Basic Workflow Simple Prompt Testing: Krea2 vs MOSS vs Qwen2.1 vs Flux3
retained text
Finally had some time today to tinker around with the latest open source releases, put together a gallery of results. Each model/prompt combination got 3 tries across different seeds and I picked the best of each for this post. Mostly using default ComfyUI workflows and settings, some basic prompts with some style checks (see comments for full prompts) **Krea2** ***(turbo, 12 steps, 1 CFG, euler, beta)***: solid all-around model, innately knows a lot of styles. Old news to some, but I haven't had much time to play with it. Weird patterns in many images (especially if prompting for grain), but plenty of lora options already that help a lot with that. I'm sure better workflows exist too by now. **MOSS** ***(25 steps, 2.5 CFG, er\_sde, simple)***: imperfect, has some flaws and total duds, but there's just SOMETHING there, like some artistic flair that's been missing in most image models. excited to spend more time with this one, definitely has potential. **Qwen 2.1** ***(25 steps, 1 CFG, euler, simple)***: only T2I in this gallery, prompt adherence is solid and results are polished although overbaked. outside this in limited editing scenarios I've had pretty good luck with it so far too. **Flux3** ***(50 steps, 3.5 CFG, euler, simple)***: Wow, absolutely blew me away how garbage this is. I guess a few turned out if you yearn for the aesthetic of AI images circa 2023. May be top tier if you want to dabble in horror generations
document fetched 2026-10-20 00:32 · canonical https://reddit.com/r/StableDiffusion/comments/1x9oi6x/basic_workflow_simple_prompt_testing_krea2_vs