I used the cover artwork from Tom Dimuzio's "Markoff Process" cd as input to a one shot image to image resynthesis algorithm using Versatile Diffusion, which is a new hot off the press multimodal model that was released this week. An amazing improvement over the CLIP image embedding guidance transfer learning retrained Stability Diffusion model i've been using for this.
The rerun of Tom's promo photo on his bandcamp page using this new one shot image resynthesis approach is shown below. You can compare with the old technique here. Note that all i was able to get out of the old approach for the Markoff Process artwork was basically noise, again showing off the strengths of the new approach. Although it needs more control over generative variability, a little too tight for some uses.
No comments:
Post a Comment