Not a bad haul for today, with learning the AI called Flux Kontext. I learned how to…
* Speed up the slow Flux Kontext x 2 (turbo LoRA for 12 steps rather than 24, no noticeable difference in output).
* Combine two images into a new prompted composition. That was a feature I hadn’t yet investigated. Got a cat and dog running along a beach (a stress-test it just about managed, from two random stock photos), but the more mundane use would be two talking heads for comic-book yak-yak dialogue. Poser can do this anyway, but the widescreen Poser render you might ideally need for that might not be suitable for input/output in Kontext.
* Zoom the camera in and out in Kontext via a LoRA + prompt, while keeping the central character fixed and any background more or less similar. Again, you’re duplicating what Poser can do anyway in a render, but it’s good to know how to do it in Kontext.
* Generate convincing ‘rough pencils’ line-art from a Poser render, which can then be combined in Photoshop to firm up outlines on a Kontext watercolour render of a Poser figure. Registration is exact when the layers are blended in Photoshop.
Here the source is a Poser render of Nursoda’s ‘Ronk’ Poser figure and his snail, which I’ve shown here before. The above is the ‘rough pencils’ output at 1024px.
* Earlier in the week I also got a universal Controlnet (mistoLine_rank256.safetensors) working for Kontext.
And I now have saved ComfyUI workflows for the above.
Also, it looks like I moved up to Windows 11 just in time, which over the last few weeks has caused me to go get the best of SDXL, Illustrious and Flux style LoRAs and accessories. CivitAI is to be effectively banned here in the UK, from next week.