Closed wencxxxxxx closed 2 months ago
impossible to say without looking at what you are exactly doing
impossible to say without looking at what you are exactly doing
like this, What's the problem, please? thank you.
that is correct. you are removing over 50% of the image and keep a weight of 1.0.
you need to fill what is not covered by the clip vision with some other conditioning (another IPAdapter for example or stronger text prompt) and/or lower the weight.
I see. The meaning of masking is not to select a specific part of the reference image to affect the sampling result, but to blacken a certain part of the image to get a new reference image to affect the sampling result. Why not change the meaning of the mask to select specific parts of the reference diagram to affect the sampling results, and the other parts do not affect the sampling results.
When I used the mask of the IPAdapter Encoderd node, the generated image turned yellow and gray, which was not the normal sampling result