CivArchive
    ← All articles
    Published September 9, 2026by bvga

    How to Boost Image Detail in ComfyUI (for Anima and Illustrious)

    446 views15 reactions2 comments on CivitAI4 collected
    generation guide

    I haven't been satisfied with the standard quality of my images in Illustrious and Anima. When characters are close-up they look good, but as soon as they started pushing further out the textures could become flat and details could become mushy. On Illustrious I struggled to get proper looking full-body shots because of this.

    A common approach people use is running an "ADetailer", which automatically detects and masks features like character eyes and runs a higher resolution pass over them to improve detail.

    However, what I needed wasn't just a feature or two fixed. I needed an entire rendering pass to enhance details and textures across the whole image.

    After a bit of research and experimentation I found this was achievable. I don't know why this isn't talked about much; I had seen a lot of people mention ADetailer, but until I actually went searching for it I hadn't heard anyone mention how to do full detailing passes, or just how much this improves the quality of images.

    The concept is pretty simple: You upscale the latent or image and then feed it into another KSampler with lower noise. What this will do is focus on refining the detail of the image at the higher resolution.

    However, how to set that up properly is technical. You need to select how many steps, the denoise amount, what sampler and scheduler you want. You can chain multiple KSamplers together, each with their own configuration. And all of this needs to be balanced just right, otherwise you'll get grainy noise or ghosted textures.

    But as you can see here with Anima, the boost in quality is significant.

    I've also picked up on a few additional tricks:

    I see most people using a normal KSampler and its denoise setting. However, if you use the advanced KSampler instead with its start_at_step function, you can get nearly the same effect in much fewer generation steps.

    With Anima I've also found that you can use a high-and-low sampler approach (picture what Wan 2.2 does), placing a latent upscale between the samplers. This improves the detail a fair amount. And then you can additionally pass it into a denoised sampler to push the detail to the max.

    I included this sampler approach in my newest Anima workflow, which is what I personally use to generate images.

    I've also put together a basic workflow attached to this article, which contains the detailing approach I've been using for both Illustrious and Anima. I'm sure the settings could be further optimized, but it works pretty well as-is.

    Attachments

    Archived from CivitAI · Updated September 21, 2026View source