39
PART III · HANSEN BY TIMESTAMPS · 12:53
Generation Mode 2 — control the change instead of merely launching a second preset
Hansen returns to the control/config area and Mode 2 generation. The key learning skill here is to separate source-preservation controls from aesthetic controls and understand which parameters truly change the freedom of generation.
CONFIRMEDVideo review around 12:53 shows the control area and Stage 1 preview. The source graph confirms working resolution, generation mode, denoise, ControlNet strengths, prompt selector and shared seed/steps controls.
Separate controls by responsibility
| Family | Examples | Question |
|---|---|---|
| Geometry preservation | Mode, denoise, Depth/Canny strength | How much freedom does AI have? |
| Canvas | Working resolution 702 | At what size does the pipeline operate? |
| Stochastic | Seed 231, steps 771 | How can the sampling state be reproduced? |
| Prompt/style | Prompt selector 453, IPA weight 721 | Which visual intent is applied? |
| Detail conservation | 720 | How much source detail is restored? |
For learning, change one control family at a time
If denoise, seed, prompt and ControlNet strength are changed together, the result cannot be interpreted. A valid A/B test changes one factor while holding the other controls fixed.
Stage 1 preview is a mandatory stop
Mode 2 should first prove that base generation preserved the required architecture. PEOPLE, main FLUX and upscale are not part of that proof.