CivArchive
    ← All articles
    Published November 5, 2026by arsibalt

    Moss-Image and fast illustrations

    80 views4 reactions0 comments on CivitAI0 collected
    moss-imagecomparative studyillustration

    Don't sleep on the new Moss-Image model as a base for your illustrations - there are a ton of good things about it, and speed alone may be a great reasons to consider it as an alternative to both Krea 2 and Anima (both of which I still use for illustrations and value a lot)

    Moss-Image has a growing rep among photo enthusiasts for a "real photo" look with diverse faces. Some with that particular focus only use Moss-Image Base instead of Moss-Image SFT. To maximize quality they may also avoid DMD (lightning) LoRAs. The post-training of SFT and the speed gains of a DMD LoRA risk sliding toward an "AI look".

    They have a point. Here is the god Thor working in a coffee shop, according to Moss-Image Base at 50 steps, or Moss-Image SFT with an 8-step DMD LoRA:


    If your goal is candid photography style, one of those may definitely be worth the extra diffusion steps!

    But in this and many other instances, the rules of photography don't perfectly apply to illustrations. Is one of the following better than the other? Do you know which rendered in 30 seconds over 50 diffusion steps, and which in 4 seconds over 8 steps?

    Are they both a little confused about espresso machines?

    Moss-Image SFT with an 8-step DMD LoRA can produce very high quality illustrations very fast. The post-training of the SFT model makes it better than Base at handling text. The things it supposedly loses - some diversity of face shape, subsurface scattering of light through skin, a bunch of other important photo thingies - are just not as big an issue in most illustration styles. It's almost as good at following prompts as Krea 2. It's about as fast as Anima when both are turbo'd up, and it has Anima or Illustrious amounts of seed diversity too - so it's both possible and practical to run a batch, find fun and different expressions of your prompt, and pick one you like best. It's cheap to run even for folks with low-VRAM cards.

    I could go on, but you're late for an appointment making a bunch of illustrations with Moss-Image :)

    Happy illustrating!

    Archived from CivitAI · Updated November 5, 2026View source