>>24921
I'm using grok as a chat bot LLM to help generate prompts, i fed it the style guides for krea2 and minimaxh3 to help it format the prompts. I give a generic description of what i want and then tell it to fill in the details.

i feed that into the krea2 workflow i yoinked from the other thread on this board, adjusting LoRAs until the slop it churns out is close to what i want. if it doesn't get close i'll feed the output in as a new latent with different amounts of denoise to try and get it dialed in. once i get an image that's good i feed it into minimax h3.

for minimax h3 i'm using some workflow i found on civitai by DaSiWa. I feed in the image as either a first frame (I2VA) or last frame (L2VA). results are usually better with I2VA. if I was better at photo editing/generation I'd do First and Last frames (FL2VA). sometimes I'll also feed the image into grok to help refine the video prompt. Minimax's prompt coherence is actually pretty outstanding, i typically don't need LoRAs for much.