Flux has clear strengths, especially in coherence and context. But I do think it’s worth pointing out that SDXL + the right nodes is still a strong option for artists who want to work locally, on their own hardware.
That’s where I’ve been putting my energy: building ComfyUI nodes that solve SDXL’s weak spots — safe resizing (no more weird vertical lines), ControlNets that respect CC/iClone renders, and my ComfyUI-SDXL-Adherence node to keep things on-prompt.
A bit of context about me:
- Programming since 2001
- Python since 2009
- Go since 2024
- AI tools since 2023
- ComfyUI custom nodes & tools since 2024
So this is me scratching my “TD itch” — building workflows that make Render AI more reliable for everyday artists, not just those with $10K GPUs or always-on cloud subscriptions. If you are a production house, of course, either the pro-level cards or cloud servers make sense…but for us mere mortals… It’s kind of hard to justify
Curious what others here think: after the webinar, are you leaning more toward Flux, or sticking with SDXL locally?
4 Likes
I attended the webinar (the one organized by RL), but I must admit that it seemed fairly complicated. Part of it is that I haven’t done much with AI models, except what is included in some online services that I use.
So in that sense, I don’t really have a preference. But what you propose seems more suitable for an everyday artist like myself. I’m very interested in the developments, but none from what I saw in the webinar was “turn-key”.
So I think what you are proposing, working with a model that can run locally and that is further enhanced by you to make it specifically suitable for AI Render would be very valuable.
I’m not dismissing other models outright, because the demands for animation are more stringent because it requires consistency between frames.
I have used iClone since 2006 (almost 20 years!) and I like it. There are things missing but overall the program has served me well.
What iClone especially misses is the ability to create large environments and simulations like water etc.
So if AI Render could be developed into a hybrid approach where the characters are animated but the look would be AI-enhanced with a particular style. Then the environment like a city for example could be indicated with a prompt or so but would be AI generated.
Anyway, I’m just throwing out some things here.
1 Like
@animagic
I watched the AI Render webinar afterwards, which caused me a lot of headaches.
A lot was hinted at, but unfortunately not shown step by step. I don’t have much experience with AI.
I think it’s good that RL wants to improve rendering with AI.
But the many different platforms that were shown.
And above all the nodes, oh my God.
If you could do everything directly from IC8 or CC4/CC5,
that would also be an option for me.
Regards, Robert
1 Like
This is the new SDXL straight out of iClone locally. I didn’t even try hard and the quality is insane.
I am an animator 2D & 3D.
I was initially excited about Reallusion AI render but so far I see no impressive examples of HD animation that would make this a viable alternative to me just rendering in Blender or just using
something like GROK for animating
Still NPR images with fairly consistent backgrounds.
