Good afternoon yall,

I’m having trouble generating fliers, greeting card interiors, and the like. The model reeeally wants to generate the text but I reaaally need it not to so that I can layer it in, in post. Any tips, workflows, prompts, etc you all would recommend?

Thanks in advance!

Edit: I’m working with: SDXL Base + Refiner, DPMM2++ Ancestral comfyui I’m using the same prompt for L and G for both positive and negative and am obviously using text and words in my negative prompt (using a weight of 1.3)

  • tal
    2·
    1 year ago

    Put the token “text” in the negative prompt? Maybe “writing”, “signature”, “watermark”, “label”, etc.

    EDIT: A maybe less-shotgun thing to try: use an image-editing program to cut out just the portion with the text. Install the Clip Interrogator extension in Stable Diffusion. Take the cut-out label, and insert it into img2img in Stable Diffusion. Then use Clip Interrogator. That should give you a list of tokens that will generate something like the image that you’re seeing. That’ll probably give you a better idea than my guesses as to what tokens are associated with the text there. Maybe it’s “sign” or something like that.

    • stevedidWHAT@lemmy.worldOP
      1·
      1 year ago

      Yeah it’s likely to do with it being a “flier” the results are so good, it’s just the text generated is so bad. Hate to waste so much of it but it’s pretty non -workable. I’ll try out that clip interrogator. I think overall enrichment or smarter prompting will likely get me there. Thanks for commenting!