H3 - T2VA character inception/sheets+RefMods
retained text
TLDR: You can use RefMods with T2VA. Finally decided to give the RefMod a test run yesterday. I tend to experiment and generate many actors via prose in T2VA and recast them later, see the Tanya bombshell circulating in a few of my videos. Since my workflow mainly revolves around T2VA it make a lot of sense to work with native latent as much as possible. However, instead of providing 10-20 images to make a redmod, I just use the latent from my T2VA character generations. Once you have a latent of anything, you can extend, prepend, or even bridge it as many times as you like, even joining latents of same canvas sizes. The latents can be clipped, cropped, scaled, and upscaled as well. I prefer T2VA due to it's superior image quality compared to R2VA. Anyways, just having fun and wanted to show Vera being recast over various T2VA generations, int8/20 steps, no turbo loras. How do you prefer creating your actors? Krea2? Flux? Qwen image 2.1? MiniMax H3? I am thinking using latents from a MiniMax h3 single-frame generation is going to be pretty good work flow, though I prefer a 2second moving character sheet.
document fetched 2026-10-05 00:29 · canonical https://reddit.com/r/StableDiffusion/comments/1wxlw8w/h3_t2va_character_inceptionsheetsrefmods