Wan 2.1 Text-to-Image!
Who knew Wan 2.1 was an absolute beast at generating stunning, single-frame text-to-image outputs? Well… you do now.
Originally trained for rendering video, Wan 2.1 wasn’t meant to be a full-blown T2I model - but it turns out, this thing absolutely slaps when it comes to creating high-detail, expressive, and stylish compositions from simple prompts. Anime scenes, suggestive portraits, or moody cinematic stills, Wan 2.1 brings an uncensored edge and a surprising amount of depth, lighting finesse, and expressiveness to every gen.
This model card contains everything you need to get started using Wan 2.1 as an image generator with ~12 to 16 GB VRAM, utilizing GGUF models. You will need some custom ComfyUI nodes, so make sure ComfyUI is up to date, and pull those in with the Comfy Manager!
Description
Q5_K_M GGUF
FAQ
Comments (18)
Quite pleasantly surprised by the renderings of Wan 2.1, which is supposed to do T2V, I find it performs significantly better than some T2I models, these images even seem more realistic than those from Flux1.Dev.
It runs well on an 8 GB VRAM/32 GB RAM configuration, even faster than Flux.
I obtained good results with DPM++ 2M Karras 20 steps.
Thank you for sharing this original approach.
how it faster than flux if we said we run both at 20 steps for same resolution ?
@amazingbeauty faster than flux with the same number of steps, and higher resolution
Yeah, many things are better than flux. Illustrious and NOobAI are worlds better. No idea why people ride Flux so hard.
@DaddyWolfgang i can use it with the txt2vid wan 14b ? just pointing to a single frame ? which model best 720p or 480p ?
Yep I knew, That's why I am trying to improve Wan to be the ultimate AI gen, T2I, T2V, I2V and V2V
I have this error:
No clue why those aren't working, but you don't really need the Clean VRAM Used and Clear Cache All nodes - just connect the image output from Fast Film Grain directly to the Save Image.
This is probably the best model for creating high‑resolution images with great detail. For now, I’ve put Flux on hold.
Will LORAs work with it? Like the Wan LORAs?
They do!
theally Thank you
are there character loras or just video loras?
I get this error:
'ModelSamplingAdvanced' object has no attribute 'log_sigmas'